Systems with Application to Robotics and
Aerospace Vehicles
by
Reza OlfatiSaber
Submitted to the Department of Electrical Engineering and Computer
Science
in partial fulﬁllment of the requirements for the degree of
Doctor of Philosophy in Electrical Engineering and Computer Science
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
February 2001
c Massachusetts Institute of Technology 2001. All rights reserved.
Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Department of Electrical Engineering and Computer Science
January 15, 2000
Certiﬁed by. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Alexandre Megretski, Associate Professor of Electrical Engineering
Thesis Supervisor
Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Arthur C. Smith
Chairman, Department Committee on Graduate Students
Nonlinear Control of Underactuated Mechanical Systems
with Application to Robotics and Aerospace Vehicles
by
Reza OlfatiSaber
Submitted to the Department of Electrical Engineering and Computer Science
on January 15, 2000, in partial fulﬁllment of the
requirements for the degree of
Doctor of Philosophy in Electrical Engineering and Computer Science
Abstract
This thesis is devoted to nonlinear control, reduction, and classiﬁcation of underac
tuated mechanical systems. Underactuated systems are mechanical control systems
with fewer controls than the number of conﬁguration variables. Control of underactu
ated systems is currently an active ﬁeld of research due to their broad applications in
Robotics, Aerospace Vehicles, and Marine Vehicles. The examples of underactuated
systems include ﬂexiblelink robots, mobile robots, walking robots, robots on mo
bile platforms, cars, locomotive systems, snaketype and swimming robots, acrobatic
robots, aircraft, spacecraft, helicopters, satellites, surface vessels, and underwater ve
hicles. Based on recent surveys, control of general underactuated systems is a major
open problem.
Almost all reallife mechanical systems possess kinetic symmetry properties, i.e.
their kinetic energy does not depend on a subset of conﬁguration variables called
external variables. In this work, I exploit such symmetry properties as a means of
reducing the complexity of control design for underactuated systems. As a result,
reduction and nonlinear control of highorder underactuated systems with kinetic
symmetry is the main focus of this thesis. By “reduction”, we mean a procedure
to reduce control design for the original underactuated system to control of a lower
order nonlinear or mechanical system. One way to achieve such a reduction is by
transforming an underactuated system to a cascade nonlinear system with structural
properties. If all underactuated systems in a class can be transformed into a speciﬁc
class of nonlinear systems, we refer to the transformed systems as the “normal form”
of the corresponding class of underactuated systems.
Our main contribution is to ﬁnd explicit change of coordinates and control that
transform several classes of underactuated systems, which appear in robotics and
aerospace applications, into cascade nonlinear systems with structural properties that
are convenient for control design purposes. The obtained cascade normal forms are
three classes of nonlinear systems, namely, systems in strict feedback form, feedfor
ward form, and nontriangular linearquadratic form. The names of these three classes
are due to the particular lowertriangular, uppertriangular, and nontriangular struc
ture in which the state variables appear in the dynamics of the corresponding nonlin
ear systems. The triangular normal forms of underactuated systems can be controlled
using existing backstepping and feedforwarding procedures. However, control of the
2
nontriangular normal forms is a major open problem. We address this problem for
important classes of nontriangular systems of interest by introducing a new stabiliza
tion method based on the solutions of ﬁxedpoint equations as stabilizing nonlinear
state feedback laws. This controller is obtained via a simple recursive method that is
convenient for implementation. For special classes of nontriangular nonlinear systems,
such ﬁxedpoint equations can be solved explicitly.
As a result of the reduction process, one obtains a reduced nonlinear subsystem
in cascade with a linear subsystem. For many classes of underactuated systems, this
reduced nonlinear subsystem is physically meaningful. In fact, the reduced nonlinear
subsystem is itself a Lagrangian system with a welldeﬁned lowerorder conﬁgura
tion vector. In special cases, this allows construction of Hamiltoniantype Lyapunov
functions for the nonlinear subsystem. Such Lyapunov functions can be then used
for robustness analysis of normal forms of underactuated systems with perturbations.
The Lagrangian of the reduced nonlinear subsystem is parameterized by the shape
variables (i.e. the complement set of external variables). It turns out that “a control
law that changes the shape variables” to achieve stabilization of the reduced nonlinear
subsystem, plays the most fundamental role in control of underactuated systems.
The key analytical tools that allow reduction of highorder underactuated systems
using transformations in explicit forms are “normalized generalized momentums and
their integrals” (whenever integrable). Both of them can be obtained from the La
grangian of the system. The diﬃculty is that many reallife and benchmark examples
do not possess integrable normalized momentums. For this reason, we introduce a
new procedure called “momentum decomposition” which uniquely represents a non
integrable momentum as a sum of an integrable momentum term and a nonintegrable
momentumerror term. After this decomposition, the reduction methods for the in
tegrable cases can be applied. The normal forms for underactuated systems with
nonintegrable momentums are perturbed versions of the normal forms for the inte
grable cases. This perturbation is the timederivative of the momentumerror and
only appears in the equation of the momentum of the reduced nonlinear subsystem.
Based on some basic properties of underactuated systems as actuation/passivity
of shape variables, integrability/nonintegrability of appropriate normalized momen
tums, and presence/lack of input coupling; I managed to classify underactuated sys
tems to 8 classes. Examples of these 8 classes cover almost all major applications in
robotics, aerospace systems, and benchmark systems. In all cases, either new con
trol design methods for open problems are invented, or signiﬁcant improvements are
achieved in terms of the performance of control design compared to the available
methods. Some of the applications of our theoretical results are as the following: i)
trajectory tracking for ﬂexiblelink robots, ii) (almost) global exponential tracking
of feasible trajectories for an autonomous helicopter, iii) global attitude and posi
tion stabilization for the VTOL aircraft with strong input coupling, iv) automatic
calculation of diﬀerentially ﬂat outputs for the VTOL aircraft, v) reduction of the
stabilization of a multilink planar robot underactuated by one to the stabilization of
the Acrobot, or the Pendubot, vi) semiglobal stabilization of the Rotating Pendulum,
the BeamandBall system, using ﬁxedpoint state feedback, vii) global stabilization
of the 2D and 3D CartPole systems to an equilibrium point, viii) global asymptotic
3
stabilization of the Acrobot and the InertiaWheel Pendulum.
For underactuated systems with nonholonomic velocity constrains and symmetry,
we obtained normal forms as the cascade of the constraint equation and a reduced
order Lagrangian control system which is underactuated or fullyactuated. This de
pends on whether the sum of the number of constraints and controls is less than,
or equal to the number of conﬁguration variables. This result allows reduction of a
complex locomotive system called the snakeboard. Another result is global exponen
tial stabilization of a twowheeled mobile robot to an equilibrium point which is far
from the origin ( <1), using a smooth dynamic state feedback.
Thesis Supervisor: Alexandre Megretski
Title: Associate Professor of Electrical Engineering
4
Acknowledgments
First and foremost, I would like to thank my advisor Alexandre Megretski. Alex’s
unique view of control theory and his mathematical discipline taught me to see non
linear control from a new perspective that enhanced my understanding of what con
stitutes an acceptable solution for certain fundamental problems in control theory. I
can never thank him enough for giving me the freedom to choose the theoretical ﬁelds
of my interest and helping me to pursue my goals. I have always valued his advice
greatly.
I would like to thank Munther Dahleh for our exciting discussions during the group
seminars and for giving me many useful comments and advice on my work.
I wish to give my very special thanks to both George Verghese and Bernard
Lesieutre who have been a constant source of support and encouragement for me
from the beginning of my studies at MIT.
The past years at MIT were a wonderful opportunity for me to get engaged in
interesting discussions with several MIT professors and scientists. Sanjoy Mitter has
always amazed me with his brilliant view of systems and control theory. I am very
grateful for his invaluable advice on many important occasions. I particularly would
like to thank Eric Feron for his encouragement and for enthusiastically showing me
simulation results and test models related to my work. I wish to thank Dimitri
Bertsekas for being a great teacher and giving me important advice as my academic
advisor during the past years.
Many thanks goes to Nicola Elia and Ulf J¨ onsson for our long hours of discussions
and their valuable technical comments. I would also like to thank Emilio Frazzoli
for many discussions on helicopters and nonlinear control. I would like to take this
opportunity to thank all my friends at LIDS for their encouragement, support, and
the fun times we shared during our group seminars.
I owe my sincere gratitude to Milena Levak, the Dean of the International Students
Oﬃce at MIT, who has greatly helped me during very important situations and times.
I am incredibly grateful to my family. My parents, Hassan and Soroor, and my
sister, Negin, gave me their unconditional love and support during all these years.
While I never had the opportunity to see my family during my ﬁve and a half years of
studies at MIT, my heart has always been with them. They never stopped believing
in me and did everything they could to help me achieve my goals.
At last, my most special thanks goes to Manijeh. Her beautiful smile, emotional
support, and strong encouragement were the driving force of my eﬀorts. With her
creative mind, she gave me many examples of underactuated systems in nature that
I had not thought of myself. Without her help and enthusiasm, this work certainly
would not have been possible.
5
Contents
1 Introduction 14
1.1 Mechanics and Control Theory . . . . . . . . . . . . . . . . . . . . . 14
1.2 Nonlinear Control Systems . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.1 Underactuated Systems . . . . . . . . . . . . . . . . . . . . . . 15
1.2.2 Highly Nonlinear Systems . . . . . . . . . . . . . . . . . . . . 17
1.2.3 Cascade Nonlinear Systems . . . . . . . . . . . . . . . . . . . 19
1.3 Statement of the Contributions . . . . . . . . . . . . . . . . . . . . . 20
1.3.1 Normal Forms for Underactuated Systems . . . . . . . . . . . 22
1.3.2 Reduction and Control of LowOrder Underactuated Systems . 23
1.3.3 Reduction and Control of HighOrder Underactuated Systems 24
1.3.4 Reduction and Control of Underactuated Systems with Non
holonomic Velocity Constraints . . . . . . . . . . . . . . . . . 26
1.3.5 Applications in Robotics and Aerospace Vehicles . . . . . . . . 27
1.3.6 Control of Nontriangular Cascade Nonlinear Systems . . . . . 27
1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2 Mechanical Control Systems 28
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Simple Lagrangian Systems . . . . . . . . . . . . . . . . . . . . . . . 28
2.3 Fullyactuated Mechanical Systems . . . . . . . . . . . . . . . . . . . 30
2.4 Underactuated Mechanical Systems . . . . . . . . . . . . . . . . . . . 31
2.5 Flat Mechanical Systems . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.6 Nonholonomic Mechanical Systems . . . . . . . . . . . . . . . . . . . 32
2.7 Symmetry in Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.8 Normalized Momentums and Integrability . . . . . . . . . . . . . . . 35
3 Normal Forms for Underactuated Systems 37
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Dynamics of Underactuated Systems . . . . . . . . . . . . . . . . . . 39
3.3 Examples of Underactuated Systems . . . . . . . . . . . . . . . . . . 39
3.3.1 The Acrobot and the Pendubot . . . . . . . . . . . . . . . . . 40
3.3.2 The CartPole System and the Rotating Pendulum . . . . . . 41
3.3.3 The BeamandBall System . . . . . . . . . . . . . . . . . . . 42
3.3.4 The TORA System . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3.5 The InertiaWheel Pendulum . . . . . . . . . . . . . . . . . . 43
6
3.3.6 The VTOL Aircraft . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Collocated Partial Feedback Linearization . . . . . . . . . . . . . . . 45
3.5 Noncollocated Partial Feedback Linearization . . . . . . . . . . . . . 46
3.6 Partial Feedback Linearization Under Input Coupling . . . . . . . . . 47
3.7 Normal Forms for Underactuated Systems . . . . . . . . . . . . . . . 49
3.8 Classes of Structured Nonlinear Systems . . . . . . . . . . . . . . . . 51
3.9 Normal Forms for Underactuated Systems with 2 DOF . . . . . . . . 53
4 Reduction and Control of HighOrder Underactuated Systems 58
4.1 Shape Variables and Kinetic Symmetry . . . . . . . . . . . . . . . . . 58
4.2 Underactuated Systems with Noninteracting Inputs and Integrable Mo
mentums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2.1 Underactuated Systems with Actuated Shape Variables . . . . 60
4.2.2 Underactuated Systems with Unactuated Shape Variables . . 63
4.3 Underactuated Systems with Input Coupling . . . . . . . . . . . . . . 74
4.4 Underactuated Systems with Nonintegrable Momentums . . . . . . . 80
4.5 Momentum Decomposition for Underactuated Systems . . . . . . . . 86
4.6 Classiﬁcation of Underactuated Systems . . . . . . . . . . . . . . . . 100
4.7 Nonlinear Control of the Core Reduced System . . . . . . . . . . . . 106
5 Applications to Robotics and Aerospace Vehicles 112
5.1 The Acrobot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.2 The TORA System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.3 The InertiaWheel Pendulum . . . . . . . . . . . . . . . . . . . . . . 127
5.4 The CartPole System . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.4.1 Global Stabilization to an Equilibrium Point . . . . . . . . . . 133
5.4.2 Aggressive Swingup of the Inverted Pendulum . . . . . . . . . 137
5.4.3 Switchingbased Controller . . . . . . . . . . . . . . . . . . . . 139
5.5 The Rotating Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.5.1 Aggressive Swingup of the Rotating Pendulum . . . . . . . . 143
5.6 The Pendubot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.7 The BeamandBall System . . . . . . . . . . . . . . . . . . . . . . . 147
5.8 Control of Multilink Underactuated Planar Robot Arms . . . . . . . 149
5.9 Trajectory Tracking for a Flexible OneLink Robot . . . . . . . . . . 150
5.9.1 Model of a Flexible Link . . . . . . . . . . . . . . . . . . . . . 151
5.9.2 The Noncollocated Output . . . . . . . . . . . . . . . . . . . . 152
5.9.3 Tracking Control . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.10 Conﬁguration Stabilization for the VTOL Aircraft . . . . . . . . . . . 157
5.11 Trajectory Tracking and Stabilization of An Autonomous Helicopter . 161
5.11.1 Dynamic Model of A Helicopter . . . . . . . . . . . . . . . . . 161
5.11.2 Input Moment Decoupling for A Helicopter . . . . . . . . . . . 164
5.11.3 Nontriangular Normal Form of a Helicopter . . . . . . . . . . 170
5.11.4 Feedback Linearization of the Unperturbed Model of the Heli
copter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.11.5 Desired Control Attitude R
d
for Position Tracking . . . . . . . 172
7
5.11.6 Asymptotic Tracking of a Desired Attitude R
d
∈ SO(3) . . . . 173
5.11.7 Thrust for Exponential Attitude Stabilization . . . . . . . . . 181
6 Reduction and Control of Underactuated Nonholonomic Systems 183
6.1 Nonholonomic Systems with Symmetry . . . . . . . . . . . . . . . . . 184
6.2 Applications to Nonholonomic Robots . . . . . . . . . . . . . . . . . 187
6.2.1 A Rolling Disk . . . . . . . . . . . . . . . . . . . . . . . . . . 188
6.2.2 A Nonholonomic Mobile Robot . . . . . . . . . . . . . . . . . 189
6.2.3 A Car . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
6.2.4 The Snakeboard . . . . . . . . . . . . . . . . . . . . . . . . . . 194
6.3 Notions of Stabilization and Tracking . . . . . . . . . . . . . . . . 205
6.4 Global Exponential Stabilization/Tracking for a Mobile Robot . . . 207
6.4.1 Dynamics of a Mobile Robot in SE(2) . . . . . . . . . . . . . . 207
6.4.2 NearIdentity Diﬀeomorphism . . . . . . . . . . . . . . . . . . 208
6.4.3 Control Design for Stabilization/Tracking . . . . . . . . . . 210
7 Control of Nonlinear Systems in Nontriangular Normal Forms 216
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
7.2 Structure of Nontriangular Normal Forms of Underactuated Systems 222
7.3 Stabilization of Nonlinear Systems in Feedback Form . . . . . . . . . 224
7.3.1 Standard Backstepping Procedure . . . . . . . . . . . . . . . . 224
7.3.2 Cascade Backstepping Procedure . . . . . . . . . . . . . . . . 226
7.4 Stabilization of Nonlinear Systems in Nontriangular Forms . . . . . . 231
7.4.1 Nontriangular Nonlinear Systems Aﬃne in (ξ
1
, ξ
2
) . . . . . . . 232
7.4.2 Nontriangular Nonlinear Systems Nonaﬃne in (ξ
1
, ξ
2
) . . . . 242
7.4.3 Global Existence of FixedPoint Control Laws . . . . . . . . . 246
7.4.4 Slightly Nontriangular Nonlinear Systems . . . . . . . . . . . 253
7.4.5 Notion of Partial Semiglobal Asymptotic Stabilization . . . . . 258
7.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
7.5.1 The CartPole System with Small Length/Strong Gravity Eﬀects259
7.5.2 The CartPole System with Large Length/Weak Gravity Eﬀects 262
7.5.3 The Rotating Pendulum . . . . . . . . . . . . . . . . . . . . . 263
7.5.4 The Generalized BeamandBall System . . . . . . . . . . . . 265
8 Conclusions 268
8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
8.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
8.2.1 Hybrid Lagrangian Systems . . . . . . . . . . . . . . . . . . . 270
8.2.2 Uncertainty and Robustness . . . . . . . . . . . . . . . . . . . 272
8.2.3 Output Feedback Stabilization and Tracking . . . . . . . . . . 272
8.2.4 The Eﬀects of Bounded Control Inputs . . . . . . . . . . . . . 272
8.2.5 Small Underactuated Mechanical Systems (SUMS) . . . . . . 272
8.2.6 Underactuated Systems in Strong/Weak Fields . . . . . . . . . 272
8
A Dynamics of Mechanical Systems 274
A.1 Underactuated Mechanical Systems with Two DOF and Kinetic Sym
metry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
A.2 A Threelink Planar Robot . . . . . . . . . . . . . . . . . . . . . . . . 275
A.3 Twolink Planar Robots: the Acrobot and the Pendubot . . . . . . . 279
A.4 The BeamandBall System . . . . . . . . . . . . . . . . . . . . . . . 280
A.5 The CartPole System . . . . . . . . . . . . . . . . . . . . . . . . . . 282
A.6 The Rotating Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . 283
A.7 The 3D CartPole System with Unactuated Roll and Pitch Angles . . 284
A.8 An Underactuated Surface Vessel . . . . . . . . . . . . . . . . . . . . 286
B Symbolic Calculation of the Flat Outputs of the Helicopter 289
C Proofs of Theorems in Chapter 7 290
C.1 Proof of Theorem 7.4.3: . . . . . . . . . . . . . . . . . . . . . . . . . 290
C.2 Proof of the Stability of µSubsystem in Theorem 7.4.4 . . . . . . . . 291
C.3 Proof of Lemma 7.4.4: . . . . . . . . . . . . . . . . . . . . . . . . . . 291
C.4 Proof of Theorem 7.4.4: . . . . . . . . . . . . . . . . . . . . . . . . . 292
C.5 Proof of Theorem 7.4.5: . . . . . . . . . . . . . . . . . . . . . . . . . 295
9
List of Figures
31 (a) The Acrobot, (b) The Pendubot. . . . . . . . . . . . . . . . . . . 40
32 (a) The CartPole system, (b) The Rotating Pendulum. . . . . . . . . 41
33 The BeamandBall System. . . . . . . . . . . . . . . . . . . . . . . . 42
34 The TORA system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
35 The Inertia Wheel Pendulum . . . . . . . . . . . . . . . . . . . . . . 44
36 The VTOL aircraft. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
41 A ﬂexible onelink robot arm. . . . . . . . . . . . . . . . . . . . . . . 73
42 A ﬂying bird. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
43 A triplelink robot arm with two actuators. . . . . . . . . . . . . . . . 81
44 An inverted pendulum on a moving platform with 4 DOF and 2 controls. 84
45 A triplelink robot arm with actuated shape variables. . . . . . . . . . 90
46 A planar threelink robot arm. . . . . . . . . . . . . . . . . . . . . . . 94
47 Three possible actuation conﬁguration for a planar threelink robot arm. 96
51 The Acrobot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
52 The function γ(q
2
) for a = 20/3, b = 4, c = 4/3. . . . . . . . . . . . . . 115
53 The function α(q
r
) for m
1
= m
2
= 1 and l
1
= l
2
= 1. . . . . . . . . . . 119
54 Trajectory of (q
1
, q
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π/3, 0, 0) and control input p
2
. . . . . . . . . . . . . . . . . . . . . . 120
55 Trajectory of (p
1
, p
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π/3, 0, 0) and control input p
2
. . . . . . . . . . . . . . . . . . . . . . 120
56 Trajectory of (q
1
, q
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π, 0, 0) and control input p
2
. . . . . . . . . . . . . . . . . . . . . . . 121
57 Trajectory of (p
1
, p
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π, 0, 0) and control input p
2
. . . . . . . . . . . . . . . . . . . . . . . 121
58 Trajectory of (q
1
, q
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(0, π, 0) and control input p
2
. . . . . . . . . . . . . . . . . . . . . . . 122
59 Trajectory of (p
1
, p
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(0, π, 0) and control input p
2
. . . . . . . . . . . . . . . . . . . . . . . 122
510 The TORA system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
511 State trajectories and control input of the TORA system for the ini
tial condition (q
1
, q
2
, p
1
, p
2
) = (1, 0, 0, 0): (a) (q
1
, p
1
), (b) (q
2
, p
2
), (c)
(q
r
, p
r
), and (d) u. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
512 The InertiaWheel Pendulum . . . . . . . . . . . . . . . . . . . . . . 128
10
513 For the initial condition (π, 0, 0), (a) shows the state trajectories for
(q
1
, ˙ q
1
), (b) shows the state trajectory in (q
1
, ˙ q
1
) plane, (c) shows ˙ q
2
,
and (d) shows the control input τ. . . . . . . . . . . . . . . . . . . . . 131
514 For the initial condition (π/2, 0, 0), (a) shows the state trajectories for
(q
1
, ˙ q
1
), (b) shows the state trajectory in (q
1
, ˙ q
1
) plane, (c) shows ˙ q
2
,
and (d) shows the control input τ. . . . . . . . . . . . . . . . . . . . . 132
515 The CartPole system. . . . . . . . . . . . . . . . . . . . . . . . . . . 133
516 (a) State trajectory of the cartpole system for the initial condition
(5, 0, π/3, 0) and σ
i
() = sat
1
(), (b) The force F and control input u. 136
517 (a) State trajectory of the cartpole system for the initial condition
(5, 0, π/3, 0) and σ
i
() = sat
0.1
(), (b) The force F and control input u. 136
518 The deadzone nonlinearity. . . . . . . . . . . . . . . . . . . . . . . . 138
519 (a) Aggressive swingup of an inverted pendulum on a cart from the
downward position (q
2
, p
2
) = (π, 0), (b) The path of the pendulum in
(q
2
, p
2
)plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
520 (a) State trajectory of the cartpole system for the initial condition
(0, 0, π, 0), (b) The input force F and the trajectory in (q
2
, p
2
)plane. 141
521 The Rotating Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . 141
522 (a) Aggressive swingup of an inverted pendulum on a rotating arm
from the downward position (q
2
, p
2
) = (π, 0), (b) Trajectory of the
pendulum in (q
2
, p
2
)plane. . . . . . . . . . . . . . . . . . . . . . . . . 145
523 The Pendubot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
524 The BeamandBall System. . . . . . . . . . . . . . . . . . . . . . . . 147
525 A ﬂexible onelink robot arm. . . . . . . . . . . . . . . . . . . . . . . 151
526 The VTOL aircraft. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
527 (a) The state trajectories of the VTOL aircraft, and (b) The path of
the center of mass of the aircraft in (x
1
, y
1
)plane. . . . . . . . . . . . 160
528 A helicopter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
61 A vertical rolling disk . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
62 A mobile robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
63 The model of a car. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
64 The Snakeboard. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
65 The geometry of the paths of the rear and front wheels of the car with
a radius of rotation ρ. . . . . . . . . . . . . . . . . . . . . . . . . . . 202
66 The geometry of the paths of the rear and front wheels and the center
of mass of the snakeboard with a radius of rotation ρ. . . . . . . . . . 204
67 A near identity transformation with a parameter [λ[ <1. . . . . . . . 209
68 Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane for ini
tial position x = (2, 3)
T
(v = 0) and orientation angles θ = 0, π/4, π/2, 3π/4.211
69 Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane for ini
tial position x = (2, 3)
T
(v = 0) and orientation angles θ = π, 5π/4, 3π/2, 7π/4.212
610 Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane starting
at x = (2, 3)
T
(v = 0) with angle θ = π/4 . . . . . . . . . . . . . . . . 213
11
611 Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane for
initial position x = (0, 2)
T
(v = 0) and initial orientation angles
θ = 0, π/4, π/2, 3π/4. . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
612 Evolution of position and controls for the nonholonomic mobile robot
in (x
1
, x
2
)plane starting at x = (0, 2)
T
(v = 0) with angle θ = 0 . . . 214
613 Trace trajectories of a twowheel nonholonomic robot with initial po
sition (5, 4) and heading angles (a) θ = 0, (b) θ = π/2, (c) θ = π, and
(d) θ = 3π/2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
614 Position trajectory tracking for a mobile robot from the initial point
(4, 4, π/2). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
71 A C
2
smooth sigmoidal function with ﬂat tails with its derivatives. . 248
72 Sigmoidal ribbon o and compact set /. . . . . . . . . . . . . . . . . . 250
73 The CartPole System . . . . . . . . . . . . . . . . . . . . . . . . . . 260
74 The state trajectories of the CartPole system. . . . . . . . . . . . . . 261
75 The Rotating Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . 263
76 The state trajectories for the Rotating Pendulum. . . . . . . . . . . . 265
77 The state trajectories of the generalized beamandball system with
initial condition (1, 0, −π/6, 0) for g = −9.8 and ν = 1, 2, 3. . . . . . . 267
A1 A planar threelink robot arm. . . . . . . . . . . . . . . . . . . . . . . 275
A2 The BeamandBall System. . . . . . . . . . . . . . . . . . . . . . . . 281
A3 The CartPole system. . . . . . . . . . . . . . . . . . . . . . . . . . . 282
A4 The Rotating Pendulum. . . . . . . . . . . . . . . . . . . . . . . . . . 283
A5 The 3D CartPole system with unactuated roll and pitch angles. . . . 285
A6 An underactuated surface vessel. . . . . . . . . . . . . . . . . . . . . 286
12
List of Tables
4.1 Classiﬁcation of Underactuated Systems . . . . . . . . . . . . . . . . 101
5.1 The Values of F
∗
(, n
0
). . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.2 The Values of τ
∗
(, n
0
, 1). . . . . . . . . . . . . . . . . . . . . . . . . . 144
13
Chapter 1
Introduction
1.1 Mechanics and Control Theory
Control of mechanical systems is currently among one of the most active ﬁelds of
research due to the diverse applications of mechanical systems in reallife. Though,
the study of mechanical systems goes back to Euler and Lagrange in the 1700’s, it was
not until 1850’s that mechanical control systems came to the picture in regulation of
steam engines. During the past century, a series of scientiﬁc, industrial, and military
applications motivated rigorous analysis and control design for mechanical systems.
On the other hand, theoretically challenging nature of analysis of the behavior of non
linear dynamical systems attracted many mathematicians to study control systems.
As a result, the eﬀorts of engineers and scientists together led to creation of Linear
Control, Optimal Control, Adaptive Control, and Nonlinear Control theories. More
recently, Robust Control theory was born and added to the above picture because of
an inevitable need to deal with the presence of uncertainties in reallife control sys
tems. Before the creation of most of these theories, humans managed to fulﬁll their
longtime dream to travel in the space and land on the moon. This is one of the great
est accomplishments of control scientists and engineers of the ﬁrst half of the past
century. Aerospace Vehicles and Robotics Systems with ability to explore the surface
of other planets are still among the most complex machines built by humans. For
the past 50 years, Aerospace and Robotics applications remained as some of the most
inﬂuential sources of motivation for rigorous analysis and control of both mechanical
systems and nonlinear systems. All along the way, advances achieved by researchers
in Mechanics and Nonlinear Control Theory mutually aﬀected and enhanced each
other. Control of Multibody Systems [53] in mechanics and rigorous analysis (e.g.
controllability and observability) and control design (e.g. stabilization and output
tracking) for nonlinear systems aﬃne in control [36, 65], which shaped the ﬁeld of
nonlinear control theory, were just two theoretical consequences of this eﬀort.
14
1.2 Nonlinear Control Systems
In this section, we provide an overview of the past research on classes of nonlinear
control systems and mechanical systems that are relevant to this work. In this thesis,
our main focus in is on control design for nonlinear systems that arise from control
of an important and broad class of mechanical systems known as underactuated sys
tems. Underactuated Systems are mechanical control systems with fewer actuators
(i.e. controls) than conﬁguration variables. One of the main contributions of this
thesis is explicit transformation of highorder underactuated systems into cascade
nonlinear systems with both triangular and nontriangular structural properties. This
transformation is performed using a global/semiglobal change of coordinates obtained
from the Lagrangian of the system in closedform. After applying this transforma
tion, control of the original (possibly) highorder underactuated system, reduces to
control of lowerorder nonlinear systems (possibly) nonaﬃne in control. This moti
vated us to develop new control design methods for both cascade nonlinear systems
with nontriangular structures and classes of nonlinear systems nonaﬃne in control,
i.e. highly nonlinear systems. As a result, the main body of this thesis involves the
following topics:
i) Dynamics, Reduction, and Control of Underactuated Systems
ii) Control of Highly Nonlinear Systems
iii) Nonlinear Control of Cascade Systems with Nontriangular Structures
In the following, we present the stateoftheart of research in each of the above topics.
1.2.1 Underactuated Systems
Underactuated mechanical systems are systems that have fewer control inputs than
conﬁguration variables. Underactuated systems appear in a broad range of appli
cations including Robotics, Aerospace Systems, Marine Systems, Flexible Systems,
Mobile Systems, and Locomotive Systems. The “underactuation” property of under
actuated systems is due to the following four reasons: i) dynamics of the system
(e.g. aircraft, spacecraft, helicopters, underwater vehicles, locomotive systems with
out wheels), ii) by design for reduction of the cost or some practical purposes (e.g.
satellites with two thrusters and ﬂexiblelink robots), iii) actuator failure (e.g. in a
surface vessel or aircraft), iv) imposed artiﬁcially to create complex loworder nonlin
ear systems for the purpose of gaining insight in control of highorder underactuated
systems (e.g. the Acrobot, the Pendubot, the BeamandBall system, the CartPole
system, the Rotating Pendulum, the TORA system (all are described in chapter 3)).
The main control methods applied to examples of invertedpendulum type under
actuated systems is based on swingup of the pendulum from its downward position
and then switching to a balancing controller that is designed using a linearization
technique or gain scheduling to balance the pendulum [94]. This includes swing up
control using energybased methods [90] for the Acrobot (i.e. a twolink planar robot
15
with an actuator at the elbow) [86], the Pendubot (i.e. a twolink planar robot with
an actuator at the shoulder) [92], the cartpole system [21] , a triplelink inverted pen
dulum [90], and the rotating pendulum [6]. The balancing controller for the Acrobot
using spline functions can be found in [13, 14].
Due to its complexity, the beamandball has been the focus of study among re
searchers with diverse interests including approximate feedback linearization methods
by Hauser et al. [34], using small nested saturations for stabilization of feedforward
cascade nonlinear systems due to Teel [102] and stabilization by output feedback [103].
Moreover, stabilization of the beamandball by construction of Lyapunov functions
is addressed in [80]. This is done based on the original work due to Mazenc and Praly
in [56]. Recently, global stabilization of the beamandball system is achieved in [79].
Passivitybased method is mostly used for swingup control design of underactu
ated systems with invertedpendulums [28, 90]. Moreover, a passivitybased approach
was employed for a special example of a pendulum on a cart (i.e. the TORA sys
tem) that was transformed to a cascade form [37]. The main drawback of these
passivitybased methods is their narrow range of applications. In fact, to the best of
my knowledge, so far no applications of energybased methods in control of reallife
underactuated systems in Robotics, Aerospace, and Marine applications is known.
The VTOL (vertical take oﬀ and landing) aircraft is another example of an under
actuated system that has been extensively used as a testbed for diﬀerent methods of
trajectory tracking and conﬁguration stabilization. This includes tracking for slightly
nonminimum phase systems [35] and hybrid/switching based control methods [58].
Exponential stabilization of examples of underwater vehicles and surface vessels
that are underactuated was achieved in [27] and [75] using appropriate coordinate
transformations and analysis of a timevaring linear system. A similar type result for
attitude control of an underactuated spacecraft is given in [62].
The role of secondorder nonholonomic constrains in the necessity of the use of
discontinuous stabilizing feedback laws for stabilization of underactuated systems is
discussed in [111]. This is mainly based on the famous condition of Brockett on
stabilizability of nonlinear systems using timeinvariant continuously diﬀerentiable
state feedbacks [15]. In addition to this issue, accessibility of classes of underactuated
mechanical systems has been recently addressed in a recent important work by Rey
hanoglu et al. [78]. This is based on a framework applied to analysis of controllability
of nonholonomic systems [11] and a famous controllability theorem due to Sussmann
[97]. An example of a discontinuous stabilizing feedback for a system with an internal
unactuated degree of freedom is given in [77].
Adaptive control [33] and sliding mode control techniques [96] have been also
applied to underactuated mechanical systems for rather limited applications.
Flexiblelink robots are an important class of underactuated systems that are
appropriate for space applications due to their lightweight and fast execution of com
mands. The EulerBernoulli model for a ﬂexible arm is an inﬁnite dimensional system
[25]. A truncated modal analysis can be used to obtain a ﬁnite dimensional statespace
model for ﬂexible robots [25, 107]. Trajectory tracking for ﬂexible robots is rather
complicated and common measurements like the angle of rotation or the position of
the tip, respectively, lead to a poor performance and nonminimum phase zero dy
16
namics (see [20, 24] and references therein). In [20], a noncollocated minimumphase
output is proposed based on an analysis of the initial inﬁnite dimensional model and
then a ﬁniteorder compensator is deigned for trajectory tracking. Here, we obtain
a nonlinear noncollocated minimumphase output for a ﬂexible onelink robot arm
from a ﬁniteorder EulerLagrangian equations of the system.
The method of controlled Lagrangians (i.e. applying a control input that preserves
the Lagrangian structure of a mechanical system) has been recently applied to local
stabilization of the cartpole system and the rotating pendulum to an equilibrium
manifold [10]. However, so far this method has been unable to stabilize the Rotating
Pendulum or more general underactuated systems to an equilibrium point.
In addition to more traditional methods, recently, hybrid and switchingbased
control methods are ﬁnding their way in control of underactuated mechanical systems
[7, 41] and bipedal locomotion of walking robots [88, 89, 91].
In conclusion, apart from linearizationbased techniques, control of underactuated
mechanical systems has been mainly along the line of stabilization of special exam
ples of cascade nonlinear systems and using energybased methods combined with a
supervisorybased switching control. The stateoftheart of research in control of un
deractuated systems is currently very far from our goal to ﬁnd control design methods
for broad classes of highorder underactuated systems that are eﬀective for Robotics
and Aerospace applications.
1.2.2 Highly Nonlinear Systems
To describe what we mean by “highly nonlinear systems”, ﬁrst we present an evolution
of control systems from linear systems to nonlinear systems. We start with linear
timeinvariant control systems in the form
˙ x = Ax + Bu , y = Cx (1.1)
where x ∈ R
n
, y ∈ R
m
, u ∈ R
p
. Questions regarding controllability, observability,
stabilization, and tracking for this system using state or output feedback has been
quite wellunderstood for a long time. However, adding constraints or further speciﬁ
cations to the description of the system might make the control design for the system
rather complicated. For example, if the system is controllable with bounded control
inputs, is there any static or dynamic state feedback that asymptotically stabilizes
the system? or what if A, B, C are known up to an uncertainty (i.e. A = A
0
+ ∆
where A
0
is known and ∆ is unknown but normbounded). None of these problems
can be addressed as simple as the control problems for the original linear system (1.1).
One can observe that minor deviations from the standard problem of stabilization of
an LTI control system and additional constraints make the system rather complex.
As a further step, consider the following linear system in feedback connection with a
memoryless nonlinearity φ
˙ x = Ax +Bu , y = φ(Cx) (1.2)
17
where φ(0) = 0 and ∃c
2
> c
1
≥ 0 : c
1
z
2
≤ z φ(z) ≤ c
2
z
2
, ∀z ∈ R. We call these
systems slightly nonlinear systems. Control design and analysis for systems in the form
(1.2) with diﬀerent types of nonlinearities led to Absolute Stability theory in 60’s. An
inputoutput stability approach or a frequency domain analysis are the dominant tools
in dealing with slightly nonlinear systems. In mideighties, this notion of feedback
interconnection of a linear system and a nonlinearity was generalized to feedback
of an LTI system and an uncertainty (or operator) that has bounded gain. This
led to Robust Stability theory [23] and more recently Integral Quadratic Constraints
(IQC’s) [61] due to Megretski and Rantzer. Though, these methods are successful in
dealing with linear systems, uncertain linear systems, and slightly nonlinear systems,
they are not applicable to truly nonlinear systems (i.e. systems with nonlinear time
evolution of the state) that do not include any basic linear parts. To be more precise,
a modiﬁcation of (1.2) as
˙ x = σ(Ax + Bu) , y = Cx (1.3)
called a recurrent neural network with a saturationtype (i.e. sigmoidal) nonlinearity
σ, has no fundamental similarities in terms of controllability and observability to an
LTI control system or a slightly nonlinear system. In addition, the linearity of the
output does not simplify analysis of the system due to the fact that timeevolution of
the system follows a nonlinear law. Dynamic neural networks are examples of highly
nonlinear systems and only recently around midnineties conditions for controllability
(only the discretetime case) and observability of these systems were introduced by
Albertini and Sontag [2, 3]. The analysis method employed in the preceding work was
a rather involved timedomain analysis. Obviously, due to the fact that a frequency
domain analysis can only deal with systems that have linear timeevolution of the
state.
A rather standard form for nonlinear systems aﬃne in control in analogy to (1.1)
is
˙ x = f(x) + g(x)u , y = h(x) (1.4)
where f, g, h are nonlinear smooth functions. By a highly nonlinear system, we mean
a system in the following form
˙ x = f(x, u) , y = h(x) (1.5)
where f is a nonlinear function of (x, u) (regardless of linearity or nonlinearity of h(x))
such that there exist no diﬀeomorphism (z, v) = (T
1
(x), T
2
(x, u)) and matrices A, B
satisfying ˙ z = Az + Bv. A comprehensive local theory regarding controllability, ob
servability, stabilization, tracking, and disturbance decoupling for nonlinear systems
in (1.4) can be found in [36] (see references therein for further details). The main
tools to address these control problems were diﬀerentialgeometry and Lie theory that
became very common in the literature around late 70’s and early 80’s. Though these
methods were rather successful in local analysis of nonlinear systems aﬃne in control
they usually fail to work for a global analysis and nonlinear systems that are non
aﬃne in control. Moreover, lie algebraic conditions are not robust to uncertainties in
18
f, g, h.
InputtoState Stability (ISS) theory, essentially developed and introduced by Son
tag [82, 84], combines both Absolute Stability and Robust Stability theories in one
for highly nonlinear systems in the general form (1.5). The main tools in this the
ory for robustness analysis to disturbances are control Lyapunov functions (CLF’s).
The problem is that in general it is not easy to construct CLF’s for highly nonlinear
systems.
In many control applications, a global/semiglobal control design and analysis is
required. In addition, after applying certain nonlinear coordinate transformation to
the dynamics of nonlinear systems aﬃne in control, the transformed system or its
subsystems could be nonlinear systems that are nonaﬃne in control. This motivated
us to consider global/semiglobal stabilization and analysis of highly nonlinear systems
that arise from the study of underactuated mechanical systems and nonholonomic
systems. This topic is covered in chapters 4, 5, 6, and 7.
1.2.3 Cascade Nonlinear Systems
Cascade nonlinear systems arise in many control applications either naturally after
some change of coordinates, or due to design of an output feedback or a dynamic
state feedback. In general, they are in the following form
˙ z = f(z, ξ)
˙
ξ = g(ξ, u)
(1.6)
The most wellknown results for cascade systems are related to nonlinear systems in
triangular forms. Namely, nonlinear systems in strict feedback forms and feedforward
forms. The backstepping procedure has proven to be successful for systems in strict
feedback form [57, 36, 40, 80]. Control of feedforward cascade nonlinear systems is
also wellstudied. The simplest example of a feedforward nonlinear system is a per
turbed chain of integrators that can be stabilized using small nested saturations [102].
Then, more general classes of feedforward systems were either controlled using a non
linear smallgain theorem [101] or construction of Lyapunov functions [56]. However,
control of cascade nonlinear systems with nontriangular structures has proven to be
rather problematic. This is due to a counterexample by Sussmann [98] and existence
of the peaking phenomenon [100]. It turns out that if some coupling terms in the
dynamics of cascade nonlinear systems satisfy appropriate growth conditions then it
is possible to stabilize them using lowgain or highgain feedback laws [80], [79]. In
chapter 5, we show that the Rotating Pendulum and the Pendubot are examples of
underactuated systems that have nontriangular structures and they do not satisfy any
of the aforementioned growth conditions. This suggests developing new stabilization
methods for cascade nonlinear systems with nontriangular structures. This stabiliza
tion method is one of the main contributions of this work which will be presented in
chapter 7.
19
1.3 Statement of the Contributions
In this section, we brieﬂy present the main contributions of this thesis. More detailed
statement of the contributions are given in the subsequent sections.
The main focus in this thesis is on nonlinear control of underactuated mechani
cal systems. This is motivated by broad applications of underactuated systems and
theoretically challenging problems that they have to oﬀer. Though, control of me
chanical systems with nonholonomic ﬁrstorder (or velocity) constraints is very well
studied, there is very little known about control of underactuated systems with non
holonomic secondorder (or acceleration) constraints. In fact, based on two recent
surveys [78, 87], control of general underactuated mechanical systems is currently
considered a major open problem.
Our main contribution in this thesis is to provide analytical tools that allow us
to translate and solve control problems for broad classes of highorder underactuated
systems to equivalent control problems for lowerorder nonlinear systems. We call this
systematic translation procedure reduction of underactuated systems. Such a reduc
tion process requires an explicit transformation, i.e. a global change of coordinates (or
diﬀeomorphism) in closedform. We introduce these explicit change of coordinates for
broad classes of underactuated systems including both reallife examples in robotics
and aerospace applications and benchmark nonlinear systems. Under such a diﬀeo
morphism, the original underactuated system transforms into a nonlinear system that
is referred to as a normal form. It turns out that underactuated systems with symme
try properties can be reduced to normal forms that are structured cascade nonlinear
systems. These cascade normal forms are nonlinear systems in strict feedback form
[57], feedforward form [102], and nontriangular linearquadratic form [69] (see section
3.8 for deﬁnitions). This in turn implies that these normal forms can be controlled
using backstepping procedure [57], feedforwarding [102, 101, 56, 80], and ﬁxedpoint
state feedback laws [67] (presented in chapter 7), respectively. Therefore, this thesis
includes a complete package for control design and analysis of both loworder and
highorder underactuated systems with symmetry by providing
i) Classiﬁcation of underactuated systems,
ii) Reduction of each class to its normal form using an explicit transformation,
iii) Control design for each reduced normal form (or class).
All three types of aforementioned cascade normal forms are special classes of
ByrnesIsidori normal form with a doubleintegrator linear part [36]. However, the
special structure of each type of normal form allows eﬀective use of the existing and
recently developed nonlinear control design methods. This is very important, since
even stabilization of the general ByrnesIsidori normal form with a doubleintegrator
linear part is currently an open problem in nonlinear control theory.
Nearly all current applications of underactuated systems possess certain kinetic
symmetry properties. By kinetic symmetry, we mean invariance of the kinetic energy
under the action of a cyclic group. Of course, the reader should notice that this
20
is a diﬀerent notion than the famous classical symmetry in mechanics [1, 54] which
in a simpler form was familiar to Lagrange. The classical symmetry is deﬁned as
the invariance of the Lagrangian of a system under a symmetry group action. In
simple words, mechanical systems that their inertia matrix is independent of a subset
of conﬁguration variables possess kinetic symmetry property w.r.t. this subset of
variables. We call this subset of variables external variables. The compliment of
the set of external variables are called shape variables. Thus, shape variables are
conﬁguration variables that appear in the inertia matrix of a (simple) mechanical
system. The key tools in reduction of highorder underactuated systems are normalized
generalized momentums conjugate to external or shape variables together with their
integrals (see section 2.8 for the deﬁnitions). It turns out that actuation or lack of
actuation of the shape variables of an underactuated system and also integrability
of the normalized momentums play the fundamental role in the structure of the
corresponding normal form of the system. These two properties together with another
two basic properties of underactuated systems led us to classiﬁcation of underactuated
mechanical systems with kinetic symmetry to 8 classes. For each class, we provide
its associated normal form, control design method, and examples from reallife and
benchmark nonlinear systems. A surprising result is that the reducedorder subsystem
of the normal form, for several classes of underactuated systems is a Lagrangian
system satisfying forced (or unforced) EulerLagrange equations. In other words, I
obtain ByrnesIsidori normal forms for underactuated systems with zerodynamics
that are welldeﬁned lowerorder Lagrangian systems. In certain cases, this allows us
to use the Hamiltonian/Lagrangian of the reduced system as a Lyapunov function.
Several examples of reallife and benchmarktype highorder underactuated sys
tems possess nonintegrable normalized momentums (e.g. a ﬂexiblelink robot and
a 3D CartPole system). We introduce a method called Momentum Decomposition
which uniquely decomposes a nonintegrable normalized momentum as the summa
tion of an integrable momentum and a nonintegrable momentum. This class of
underactuated systems with nonintegrable momentums can be then reduced to cas
cade normal forms plus structured perturbation terms. A possible way to deal with
this problem is to design a nonlinear controller for the nominal unperturbed system
that renders the desired equilibrium point of the underactuated system asymptoti
cally stable and makes the perturbation term exponentially vanishing (the details are
provided in chapter 4). Then, after performing a Lyapunovbased Robustness Analy
sis, the stability of the perturbed system in closedloop with the nominal nonlinear
controller can be veriﬁed. A Lyapunov function candidate for this analysis can be
calculated via the Lyapunov function of the core reduced system of the unperturbed
system.
In chapter 5, we apply our theoretical results in reduction and control of high
order underactuated systems to a number of complex underactuated systems. Here
is a list of contributions on these applications:
– Global asymptotic stabilization of the Acrobot to any arbitrary equilibrium
point of the system including the upright equilibrium using a nonlinear state
feedback.
21
– Global asymptotic stabilization of the TORA system using a nonlinear state
feedback.
– Global asymptotic stabilization of the InertiaWheel Pendulum using a nonlin
ear state feedback.
– Almost global exponential stabilization of the CartPole system, and the Ro
tating Pendulum to their upright relative equilibrium point.
– Global asymptotic and local exponential stabilization of the 2D and 3D Cart
Pole systems to their upright equilibrium point.
– Semiglobal asymptotic stabilization of the CartPole system using a ﬁxedpoint
state feedback law.
– Partial Semiglobal asymptotic stabilization of the Rotating Pendulum and the
BeamandBall system using a ﬁxedpoint state feedback law.
– Global exponential stabilization of a nonholonomic twowheeled mobile robot
to a point <1 far from the origin using a smooth dynamic state feedback.
– Trajectory tracking for a ﬂexible onelink robot arm using a nonlinear noncol
located output.
– Reduction of stabilization of a planar nlink robot underactuated by one to
stabilization of the Acrobot or the Pendubot.
– Global position and attitude stabilization for the VTOL aircraft with strong
input coupling.
– Almost global exponential tracking of the feasible trajectories of an autonomous
helicopter.
– Automatic calculation of diﬀerentially ﬂat outputs for the VTOL aircraft as a
byproduct of the decoupling of the eﬀects of the body torque in the translational
dynamics of the VTOL.
– Automatic calculation of weakly diﬀerentially ﬂat outputs for a relatively ac
curate model of an autonomous helicopter as a byproduct of the approximate
decoupling of the eﬀects of the body torque in the translational dynamics of the
helicopter.
1.3.1 Normal Forms for Underactuated Systems
Underactuated systems as mechanical systems with fewer inputs than conﬁguration
variables with a nonholonomic secondorder (or acceleration) constraint are not fully
(i.e. exact) feedback linearizable. In [90], Spong showed that actuated conﬁguration
variables of an underactuated system with noninteracting inputs (deﬁned in section
3.4) can be globally linearized using an invertible change of control. This is called
22
collocated partial feedback linearization. This is a important initial step in some of
our results and proves that a broad class of underactuated systems can be partially
linearized. However, the main diﬃculty arises after applying this change of control.
Since the new control appears both in the actuated and unactuated subsystems of the
original underactuated system. In chapter 3, we propose an appropriate structure for
a global change of coordinates that decouples these two subsystems w.r.t. the new
control. Suﬃcient conditions are provided for the existence of this diﬀeomorphism.
The condition is a simple algebraic test on the inertia matrix of the system. After ap
plying this transformation, one obtains a cascade normal form which is a special class
of ByrnesIsidori normal form with a doubleintegrator linear part [36]. The weakness
of this result is that to obtain the change of coordinates, based on Frobenius’s theo
rem, it is required to solve a ﬁrstorder partial diﬀerential equation (PDE). Solving
this PDE, in turn, reduces to solving a ﬁnite number of ODE’s which constitute the
characterization equations of the PDE. For loworder underactuated systems with 2
DOF, this PDE can be explicitly solved. Solving this PDE explicitly for highorder
underactuated is not a feasible approach. Instead, in chapter 4, we present a new
method to resolve this issue. The method mainly relays on the use of normalized
generalized momentums and their integrals (wherever possible).
Partial feedback linearization is an initial simplifying step for reduction and con
trol of underactuated, regardless of the method used for decoupling of the actuated
and unactuated subsystems. For this purpose, we extended the partial feedback lin
earization result in [90] to two other cases: i) linearization of the dynamics of the
unactuated conﬁguration variables which is called noncollocated partial feedback lin
earization, and ii) linearization of actuated conﬁguration variables for underactuated
systems with input coupling. The former one is used for partial feedback lineariza
tion of underactuated systems with unactuated shape variables (e.g. a ﬂexiblelink
robot or the Rotating Pendulum). The latter one is very useful in aerospace applica
tions, particularly, for decoupling of the eﬀects of the body torque in the translational
dynamics of a VTOL aircraft or a helicopter.
1.3.2 Reduction and Control of LowOrder Underactuated
Systems
In chapter 3, combination of a partial feedback linearization step and an explicit de
coupling change of coordinates is used for reduction of underactuated systems with
two degrees of freedom. Derivation of explicit normal forms for loworder underac
tuated systems is important, because we later show that control of special classes of
highorder underactuated systems reduces to control of a number of loworder (e.g. 2
DOF) systems. As an example, in chapter 4, we prove that stabilization of an nlink
(n > 2) planar robot with (n −1) controls reduces to stabilization of the Acrobot or
the Pendubot regardless of n.
There is a simple way to classify underactuated systems with 2 DOF and a single
shape variable. Either their shape variable is actuated, or it is unactuated. We call
these two classes of underactuated systems ClassI and ClassII systems, respectively.
23
Examples of ClassI systems are the Acrobot, the InertiaWheel Pendulum, and the
TORA system. We proved that all ClassI underactuated systems can be trans
formed into a cascade nonlinear system in strict feedback form and (possibly) globally
asymptotically stabilized using backstepping procedure. In contrast, all ClassII un
deractuated systems can be globally/semiglobally transformed into cascade systems
in nontriangular quadratic normal form (deﬁnition 3.8.5). Examples of ClassII sys
tems include ﬂexiblelink robots, the CartPole system, the Rotating Pendulum, the
Pendubot, and the BeamandBall system. Stabilization of general partiallylinear
nontriangular cascade nonlinear systems is currently an open problem [36]. We
address this problem for important classes of nontriangular linearquadratic normal
forms and more general nontriangular forms in chapter 7.
We provide suﬃcient conditions so that under a second change of coordinates,
ClassII systems can be reduced to cascade nonlinear systems in feedforward forms.
Among all the stated examples, the CartPole system is the only one that falls within
this class and can be globally asymptotically stabilized using a state feedback in the
form of nested saturations due to Teel [102]. In chapter 5, we present detailed control
design for all the aforementioned examples of underactuated systems with 2 DOF.
1.3.3 Reduction and Control of HighOrder Underactuated
Systems
Most of reallife underactuated systems are highorder systems. An aircraft or a heli
copter has 6 DOF and 4 controls. A ﬂexible nlink robot with m deformation modes at
each link has n(m+1) DOF, n controls, and nm unactuated degrees of freedom (e.g.
for n = 2, m = 3 the system has 8 DOF and 2 controls). This motivated us to focus on
reduction of highorder underactuated systems (in chapter 4). The kinetic symmetry
property of underactuated systems and the degree in which the shape variables are
actuated are two main factors in reduction of highorder underactuated systems. Our
key analytical tools here are the normalized generalized momentums conjugate to ac
tuated/unactuated conﬁguration variables and their integrals (wherever possible). It
turns out that the integrability of theses normalized momentums plays a fundamental
role in the structure of the normal forms for highorder underactuated systems.
An important property of normal forms for highorder underactuated systems is
that they are physically meaningful. To be more precise, consider a class of un
deractuated systems with n degrees of freedom, m actuated shape variables with
noninteracting inputs. Assume the normalized momentums conjugate to the (n−m)
external variables of this system are integrable. We call this class of highorder under
actuated systems ClassI systems. We prove that there exists a global diﬀeomorphism
obtained from the Lagrangian of the system that transforms the dynamics of a ClassI
underactuated system into a reducedorder system with a new conﬁguration vector of
dimension (n−m) and a welldeﬁned Lagrangian function parameterized by the shape
variables which satisﬁes the unforced EulerLagrange equations. This normal form
is in fact a cascade nonlinear system in strict feedback form with a zerodynamics
(associated with the shape variables as the output) which is a Lagrangian function
24
itself. We refer to this Lagrangian zerodynamics system as the core reduced system.
In chapter 4, we provide a method for stabilization of this core reduced system using a
nonlinear static state feedback. As a byproduct, we obtain an energytype Lyapunov
function for the closedloop core reduced system. In certain cases, the physical prop
erty of the zerodynamics subsystem allows using Hamiltonian/Lagrangian of the core
reduced system as a valid Lypunov function. These Lyapunov functions can be later
used for robustness analysis of the perturbed normal forms of ClassI underactuated
systems.
Now, consider the class of underactuated systems with integrable normalized mo
mentums conjugate to shape variables and noninteracting inputs. We call them
ClassII underactuated systems. After the reduction process, ClassII underactuated
systems can be explicitly transformed into nontriangular quadratic normal forms. In
this case, still the reducedorder subsystem (i.e. the zerodynamics) is a welldeﬁned
Lagrangian system (under minor conditions). But it satisﬁes a forced EulerLagrange
equation with a rather complex input force with a nonlinear dependence on both the
conﬁguration and momentum of the reduced Lagrangian system. The complexity of
the structure of this input force is one of the main sources of extreme diﬃculties in
control design for ClassII underactuated systems.
We provide simple algebraic suﬃcient conditions on the inertia matrix and po
tential energy of the system such that the nontriangular normal form of a ClassII
underactuated system can be transformed into a feedforward system using a second
explicit change of coordinates. The subclass of ClassII systems that satisﬁes these
suﬃcient conditions are called ClassIII systems.
The second important factor in reduction of underactuated systems, particularly
in aerospace applications and locomotive systems, is the presence of input coupling
between actuated and unactuated subsystems. For ﬂat underactuated systems in
cluding aircraft and helicopters, the reduction is rather straightforward. In chapter
4, we address reduction of nonﬂat underactuated systems with input coupling. This
class of reallife systems can potentially have very high degrees of freedom. An exam
ple is a ﬂying bird as a threebody system with 14 DOF and 8 controls. In general,
many locomotive systems in the nature fall within this class. Other examples include
a swimming ﬁsh and a walking human being. All of these three examples have ac
tuated shape variables and they locomote in their environment using the coupling
between their translational dynamics and shape variables. Therefore, the presence of
input coupling is very crucial for the purpose of locomotion of these underactuated
systems. We call these new classes of nonﬂat and underactuated systems with input
coupling ClassIV and ClassV systems, respectively.
For the reduction of ClassI and ClassII underactuated systems, we assumed the
normalized momentums are integrable. A natural question to ask is whether it is
possible to reduce these systems if their normalized momentums are nonintegrable.
The answer to this question is positive. The integrability property of the normal
ized momentums can be considered as the third most important factor in reduction
of highorder underactuated systems. To address reduction underactuated systems
with nonintegrable momentums, we introduce a new procedure called Momentum
Decomposition. This method uniquely decomposes a nonintegrable momentum as
25
the summation of an integrable momentum and a nonintegrable momentum. Let
us call the underactuated systems with nonintegrable normalized momentums corre
sponding to ClassI,II,II, respectively, ClassVI, ClassVII, and ClassVIII systems.
We prove that normal forms of ClassVI,VII,VIII underactuated systems with non
integrable momentums are perturbed versions of the normal forms of ClassI,II,III
systems with a structured perturbation that only appears in the dynamics of the
momentum of the reduced subsystem.
In conclusion, we obtain 8 classes of underactuated systems. Namely, ClassI
through ClassVIII systems which are precisely deﬁned with their subclasses in section
4.6. The result of classiﬁcation of underactuated systems is summarized in Table 4.1.
For each class, related examples and an appropriate control design method are given.
Our main contribution is that the methods for reduction and control of highorder
systems presented in chapter 4, applies to the majority of important examples of
robots, aerospace systems, and benchmark nonlinear systems.
1.3.4 Reduction and Control of Underactuated Systems with
Nonholonomic Velocity Constraints
In chapter 6, we focus on reduction of underactuated systems with kinetic symmetry
and nonholonomic velocity constraints. Our main result on reduction of nonholonomic
underactuated systems with symmetry is that under certain assumptions, the reduced
system can be represented as the cascade of the constraint equation and a lowerorder
underactuated (or fullyactuated) Lagrangian system. If the sum of the number of
controls and constraint equations is less than the dimension of the conﬁguration
vector, the reduced system is underactuated. Otherwise, it is fullyactuated. The
snakeboard example (Fig. 64) is a mobile robot with 6 degrees of freedom, 3 controls,
and 2 nonholonomic contraints [50]. Our reduction method, provides a systematic
way for calculation of the reduced system of the snakeboard. Furthermore, we prove
that the kinematic model of a snakeboard and a car are diﬀeomorphic. This implies
that the snakeboard can be transformed into a ﬁrstorder chainedtype nonholonomic
system (equation 2.8) using an explicit change of coordinates.
Based on Brockett’s result [15], the class of mechanical systems with ﬁrstorder
nonholonomic constraints can not be stabilized to the origin using a smooth static
state feedback. An example of this class is a simple twowheeled mobile robot. We in
troduce a new class of parameterized diﬀeomorphisms that are called nearidentity dif
feomorphisms. An example of a nearidentity diﬀeomorphism is ψ(x; θ, ) = x+v(θ)
where v(θ) is a unit vector and for a ﬁx θ and , ψ is clearly a diﬀeomorphism which
is equal to id(x) = x for = 0. The point x
= x + v(θ) is called an nearby
point of x. We use a nearidentity change of coordinates for global exponential
stabilization/tracking of a nonholonomic twowheeled mobile robot to an nearby
point/trajectory of a desired position/trajectory using a smooth dynamic state feed
back ( <1). This motivated us to formally deﬁne the notions of stabilization and
tracking for nonlinear systems.
26
1.3.5 Applications in Robotics and Aerospace Vehicles
In chapter 5, we provide detailed control design for several examples of loworder and
highorder underactuated systems in robotics and aerospace applications. A list of
some of the examples and their associated control task is stated in the last part of
the introduction of section 1.3.
1.3.6 Control of Nontriangular Cascade Nonlinear Systems
Stabilization of nontriangular cascade nonlinear systems is currently considered a
major open problem [36]. In chapter 7, we address this stabilization problem for
important classes of nontriangular cascade nonlinear systems, particularly, systems
in nontriangular linearquadratic normal form. This includes normal forms for Class
II, ClassIV, and ClassVII underactuated systems.
1.4 Outline
The outline of this thesis is as follows. Chapter 1, provides an introduction and
statement of the contributions. In chapter 2, we give some background on classes of
mechanical control systems and deﬁne some basic notions in mechanics. Chapter 3 is
devoted to the dynamics of underactuated system, several examples of underactuated
systems, partial feedback linearization techniques, normal forms for underactuated
systems, and classiﬁcation of loworder underactuated systems. Our main contribu
tions are presented in chapter 4. This includes reduction, classiﬁcation, and control
of highorder underactuated systems. In chapter 5, we present detailed control design
for several robotics and aerospace applications. In chapter 6, we address reduction
and control of underactuated systems with nonholonomic velocity constraints and
symmetry properties. In chapter 7, we introduce ﬁxedpoint state feedback laws
for stabilization of nontriangular cascade nonlinear systems. In chapter 8, we make
concluding remarks and state possible directions of future research.
27
Chapter 2
Mechanical Control Systems
2.1 Introduction
This chapter provides some background on mechanical control systems. We start
by deﬁning simple mechanical control systems and introduce Legendre normal form
of a simple mechanical system. Then, we discuss fullyactuated mechanical systems
and their trasformation into a set of doubleintegrators using a change of control.
We deﬁne underactuated mechanical systems in contrast to fullyactuated mechani
cal systems and demonstrate why control of underactuated systems is a challenging
open problem. A special class of mechanical systems called ﬂat mechanical systems
is described next. We present a result on controllability limitations of underactuated
ﬂat mechnaical systems in lack of gravity terms based on the Legendre normal form
of a ﬂat underactuated system. We brieﬂy review stateoftheart of research in mod
eling and control of mechnaical systems with nonholonomic velocity (i.e. ﬁrstorder)
constranits and clarify their close relations to underactuated mechanical systems that
contain nonholonomic acceleration constraints. Symmetry in mechanics plays a cru
cial role in some of the main results of this thesis regarding reduction of underactuated
mechanical systems. Therefore, the role of symmetry and conservation laws in me
chanics and their relation through Noether’s theorem is explained in the ﬁnal section.
2.2 Simple Lagrangian Systems
In this section, we state forced EulerLagrange equations of motion as the model of
control mechanical systems and introduce Legendre normal form for simple mechan
ical systems.
Simple mechanical systems are systems that their Lagrangian is in the form of the
diﬀerence between a (positive semideﬁnite) kinetic energy and a potential energy
L(q, ˙ q) = K −V =
1
2
˙ q
T
M(q) ˙ q −V (q)
where q ∈ Q denotes the conﬁguration vector that belongs to an ndimensional conﬁg
uration manifold Q, M(q) is the inertia matrix which is a positive deﬁnite symmetric
28
matrix. K is the kinetic energy and V (q) is the potential energy of the system. Let
f
i
(q) : Q → R
n
, i = ¦1, . . . , m¦ be m linearly independent external forces applied to
the system. The EulerLagrange equation for this mechanical control system is as the
following
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= F(q)u (2.1)
where u ∈ R
m
and F(q) = (f
1
(q), . . . , f
m
(q)) denotes the matrix of external forces.
The equations of motion for this mechanical system can be derived as the following
Σ
j
m
kj
(q)¨ q
j
+ Σ
i,j
Γ
k
ij
(q) ˙ q
i
˙ q
j
+g
k
(q) = e
T
k
F(q)u, k = 1, . . . , n
where e
k
is the kth standard basis in R
n
, g
k
(q) = ∂
q
k
V (q), and Γ
k
ij
(q) are called
Christoﬀel symbols and can be deﬁned as
Γ
k
ij
(q) =
1
2
(
∂m
kj
(q)
∂q
i
+
∂m
ki
(q)
∂q
j
−
∂m
ij
(q)
∂q
k
).
in the vector form, we have
M(q)¨ q + C(q, ˙ q) ˙ q + G(q) = F(q)u (2.2)
c
ij
= Σ
n
k=1
Γ
i
kj
(q) ˙ q
k
is an element of C(q, ˙ q). The term C(q, ˙ q) ˙ q contains two types of terms involving
˙ q
i
˙ q
j
that are called Centrifugal terms (i = j) and Coriolis terms (i = j). Also,
G(q) contains the gravity terms (see [95] for more details and examples). The in
teresting relationship between the matrices M and C is that S
0
=
˙
M(q) − 2C(q, ˙ q)
is a skew symmetric matrix. In other words, because M is symmetric, one gets
˙
M(q) = C(q, ˙ q) + C
T
(q, ˙ q). Taking this property into account and noting that M(q)
is a positive deﬁnite matrix, the Legendre transform with respect to ˙ q
p =
∂L
∂ ˙ q
= M(q) ˙ q
is invertible and the dynamics of the mechanical system in (2.2) can be rewritten in
the canonical form
˙ q = M
−1
(q)p
˙ p = −G(q) +
˜
C
T
(q, p)M
−1
(q)p +F(q)u
(2.3)
where
˜
C(q, p) = C(q, M
−1
p). We call (2.3) Legendre Normal Form of a simple me
chanical system. Deﬁning the variables x
1
= q, x
2
= p, (2.3) can be rewritten as
˙ x
1
= M
−1
(x
1
)x
2
˙ x
2
= −G(x
1
) + x
T
2
Q(x
1
)x
2
+ F(x
1
)u
(2.4)
or
˙ x = f(x) +g(x)u (2.5)
29
which is a nonlinear system aﬃne in control with state x = col(x
1
, x
2
) and vector
ﬁelds
f(x) = col(M
−1
(x
1
)x
2
, −G(x
1
) + x
T
2
σ(x
1
)x
2
)
g(x) = col(0, F(x
1
))
with the property that f(x) has a mixed linearquadratic structure w.r.t. x
2
and
g(x) is independent of x
2
. Note that σ is a cubic matrix such that x
T
2
Q(x
1
)x
2
∈ R
n
.
The main advantage of the form (2.5) is that many analytical tools are available
for controllability and observability analysis and control design of nonlinear systems
aﬃne in control [36].
Remark 2.2.1. For mechanical systems, equations (2.2) and (2.3) are two equivalent
forms. However, The Legendre normal form is a ﬁrstorder ODE and (2.2) is a second
order ODE. Later, we will see that from control point of view (2.3) is more appropriate
and can be directly used for controllability analysis of a mechanical system.
Remark 2.2.2. Deﬁning the Hamiltonian of a simple Lagrangian system as
H(q, p) =
1
2
p
T
M
−1
(q)p + V (q)
the Legendre normal form in (2.3) is equivqlent to the following forced Hamiltonian
equations
˙ q =
∂H(q, p)
∂p
˙ p = −
∂H(q, p)
∂q
+F(q)u
However, the Hamiltonian form is not appropriate for direct controllability analysis
and control design compared to the Legendre normal form.
In the following sections, we describe several important classes of mechanical con
trol systems.
2.3 Fullyactuated Mechanical Systems
Consider the control mechanical system in (2.1). We call (2.1) a fullyactuated me
chanical system if m = rank F(q) = n, i.e. F(q) is an invertible matrix. For fully
actuated systems the number of control inputs is equal to the dimension of their con
ﬁguration manifold. Therefore, fullyactuated mechanical systems are exact feedback
linearizable (i.e. they have no zerodynamics [36]). This can be proved by applying
the following change of control
u = F(q)
−1
(M(q)v + C(q, ˙ q) ˙ q + G(q))
and redeﬁning the variables as x
1
= q, x
2
= ˙ q, to obtain
˙ x
1
= x
2
˙ x
2
= v
30
which is clearly a (vector) double integrator. This is why most of problems for fully
actuated mechanical systems can be reduced to equivalent problems for linear systems.
This fact suggests that control of mechanical systems is challenging either in the
presence of uncertainties (e.g. parametric uncertainties in M, C, G), or when the
number of control inputs is less than n which in both cases exact feedback linearization
is not possible. Moreover, under any holonomic or nonholomic constraint (both to
be deﬁned), control and analysis of mechanical systems could potentially be rather
complicated. The same is true if the control input of a mechanical system is required
to be bounded.
2.4 Underactuated Mechanical Systems
A control mechanical system with conﬁguration vector q ∈ Q and Lagrangian L(q, ˙ q)
satisfying the EulerLagrange equation
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= F(q)u (2.6)
is called an Underactuated Mechanical System (UMS) if m = rankF(q) < n = dim(Q)
(see the deﬁnitions for equation (2.1)). In other words, underactuted systems are
mechanical systems that have fewer actuators than conﬁguration variables. This
restriction of the control authority does not allow exact feedback linearization of
underactuated systems. As a special case, assume F(q) = (0, I
m
)
T
. Then, the ﬁrst
(m−n) equations in (2.6) can be expressed as a secondorder dynamic equation
ϕ(q, ˙ q, ¨ q) = 0 (2.7)
which contains Coriolis, centerifugal, and gravity terms that can be highly nonlinear.
If there exists no function h such that
¨
h = ϕ(q, ˙ q, ¨ q), equation (2.7) is called a second
order nonholonomic constraint, or a nonholonomic acceleration constraint (e.g. see
section A.8). We assume all underactuated systems throughout this work have second
order nonholonomic constraints unless otherwise is stated.
For the special case of F(q) = (0, I
m
)
T
, we refer to the ﬁrst (n −m) equations of
(2.6) as the unactuated subsystem and to the last m equations of (2.6) as the actuated
subsystem. In [90], Spong has shown that the actuated subsystem of a general un
deractuated system with F(q) = (0, I
m
)
T
can be linearized using an invertible change
of control. This procedure is called partial feedback linearization. However, after
partial linearization, the unactuated subsystem of (2.6) still remains as a nonlinear
system that is coupled with the linearized actuated subsystem through both the new
control input and other nonlinear terms. This highly complicates control design for
underactuated systems. One of the main contributions of this thesis is to decouple the
actuated and unactuated subsystem of the underactuated system in (2.6) with respect
to the new control (see section 3.7). The main focus of this thesis is control design
and analysis of underactuated mechanical systems. A comprehensive treatment of
control of underactuated system will be presented in the subsequent chapters.
31
2.5 Flat Mechanical Systems
Mechanical systems that their inertia matrix is constant are called Flat Mechanical
Systems. One of the characters of a ﬂat mechanical system is that the Christoﬀel
symbols associated with their inertia matrix vanishes identically and therefore the
matrix C(q, ˙ q) vanishes as well. Now, assume that for a mechanical system with inertia
matrix M(q) there exists a diﬀeomorphism z = ψ(q) such that dψ(q) = M
1/2
(q)dq,
then we have
˙ z = M
1/2
˙ q
and the Lagrangian of the system can be written as
L(z, ˙ z) =
1
2
˙ z
T
I ˙ z −V (ψ
−1
(z))
Thus, M(z) = I is constant and in new coordinates the mechanical system is ﬂat.
The Legendre normal form for a ﬂat mechanical system with B(q) = I is as the
following
˙ q = M
−1
p
˙ p = τ −G(q)
This is due to the fact that
˜
C(q, p) = C(q, M
−1
(q)p) = 0. The following proposition
reveals a major controllability limitation of ﬂat underactuated mechanical systems.
Proposition 2.5.1. Consider an underactuated ﬂat mechanical system with the grav
ity term G(q) = (g
1
(q), . . . , g
n
(q))
T
. Suppose for some 1 ≤ i ≤ n, q
i
is unactuated
and g
i
(q) = 0, then the system is not controllable/stabilizable.
Proof. Based on the assumptions, ˙ p
i
= 0 and p
i
is a conserved quantity, i.e. p
i
(t) =
p
i
(0), ∀t ≥ 0. If p
i
(0) = 0 then the system cannot be stabilized to any equilibrium
point/manifold with p = 0.
2.6 Nonholonomic Mechanical Systems
A Lagrangian mechanical system with m < n velocity constraints
W
T
(q) ˙ q = 0
(W is an m n matrix) that are nonintegrable (i.e. h(t) :
˙
h = W
T
(q) ˙ q) is called
a Mechanical System with Firstorder Nonholonomic Constraints. Control of non
holonomic mechanical systems with velocity constraints has been extensively studied
in the recent years by numerous researchers. The history of research in this ﬁeld
goes back to early 1900’s. Formulation of the dynamics of nonholonomic systems
can be found in [64]. Controllability and stabilization of nonholonomic systems is
considered in [11]. In an important work, Bloch and Crouch addressed modeling of
nonholonomic systems on Riemmanian manifolds [8]. More recently, many advances
have been made on controllability of nonholonomic systems by Lewis and Murray
32
[49], locomotion and control design for nonholonomic systems by Ostrowski in [74],
and reduction of nonholonomic systems with symmetry due to Bloch et al. [9].
Following the line of formulation in [11] and based on [64], the dynamics of non
holonomic systems with velocity constraints can be described as the following:
M(q)¨ q + C(q, ˙ q) + G(q) = W(q)λ + B(q)u
W
T
(q) ˙ q = 0
where λ ∈ R
m
is the vector of Lagrange multipliers.
Remark 2.6.1. In the literature [111], [78], sometimes in contrast to nonholonomic
systems with ﬁrstorder (or velocity) constraints, underactuated mechanical systems
are called systems with secondorder (or acceleration) nonholonomic constraints. This
is due to the fact that if B(q) = (0, I)
T
is partitioned according to q = (q
1
, q
2
)
T
, then
regardless of the control input we have
d
dt
(
∂L
∂ ˙ q
1
) −
∂L
∂q
1
= 0
or
m
11
(q)¨ q
1
+ m
21
(q)¨ q
2
+ h
1
(q, ˙ q) = 0
where m
ij
’s are block matrices of the inertia matrix associated with q = (q
1
, q
2
)
T
.
However, this terminology is somewhat misleading since the Lagrangian of an under
actuated mechanical system satisﬁes EulerLagrange equations of motions without
any external diﬀerentialalgebraic constraints that requires use of Lagrangian mul
tipliers in a variational setting as in [8] for nonholonomic systems with ﬁrstorder
constraints.
Control design for broad classes of nonholonomic mechanical systems can be re
duced to control of chainedtype systems as the following
x
(k
1
)
1
= u
1
x
(k
2
)
2
= u
2
x
(k
3
)
3
= x
2
u
1
. . .
x
(kn)
n
= x
n−1
u
1
(2.8)
where k
i
≥ 1, i = 1, . . . , n are integers. Mobile robots, carlike vehicles, and multi
trailer systems [104] are examples of chainedform systems in (2.8). For k
i
= 1, system
(2.8) is nonholonomic with ﬁrstorder constraints. Recently, in a an important work
[47], (quasi) exponential stabilization of higherorder chainedtype systems was ad
dressed by Laiou and Astolﬁ based on the original work of Astolﬁ [4]. A discontinuous
dynamicstate feedback was introduced for stabilization. Traditionally, stabilization
of chainedform systems with at most ﬁrstorder nonholonomic constraints has been
done using sinusoids [63] or homogeneous timevarying state feedback [60], [59] due to
M’Closkey and Murray. Later, the theoretical results in [60] were applied to local ex
ponential stabilization of the attitude of a spacecraft with nonholonomic constraints
33
[62] and conﬁguration stabilization of a nonholonomic underwater vehicle [27].
2.7 Symmetry in Mechanics
In this section, we deﬁne symmetry in mechanics, the connection between symmetry
and conservation laws (i.e. Noether’s theorem), and the role of symmetry in reduction
of the dynamics of a mechanical system. Formally, the symmetry in mechanics is
deﬁned as the invariance of the Lagrangian of a system under the action of a left (or
right) invariant Lie group. For a complete treatment of symmetry see [54] (also [1]),
for deﬁnitions of Lie groups, left invariant Lie groups, and the action of Lie groups
on manifolds see Warner and Boothby.
Almost all reallife mechanical systems possess certain symmetry properties. For
example, the Lagrangian of a helicopter, a car, or a Satellite is independent of their
position. This gives rise to symmetries (i.e. invariance of the Lagrangian) w.r.t.
translation.
Here, we only consider symmetry of Lagrangian systems in a Euclidean space
Q = R
n
. We say the Lagrangian L(q, ˙ q) is symmetric w.r.t. the conﬁguration variable
q
i
iﬀ
∂L
∂q
i
= 0, i ∈ ¦1, . . . , n¦
Denote the ith generalized momentum by p
i
as
p
i
=
∂L
∂ ˙ q
i
and consider the unforced EulerLagrange equations of motion
d
dt
∂L
∂ ˙ q
i
−
∂L
∂q
i
= 0
It immediately follows that the symmetry of L w.r.t. q
i
imples that the ith momentum
p
i
is conserved (i.e. ˙ p
i
= 0) and vice versa. The last theorem in its most general form
is called Noether’s theorem [1] which establishes a onetoone equivalence between
the existence of symmetries and conservation laws. Lagrange himself was aware of
this fact and he called q
i
an ignorable coordinate. This is due to the fact that the
degree of the secondorder equation of motion (i.e. Hamiltonian form) for the me
chanical system reduces by 2. Therefore, the existence of symmetry properties leads
to reduction of the mechanical systems.
For the special case of a simple mechanical systems that its Lagrangian is sym
metric w.r.t. q
i
, the condition ˙ p
i
= 0 is equivalent to a ﬁrstorder constraint
W
T
(q) ˙ q = p
i
(0)
where
W(q) = (m
i1
(q), . . . , m
in
(q))
T
34
is the ith row of the inertia matrix M(q). If this constraint is nonintegrable, the
analysis of the system reduces to the analysis of a mechanical system with a ﬁrst
order nonholonomic constraint. Therefore, in addition to the conservation laws, the
existence of symmetries gives rise to holonomic/nonholonomic velocity constraints for
mechanical systems.
In this thesis, we use a diﬀerent notion of symmetry called Kinetic Symmetry that
is rather similar to classical symmetry. By Kinetic Symmetry, we mean the kinetic
energy of the system is invariant w.r.t. q
i
, i.e.
∂K
∂q
i
= 0
The kinetic symmetry is equivqlent to classical symmetry in lack of potential en
ergy. However, the majority of systems that we consider here have nonzero (i.e.
nonconstant) potential energies and the notions of kinetic symmetry and classical
symmetry do not coincide. It is important to notice that the existence of kinetic sym
metries (in presence of a potential ﬁeld) does not lead to the existence of conserved
quantities. later, we will see that the fact that the generalized momentums are not
conserved has a crucial role in controllability of large classes of underactuated me
chanical systems. In addition, we will show that the existence of kinetic symmetry
properties leads to reduction of underactuated Lagrangian systems regardless of the
lack of conserved quantities.
2.8 Normalized Momentums and Integrability
In this work, normalized momentums are the most important analytical tools in reduc
tion of underactuated systems. Consider an underactuated system with conﬁguration
vector q = (q
1
, q
2
) and Lagrangian
L(q, ˙ q) =
1
2
¸
˙ q
1
˙ q
2
T
¸
m
11
(q) m
12
(q)
m
21
(q) m
22
(q)
¸
˙ q
1
˙ q
2
−V (q)
The generalized momentum conjugate to q
i
can obtained via partial Lengendre trans
form
p
i
:=
∂L
∂q
i
= m
i1
(q) ˙ q
1
+ m
i2
(q) ˙ q
2
, i = 1, 2
The normalized momentums conjugate to q
1
and q
2
w.r.t. q
1
are deﬁned as the
following
π
1
:= m
−1
11
(q)p
1
= ˙ q
1
+ m
−1
11
(q)m
12
(q) ˙ q
2
, π
2
:= m
−1
21
(q)p
2
= ˙ q
1
+ m
−1
21
(q)m
22
(q) ˙ q
2
where π
1
is always welldeﬁned and π
2
is deﬁned wherever m
21
(q) is an invertible
square matrix. We say π = π(q, ˙ q) is an integrable normalized momentum, if there
exists a generalized conﬁguration function h = h(q) such that
˙
h = π where the
35
function
˙
h is deﬁned as
˙
h(q, ˙ q) = ∇h(q) ˙ q
We call h(q) the integral of π. Whenever h :
˙
h = π, we say π is nonintegrable.
36
Chapter 3
Normal Forms for Underactuated
Systems
3.1 Introduction
Underactuated mechanical systems appear in a variety of reallife control applications
including Robotics (e.g. lightweight ﬂexiblelink robots, nonholonomic mobile robots,
and walking robots), Aerospace Vehicles (e.g. helicopters, aircraft, spacecraft, and
satellites), Underwater Vehicles, and Surface Vessels. Due to their broad range of
applications, control of underactuated systems is extremely important. The restric
tion of the control authority for underactuated systems causes major diﬃculties in
control design for these systems. During the past decade, the challenging nature of
analysis and control of underactuated systems has attracted numerous researchers
with interests in nonlinear control theory, robotics and automation, control of auton
umous vehicles, and control of ﬂexible structures. Control of general underactuated
mechanical systems is currently considered a major open problem based on recent
surveys [78, 87].
The reasons behind the complexity of control design for underactuated systems can
be summarized as follows. Underactuated systems are not fully feedback linearizable.
Moreover, many recent and traditional methods of nonlinear control design including
backstepping [36, 57], forwarding [56, 80, 101, 102], highgain/lowgain designs [80,
79], and sliding mode control [40] are not directly applicable to underactuated systems
with the exception of a few special cases (e.g. the beamandball system and the cart
pole system). This is due to the fact that a method for transforming underactuated
systems into cascade nonlinear systems with upper/lower triangular or nontriangular
structural properties has not yet been discovered. The classes of nonlinear systems
of interest include ByrnesIsidori normal forms (equation (3.14)) [36], strict feedback
forms (deﬁnition 3.8.2) [36, 57], and feedforward forms (deﬁnition 3.8.3) [102, 101, 56].
In this chapter, our main contribution is to obtain the aforementioned cascade nor
mal forms for underactuated systems by applying global/semiglobal nonlinear change
of coordinates. The structural properties of the new normal forms allow eﬀective
use of many existing control design methods. In addition, a new class of underac
37
tuated systems is found that leads to a cascade normal form with a nontriangular
quadratic structure (see deﬁnition 3.8.5). Due to the nontriangular nature of this
normal form, backstepping/forwarding procedures [80] are not applicable to it. Sta
bilization of this class of nontriangular nonlinear systems is equivalent to control of
a special ByrnesIsidori normal form with a doubleintegrator linear part that is a
major open problem in nonlinear control theory [36, 98, 99]. In chapter 7, we address
important special cases of this stabilization problem including the cases that appear
in control of underactuated mechanical systems.
To attain insight in dealing with control design for higherorder underactuated
systems (e.g. a helicopter with 6 DOF and 4 controls), ﬁrst it is very useful to
consider calculation of cascade normal forms for lowerorder underactuated systems.
For doing so, we focus on the class of underactuated systems with two degrees of
freedom and kinetic symmetry (deﬁned in section 2.7). We prove that for this special
class of underactuated systems both the decoupling change of coordinates and the
corresponding cascade normal form can be obtained in explicit forms. The structure
of the obtained normal forms for underactuated systems with two degrees of freedom
naturally leads to their classiﬁcation in two classes. The ﬁrst class includes systems
like the Acrobot, the Inertia Wheel Pendulum, and the TORA system (see section 3.3
for the deﬁnitions) which can be transformed into cascade nonlinear systems in strict
feedback form (deﬁnition 3.8.2). The second class includes the Pendubot, the Cart
Pole system, the Rotating Pendulum, and the BeamandBall system (see section 3.3
for the deﬁnitions) which can be transformed into a nontriangular quadratic normal
form (deﬁnition 3.8.5). In addition, the cartpole system can be transformed into a
feedforward form after another transformation (deﬁnition 3.8.3).
As an initial step, our main result on normal forms for underactuated systems
partly relays on a method called collocated partial feedback linearization due to Spong
[90]. Spong showed that the dynamics of the actuated conﬁguration variables of any
underactuated mechanical system can be globally linearized by applying a change of
control. This allowed application of passivitybased/energybased methods to solving
swingup problem for special examples of underactuated systems that involve inverted
pendulums. However, after this change of control, the new control appears both in
the actuated subsystem and the unactuated subsystem of the original system. This
coupling w.r.t. the input highly complicates control design for underactuated systems.
To resolve this issue, we introduce an appropriate structure for a global change of
coordinates that decouples the linear actuated and nonlinear unactuated subsystems
w.r.t. the new control. This global change of coordinates transforms the dynamics
of the system into a cascade nonlinear system in ByrnesIsidori normal form with a
doubleintegrator linear part [36]. In addition, characterization of the cases where
this normal form reduces to (strict) feedback/feedforward forms is provided.
We introduce a procedure to linearize the dynamics of the unactuated conﬁgura
tion variables using a change of control. This is applicable under the condition that
the number of unactuated variables is less than or equal to the number of actuated
variables. We call this procedure noncollocated partial feedback linearization. This
new partial linearization method is used in obtaining the nontriangular quadratic
normal form for a class of underactuated systems with kinetic symmetry.
38
The outline of this chapter is as follows. First, we present standard models for
underactuated systems in section 3.2. Then, we provide several examples of underac
tuated systems including both benchmark systems and reallife applications in section
3.3. In sections 3.4 and 3.5, we explain collocated and noncollocated partial feedback
linearization methods. Our main results on normal forms for underactuated systems
are presented in section 3.7. In section 3.9, we focus on normal forms and classiﬁcation
of underactuated systems with two degrees of freedom and kinetic symmetry.
3.2 Dynamics of Underactuated Systems
The EulerLagrange equations of motion for an underactuated system are as the
following
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= F(q)τ
where τ ∈ R
m
is the control and F(q) ∈ R
n×m
is a nonsquare matrix of external
forces with m < n and full column rank. Here, m denotes the number of control
inputs that is less than the number of conﬁguration variables n. For the case of
simple Lagrangian systems, the equations of motion for an underactuated mechanical
system can be expressed as the following
M(q)¨ q +C(q, ˙ q) ˙ q + G(q) = F(q)τ (3.1)
Assuming F(q) = [0, I
m
]
T
, the conﬁguration vector can be partitioned as q = (q
1
, q
2
) ∈
R
(n−m)
R
m
according to F(q) where q
1
and q
2
denote the actuated and unactuated
conﬁguration vectors, respectively [90]. After partitioning the inertia matrix M(q)
accordingly, the dynamics of an underactuated system takes the form
¸
m
11
(q) m
12
(q)
m
21
(q) m
22
(q)
¸
¨ q
1
¨ q
2
+
¸
h
1
(q, ˙ q)
h
2
(q, ˙ q)
=
¸
0
τ
(3.2)
with τ ∈ R
m
. Due to the lack of control in the ﬁrst equation of (3.2), it is not possible
to fully linearize this underactuated system using a change of control. However, it is
still possible to partially linearize the system such that the dynamics of q
2
transforms
into a double integrator [90]. We demonstrate this procedure in sections 3.4 and 3.5.
3.3 Examples of Underactuated Systems
In this section, we present several examples of underactuated systems. These exam
ples include the Acrobot, the Pendubot, the CartPole system, the Rotating Pen
dulum, the BeamandBall system, the TORA (translational rotational actuator)
system, the InertiaWheel Pendulum, and the VTOL aircraft (vertical takeoﬀ and
landing). All the examples are chosen due to the complexity of their control design
and the fact that for analysis and control purposes they are of high interest in the
literature. We brieﬂy introduce each example with its related control design task and
39
later use them as applications of our theoretical results by providing control design
and analysis for each case in chapter 5. The stateoftheart of research in control
design for each example is provided as well. The inertia matrix for each system is
given for the future use. The examples of this section only include underactuated
systems with two or three degrees of freedom. Later in chapter 5, we give exam
ples of higherorder underactuated systems that appear in Robotics and Aerospace
applications.
3.3.1 The Acrobot and the Pendubot
The Acrobot [13, 86] is a twolink planar robot with revolute joints and one actuator
at the elbow as shown in Figure 3.3.1 (a). The Pendubot shown in Figure 3.3.1 (b) is
also a twolink planar robot with revolute joints and an actuator at the shoulder. The
q
q
l
1
L
1
m I
m I
2 2
1 1
1
2
x
y
q
q
l
1
L
1
m I
m I
2 2
1 1
1
2
x
y
(a) (b)
Figure 31: (a) The Acrobot, (b) The Pendubot.
Acrobot and the Pendubot graphically seem to be very similar (i.e. the share the exact
same inertia matrix). Though, later we will see that the diﬀerence in the location of
their single actuator causes a major diﬀerence in their standard representation (i.e.
normal form) and control design. The inertia matrix for both the Acrobot and the
Pendubot is the same and has the following form
M(q
2
) =
¸
a
1
+ a
2
cos(q
2
) a
3
+
1
2
a
2
cos(q
2
)
a
3
+
1
2
a
2
cos(q
2
) a
4
where the a
i
’s are positive constants. A possible control task is to stabilize the
upright equilibrium point for the Acrobot or the Pendubot from any initial condition.
A more complicated task is to stabilize the Acrobot or the Pendubot to any of their
arbitrary inﬁnite equilibrium points. In the past, the former task has been done using
swinging up these double inverted pendulums from their downward initial positions
and bringing them close to their upright position, then switching to a linear controller
around the upright equilibrium point [13, 86, 90, 94, 28]. The latter task has never
been done before to the best of our knowledge. In this work, we perform both of
40
these tasks for the Acrobot and the former task for the Pendubot using a single state
feedback.
3.3.2 The CartPole System and the Rotating Pendulum
The CartPole system shown in Figure 3.3.2 (a) consists of an inverted pendulum on
a cart. Figure 3.3.2 (b) shows the Rotating Pendulum [112, 6] which is an inverted
pendulum on a rotating arm. The inertia matrix for the CartPole system is
m
F
m , l , I
1
2 2 2
q
q
1
2
q1
q2
m2, l2, J2
m1, l1, J1
y
z
x
(a) (b)
Figure 32: (a) The CartPole system, (b) The Rotating Pendulum.
M(q
2
) =
¸
a
1
a
2
cos(q
2
)
a
2
cos(q
2
) a
3
and for the Rotating Pendulum the inertia matrix is
M(q
2
) =
¸
b
1
+ c
1
sin
2
(q
2
) b
2
cos(q
2
)
b
2
cos(q
2
) b
3
where all the a
i
, b
i
, c
i
> 0 are constants. Clearly, the only diﬀerence between the iner
tia matrices of these two mechanical systems is in the ﬁrst element m
11
(q
2
). Another
similarity between the CartPole system and the Rotating Pendulum is that both
have the same form of the potential energy
V (q
2
) = a
4
cos(q
2
)
with a
4
> 0. Swingup control design for the inverted pendulum in the CartPole sys
tem [21, 90] and the Rotating pendulum [112, 6] has been done by several researchers.
The drawback of the existing swingup design methods is that they are relatively slow.
In this work, we introduce aggressive swingup control designs for both the CartPole
system and the Rotating Pendulum using bounded control inputs. In addition, we
address asymptotic stabilization of both of these systems to their upright equilibrium
points using nonlinear state feedback.
41
3.3.3 The BeamandBall System
The BeamandBall system illustrated in Figure 33 consists of a beam and a ball on
it. The task is to bring the ball to the center of the beam by applying a torque τ
to the beam. The vertical distance between the center of mass of the ball and the
location of the external torque (i.e. control) is shown by d. The inertia matrix for
the beam and ball system takes the following form
M(q
2
) =
¸
a
1
+ a
2
q
2
2
−a
3
d
−a
3
d a
4
where all the a
i
’s are positive constants. Due to complexity of this system, stabiliza
d
r
I
τ
m , I
q
1
2
2
1
q
Figure 33: The BeamandBall System.
tion and tracking for the beamandball system using state or output feedback has
been considered by many researchers [34, 102, 103, 80, 79]. In [34], tracking for the
beamandball system was considered using approximate inputoutput linearization.
Semiglobal stabilization of the beamandball system using small nested saturations
(i.e. state feedback) was addressed by Teel in [102]. Also, stabilization of this system
using output feedback is due to Teel and Praly [103]. In [80], global stabilization of
the beamandball system with friction was achieved using a numeric version of the
method of construction of Lyapunov functions with crossterms which is originally
due to Mazenc and Praly [56]. Moreover, global stabilization of the beamandball
system as a perturbed chain of integrators with nontriangular (deﬁnition 3.8.4) ho
mogeneous higherorder perturbations was achieved in [79]. In all the aforementioned
works, the model of the beamandball system is taken from [34] with the assumption
that d = 0. In this work, we consider aggressive stabilization of the beamandball
system with d = 0 in chapter 7 and introduce the challenges in dealing with the case
d = 0.
3.3.4 The TORA System
The TORA (Translational Oscillator with Rotational Actuator) system was ﬁrst in
troduced in [109]. Figure 34 illustrates the TORA system consisting of a translational
oscillating platform with mass m
1
that is controlled via a rotational eccentric mass
42
m
2
. The inertia matrix for the system is in the form
M(q
2
) =
¸
a
1
a
2
cos(q
2
)
a
2
cos(q
2
) a
3
where the a
i
> 0 are constants and the potential energy is given by
V (q
1
, q
2
)) =
1
2
k
1
q
2
1
+ m
2
gr cos(q
2
)
Due to the fact that the TORA system is not fully feedback linearizable an an easy
m
m
r
q
k
1
2
2
τ
I
1
2
q
1
Figure 34: The TORA system.
change of coordinates was available to transform it to a cascade nonlinear system [109],
it received a tremendous attention by several researchers [37, 80, 38, 39]. In most of
these works the TORA system was used as a benchmark example for passivitybased
techniques and stabilization/tracking using output feedback in zero gravity, i.e. g = 0.
Here, we are interested in global stabilization of the TORA system in the presence of
gravity, i.e. g = 0.
3.3.5 The InertiaWheel Pendulum
The InertiaWheel Pendulum was ﬁrst introduced by Spong et al. in [93]. Figure
35 shows the inertia wheel pendulum that consists of a pendulum with a rotating
uniform inertiawheel at its end. The pendulum is unactuated and the system has
to be controlled via the rotating wheel. The task is to stabilize the pendulum in
its upright equilibrium point while the wheel stops rotating. The speciﬁc angle of
rotation of the wheel is not important. The inertia wheel pendulum is the ﬁrst example
of a ﬂat underactuated mechanical system with two degrees of freedom and a single
actuator. This is due to the constant inertia matrix of the InertiaWheel Pendulum.
In [93], an energybased method is used for the swingup of the pendulum and then
a supervisorybased switching strategy is employed to switch to a stabilizing local
nonlinear controller with a relatively large region of attraction. Here, we achieve
global stabilization of the upright equilibrium point of the inertia wheel pendulum
using a single nonlinear state feedback in analytically explicit form.
43
q
q
I
1
2
2
1
L , I
1
τ
Figure 35: The Inertia Wheel Pendulum
3.3.6 The VTOL Aircraft
The VTOL aircraft depicted in Figure 36 is a simpliﬁed planar model of a real vertical
take oﬀ and landing plane (e.g. the Harrier). The dynamics of the VTOL aircraft is
u
ε u
u
θ
1
2
2
x
y
Figure 36: The VTOL aircraft.
given in [35, 55] as the following
˙ x
1
= x
2
˙ x
2
= −u
1
sin(θ) + u
2
cos(θ)
˙ y
1
= y
2
˙ y
2
= u
1
cos(θ) + u
2
sin(θ) −g
˙
θ = ω
˙ ω = u
2
(3.3)
The VTOL aircraft is an underactuated system with three degrees of freedom and two
control inputs. In [35], it is assumed that [[ is relatively small and the VTOL aircraft
is treated as a slightly nonminimum phase system. Then, using approximate input
output linearization, bounded tracking is achieved. In section 5.10, we discuss global
conﬁguration stabilization for the VTOL aircraft with strong input coupling (i.e.
44
[[ 1). In this case, the system is strongly nonminimum phase and stabilization
or tracking for system (3.3) is a challenging problem.
3.4 Collocated Partial Feedback Linearization
The actuated conﬁguration vector for the underactuated mechanical system in (3.2)
is q
2
. The procedure of linearization of the dynamics of q
2
is called collocated partial
linearization which is due to Spong [90]. Spong has shown that all underactuated
systems in the form (3.2) can be globally partially linearized using a change of control.
Here, we restate this result as the following with our own phrasing due to applications
in this work.
Proposition 3.4.1. (Spong [90]) There exists a global invertible change of control in
the form
τ = α(q)u + β(q, ˙ q)
that partiallylinearizes the dynamics of (3.2) as the following
˙ q
1
= p
1
˙ p
1
= f
0
(q, p) + g
0
(q)u
˙ q
2
= p
2
˙ p
2
= u
(3.4)
where α(q) is an mm positive deﬁnite symmetric matrix and
g
0
(q) = −m
−1
11
(q)m
12
(q)
Proof. From the ﬁrst line of (3.2), we have
¨ q
1
= −m
−1
11
(q)h
1
(q, ˙ q) −m
−1
11
(q)m
12
(q)¨ q
2
which proves the expression for g
0
(q). Substituting this in the second line of (3.2),
we get
(m
22
(q) −m
21
(q)m
−1
11
(q)m
12
(q))¨ q
2
+ h
2
(q, ˙ q) −m
−1
11
(q)h
1
(q, ˙ q) = τ
Thus, deﬁning
α(q) = m
22
(q) −m
21
(q)m
−1
11
(q)m
12
(q)
β(q, ˙ q) = h
2
(q, ˙ q) −m
21
(q)m
−1
11
(q)h
1
(q, ˙ q)
and observing that α(q) is positive deﬁnite and symmetric ﬁnishes the proof.
The main property of the underactuated system in (3.4) is that after partial
feedback linearization, the new control u appears in the dynamics of both (q
1
, p
1
)
subsystem (i.e. nonlinear subsystem) and (q
2
, p
2
)subsystem (i.e. linear subsystem).
This is one of the main sources of the complexity of control design for underactuated
45
systems. In section 3.7, we introduce a global change of coordinates that decouples
these two subsystems while leaves the linear subsystem invariant.
Remark 3.4.1. Deﬁning x = (q
1
, p
1
, q
2
, p
2
), the dynamics of (3.4) can be rewritten as
˙ x = f(x) +g(x)u
with obvious deﬁnitions of f(x) and g(x). In [78], using explicit expressions of f(x)
and g(x), suﬃcient conditions are provided for smalltime local controllability (STLC)
[97] of a special class of underactuated mechanical systems in the form (3.4).
3.5 Noncollocated Partial Feedback Linearization
In this section, we present a partial feedback linearization procedure for underactu
ated systems that linearizes the dynamics of the unactuated conﬁguration variables.
We show this is possible if the number of the control inputs is greater or equal to the
number of unactuated conﬁguration variables. We call this procedure noncollocated
partial feedback linearization. To be more precise, consider the following underactu
ated system
m
00
(q) m
01
(q) m
02
(q)
m
10
(q) m
11
(q) m
12
(q)
m
20
(q) m
21
(q) m
22
(q)
¸
¸
¨ q
0
¨ q
1
¨ q
2
¸
¸
+
h
0
(q, ˙ q)
h
1
(q, ˙ q)
h
2
(q, ˙ q)
¸
¸
=
τ
0
τ
1
0
¸
¸
(3.5)
where q = (q
0
, q
1
, q
2
) ∈ R
n
0
R
n
1
R
n
2
with n
1
= n
2
= m and n
0
= n −2m ≥ 0.
Remark 3.5.1. For the special case where there are equal number of actuated and
unactuated variables n = 2m and n
0
= 0. Thus, q
0
has dimension zero or q = (q
1
, q
2
).
Proposition 3.5.1. Consider the underactuated mechanical system in (3.5). Then,
there exists a change of control in the form
τ = α
1
(q)u + β
1
(q, ˙ q)
where τ = col(τ
0
, τ
1
), u = col(u
0
, u
1
) and
u
1
= α
0
(q)u
0
+ α
2
(q)u
2
+β
2
(q, ˙ q)
that partially linearizes the dynamics of (3.5) as the following
˙ q
0
= p
0
˙ p
0
= u
0
˙ q
1
= p
1
˙ p
1
= f
0
(q, p) + g
0
(q)u
0
+ g
2
(q)u
2
˙ q
2
= p
2
˙ p
2
= u
2
(3.6)
46
over the set
U = ¦q ∈ R
n
[ det(m
21
(q)) = 0¦
where
f
0
(q, p) = −m
−1
21
(q)h
2
(q, p)
g
0
(q) = −m
−1
21
(q)m
20
(q)
g
2
(q) = −m
−1
21
(q)m
22
(q)
Proof. Apparently, based on collocated feedback linearization (proposition 3.4.1)
the dynamics of the actuated conﬁguration vector (q
0
, q
1
) can be linearized using a
change of control τ = α
1
(q)u + β
1
(q, ˙ q) that gives
¨ q
0
= u
0
¨ q
1
= u
1
Over the set U, m
21
(q) is an invertible matrix and from the last equation in (3.5), we
get
¨ q
1
= −m
−1
21
(q)h
2
(q, ˙ q) −m
−1
21
(q)m
20
(q)¨ q
0
−m
−1
21
(q)m
22
(q)¨ q
2
Thus, after a second change of control as
u
1
= −m
−1
21
h
2
(q, ˙ q) −m
−1
21
(q)m
20
(q)u
0
−m
−1
21
(q)m
22
(q)u
2
where u
2
is the new control, the result follows.
3.6 Partial Feedback Linearization Under Input
Coupling
Consider the underactuated mechanical control system
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= F(q)τ
with conﬁguration q ∈ R
n
and control τ ∈ R
m
and rank F(q) = m < n. Without loss
of generality, F(q) can be written as
F(q) =
¸
F
1
(q)
F
2
(q)
such that F
2
(q) is an invertible mm matrix and q can be decomposed to (q
1
, q
2
) ∈
R
n−m
R
m
according to F(q). By input coupling, we mean F
1
(q) ≡ 0 for all q. In the
following, we provide the conditions which allow partial feedback linearization for an
underactuated (simple) Lagrangian system.
Proposition 3.6.1. Consider the following underactuated system with input coupling
47
(i.e. F
1
(q) ≡ 0, det(F
2
(q)) = 0 for all q)
¸
m
11
(q) m
12
(q)
m
21
(q) m
22
(q)
¸
¨ q
1
¨ q
2
+
¸
h
1
(q, ˙ q)
h
2
(q, ˙ q)
=
¸
F
1
(q)
F
2
(q)
τ (3.7)
and assume the following matrix is invertible for all q
Λ(q) = F
2
(q) −m
21
(q)m
−1
11
(q)F
1
(q) (3.8)
Then, there exists a change of control in the form
τ = α(q)u + β(q, ˙ q)
with
α(q) = Λ
−1
(q)[m
22
(q) −m
21
(q)m
−1
11
(q)m
12
(q)]
β(q, ˙ q) = Λ
−1
(q)[h
2
(q, ˙ q) −m
21
(q)m
−1
11
(q)h
1
(q, ˙ q)]
that partially linearizes (3.7) as the following
˙ q
1
= p
1
˙ p
1
= f
0
(q, p) + g
0
(q)u
˙ q
2
= p
2
˙ p
2
= u
where
f
0
(q, p) = m
−1
11
(q)[F
1
(q)β(q, ˙ q) −h
1
(q, ˙ q)]
g
0
(q) = m
−1
11
(q)[F
1
(q)α(q) −m
12
(q)]
Proof. From the ﬁrst equation in (3.7), we get
¨ q
1
= −m
−1
11
m
12
¨ q
2
−m
−1
11
h
1
+ m
−1
11
F
1
substituting this in the second line of (3.7) gives
(m
22
−m
21
m
−1
11
m
12
)¨ q
2
+ h
2
−m
−1
11
h
1
= (F
2
−m
21
m
−1
11
F
1
)τ = Λ(q)τ
Thus, applying τ = α(q)u + β(q, ˙ q) with α, β as deﬁned in the question partially
linearizes the system and the ﬁrst equation of (3.7) reduces to
¨ q
1
= m
−1
11
(q)[F
1
(q)β(q, ˙ q) −h
1
(q, ˙ q)] +m
−1
11
(q)[F
1
(q)α(q) −m
12
(q)]u
and the result follows.
Remark 3.6.1. The partial feedback linearization procedure in proposition 3.6.1 is
particularly useful for autonomous vehicles in SE(3) like an aircraft or a helicopter
with six degrees of freedom.
48
3.7 Normal Forms for Underactuated Systems
The complexity of control design for underactuated mechanical systems is partly due
to the fact that the control input u appears in the dynamics of both the unactu
ated subsystem (i.e. (q
1
, p
1
)subsystem) and the actuated subsystem (i.e. (q
2
, p
2
)
subsystem) in (3.4). In this section, we provide a method to decouple these two
subsystems w.r.t. the control input u using a global change of coordinates. The
following result provides an appropriate structure and suﬃcient conditions for the
existence of a decoupling change of coordinates.
Theorem 3.7.1. Consider an underactuated mechanical system with an inertia ma
trix M(q) = ¦m
ij
(q)¦; i, j = 1, 2 where q = (q
1
, q
2
) and q
1
= ¦q
i
1
¦ ∈ R
n
and
q
2
= ¦q
j
2
¦ ∈ R
m
denote the unactuated and actuated conﬁguration variables, respec
tively. Denote
g
0
(q) = −m
−1
11
(q)m
12
(q)
and
g(q) =
¸
g
0
(q)
I
m×m
where g
0
(q) = (g
1
0
(q), . . . , g
m
0
(q)) with g
j
0
(q) ∈ R
n
, j = 1, . . . , m and I
m×m
is the
identity matrix. Deﬁne the following distribution
∆(q) = span¦columns of g(q)¦
that has full column rank and thus is globally nonsingular. Then, a necessary and
suﬃcient condition for the distribution ∆(q) to be globally involutive (i.e. completely
integrable) is that
∂g
j
0
(q)
∂q
1
g
i
0
(q) −
∂g
i
0
(q)
∂q
1
g
j
0
(q) +
∂g
j
0
(q)
∂q
i
2
−
∂g
i
0
(q)
∂q
j
2
= 0, ∀i, j = 1, . . . , m (3.9)
In addition, if condition (3.9) holds, there exists a global change of coordinates given
by
z
1
= Φ(q
1
, q
2
)
z
2
= (D
q
1
Φ(q)) p
1
+ (D
q
2
Φ(q)) p
2
ξ
1
= q
2
ξ
2
= ˙ q
2
that transforms the dynamics of the system into the normal form
˙ z
1
= z
2
˙ z
2
= f(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(3.10)
Remark 3.7.1. Normal form (3.10) is a special case of the famous ByrnesIsidori
49
normal form [36] with a double integrator as the following
˙ z = f(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(3.11)
Remark 3.7.2. The main advantage of the normal form (3.10) is that the control input
of the actuated subsystem of the original system does not appear in the unactuated
subsystem. This simpliﬁes control design for this underactuated system by reducing
the control of the original higher order system into control of its lowerorder nonlinear
unactuated subsystem. This is due to a recent result by the author in [67] on control
design for systems in normal forms (3.10) and (3.11) which is explained in chapter 7.
Remark 3.7.3. The main disadvantage of theorem 3.7.1 is that it is not always possible
to ﬁnd Φ(q
1
, q
2
) in an analytical explicit form. Later, we show that under kinetic
symmetry properties of the system it is possible to calculate Φ(q
1
, q
2
) in closedform.
Proof.(Theorem 3.7.1) Note that ∆(q) globally has a full column rank of m and is
therefore a globally nonsingular distribution (see [36] for deﬁnitions and notations in
this proof). Calculating the Lie bracket of the ith and jth columns of g(q), we get
[g
i
(q), g
j
(q)] =
∂g
j
0
(q)
∂q
1
∂g
j
0
(q)
∂q
2
∂e
j
∂q
1
∂e
j
∂q
2
¸
¸
¸
¸
¸
g
i
0
(q)
e
i
−
∂g
i
0
(q)
∂q
1
∂g
i
0
(q)
∂q
2
∂e
i
∂q
1
∂e
i
∂q
2
¸
¸
¸
¸
g
j
0
(q)
e
j
=
∂g
j
0
(q)
∂q
1
g
i
0
(q) −
∂g
i
0
(q)
∂q
1
g
j
0
(q) +
∂g
j
0
(q)
∂q
2
e
i
−
∂g
i
0
(q)
∂q
2
e
j
0
m×1
¸
¸
which based on condition (3.9) implies [g
i
(q), g
j
(q)] = 0 for all i, j, q. Therefore ∆(q)
is globally involutive. To prove the converse, assume ∆(q) is globally involutive.
Then, for all i, j, q, [g
i
(q), g
j
(q)] can be expressed as a linear combination of g
k
(q)’s.
But the lower m1 block of [g
i
(q), g
j
(q)] is identically zero and linearly independent
of e
k
’s. This means that [g
i
(q), g
j
(q)] = 0 and thus condition (3.9) holds. Now, we
prove the rest of the theorem. Based on Frobenius theorem (see [36]), because ∆(q)
is a globally nonsingular and involutive distribution, the following equation
∂φ
∂q
g(q) = 0
has d = (n − m) linearly independent solutions φ
k
(q
1
, q
2
), k = 1, . . . , n. Denoting
Φ(q
1
, q
2
) = (φ
1
, . . . , φ
d
), Φ satisﬁes the property
∂Φ
∂q
1
g
0
(q) +
∂Φ
∂q
2
= 0.
50
After applying the change of coordinates
z
1
= Φ(q
1
, q
2
), z
2
= ˙ z
1
we get
z
1
= Φ(q
1
, q
2
), z
2
=
∂Φ
∂q
1
p
1
+
∂Φ
∂q
2
p
2
and because ∂Φ/∂q
1
is globally nonsingular (due to the proof of Frobenius theorem),
based on implicit mapping theorem, there exists a smooth function Ψ such that
q
1
= Ψ(z
1
, ξ
1
), p
1
=
∂Φ
∂q
1
[
q=(Ψ(z
1
,ξ
1
),ξ
1
)
−1
(z
2
−ξ
2
∂Φ
∂q
2
[
q=(Ψ(z
1
,ξ
1
),ξ
1
)
)
(we drop the substitution q = (Ψ(z
1
, ξ
1
), ξ
1
) in ∂Φ/∂q
i
due to the simplicity of nota
tion). Calculating ˙ z
2
as the following
˙ z
2
=
∂Φ
∂q
1
f
0
(q, p) +
∂
2
Φ
∂q
2
1
(
∂Φ
∂q
1
)
−2
(z
2
−
∂Φ
∂q
1
ξ
2
)
2
+
∂
2
Φ
∂q
2
2
ξ
2
2
+ (
∂Φ
∂q
1
g
0
(q)) +
∂Φ
∂q
2
)u (3.12)
and noting that the coeﬃcient of u in the last equation is identically zero, we get
˙ z
2
= f(z, ξ
1
, ξ
2
)
where f is the right hand side of (3.12). Therefore, the dynamics of the system in
new coordinates is in normal form (3.10) and this ﬁnishes the proof.
Here is an important corollary of theorem 3.7.1:
Corollary 3.7.1. All underactuated mechanical systems with a single actuator that
are globally partial feedback linearizable can be transformed into normal form (3.10)
using a global invertible change of coordinates.
Proof. For this case m = 1 and i = j = 1. By symmetry w.r.t i and j indices,
condition (3.9) globally holds for all q.
3.8 Classes of Structured Nonlinear Systems
In this section, we deﬁne important classes of structured nonlinear systems that ap
pear frequently in this thesis.
Deﬁnition 3.8.1. (cascade system) We say a nonlinear system is in cascade form if
it has the following structure
˙ z = f(z, ξ)
˙
ξ = g(ξ, u)
(3.13)
where f : R
n
R
m
→ R
n
, g : R
m
R
p
→ R
m
, (z, ξ) is the composite state, and u
is the control input. When the second equation is a linear timeinvariant system, i.e.
˙
ξ = Aξ + Bu, (3.13) is called a partially linear cascade nonlinear system.
51
Deﬁnition 3.8.2. (feedback form) We say a nonlinear system is in strict feedback
form [57], if it has the following triangular structure
˙ z = f(z, ξ
1
),
˙
ξ
1
= ξ
2
,
. . .
˙
ξ
m
= u,
Deﬁnition 3.8.3. (feedforward form) We say a nonlinear system is in feedforward
form [102], if it possesses the following triangular structure
˙ x
1
= x
2
+ ϕ
1
(x
2
, . . . , x
n
, u)
˙ x
2
= x
3
+ ϕ
2
(x
3
, . . . , x
n
, u)
. . .
˙ x
n
= u + ϕ
n
(x
n
, u)
where x ∈ R
n
and the ϕ
i
’s are at least quadratic in (x, u).
Deﬁnition 3.8.4. (nontriangular form) We say a partially linear cascade nonlinear
system is in nontriangular form, if it has the following structure
˙ z = f(z, ξ
1
, ξ
2
, . . . , ξ
m
),
˙
ξ
1
= ξ
2
,
. . .
˙
ξ
m
= u,
(3.14)
where z ∈ R
n
and u ∈ R
p
. The nonlinear system in (3.14) is also called ByrnesIsidori
normal form [36].
Deﬁnition 3.8.5. (nontriangular linearqudratic form) We say a cascade nonlinear
system is in nontriangular linearquadratic form, if it has the following structure
˙ z
1
= µ(z
1
)z
2
+ η(ξ
1
)ξ
2
˙ z
2
= φ(z
1
, ξ
1
) + Σ(ξ
1
, z
2
, ξ
2
)
˙
ξ
1
= ξ
2
,
˙
ξ
2
= u,
(3.15)
where z
1
, z
2
∈ R
n
, u ∈ R
p
, µ(z
1
) is a positive deﬁnite matrix, φ : R
n
R
p
→R
n
, and
Σ : R
p
R
n
R
p
→R
n
has the following quadratic form in (z
2
, ξ
2
)
Σ(ξ
1
, z
2
, ξ
2
) =
¸
z
2
ξ
2
T
Π(ξ
1
)
¸
z
2
ξ
2
where Π = (Π
1
, . . . , Π
n
) is a cubic matrix with elements in R
n×n
and for v ∈ R
n
,
v
T
Πv := (v
T
Π
1
v, . . . , v
T
Π
n
v)
T
∈ R
n
. If η ≡ 0, (3.15) is called a nontriangular
52
quadratic form, or in case Σ ≡ 0, (3.15) is called nontrinagular linear form. In all
cases, the quadratic or linearity properties are implicitly w.r.t. ξ
2
.
Remark 3.8.1. Clearly, if Σ(ξ
1
, z
2
, ξ
2
) = 0, then the nonlinear system in (3.15) reduces
to a strict feedback system. In addition, if µ(z
1
) = I
n
, f(z
1
, ξ
1
) = ξ
1
, and Π
(1,1)
(ξ
1
) =
0 (Π
(1,1)
is the ﬁrst cubic partition of Π according to (z
2
, ξ
2
)), then (3.15) reduces to a
nonlinear system in strict feedforward form. Thus, the famous feedback/feedforward
triangular forms are special cases of the quadratic nontriangular form.
Remark 3.8.2. In the special case where z
1
, z
2
, ξ
1
, ξ
2
are all scalars, Π(ξ
1
) in deﬁnition
3.8.5 is a 22 matrix (i.e. not a cubic matrix) which is in fact the case for underactu
ated mechanical systems with two DOF. However, we need the generality of deﬁnition
3.8.5 for further applications related to normal forms of higherorder underactuated
systems.
3.9 Normal Forms for Underactuated Systems with
2 DOF
In this section, we introduce explicit cascade normal forms for underactuated me
chanical systems with two degrees of freedom and kinetic symmetry w.r.t. one
(actuated or unactuated) conﬁguration variable. The importance of these normal
forms is due to the fact that many examples of benchmark problems in nonlin
ear control design including the cartpole system [101, 56, 94], the beamandball
system [34, 102, 103, 80, 79], the acrobot [13, 86], the pendubot [92], the rotating
pendulum[6, 10], the inertiawheel pendulum [93], and the TORA example [41, 37]
all fall within the class of underactuated systems with two degrees of freedom and
kinetic symmetry. Nevertheless, so far no (global) cascade normal forms are known
for the majority of these nonlinear systems except for the beamandball system, and
the TORA example. This motivated us to provide a (controloriented) classiﬁcation
of underactuated systems with two DOF based on their associated cascade normal
forms. These cascade forms are systems in strict feedback form [57], feedforward
form [102], and nontriangular quadratic form [69] (to be deﬁned). The main ben
eﬁt of this classiﬁcation and the associated normal form is that if the system has
feedback/feedforward triangular structure (under appropriate conditions) it can be
globally asymptotically stabilized using backstepping [36, 57, 80] or forwarding pro
cedures [102, 101, 56, 80]. The only open ended case is stabilization of systems in
nontriangular quadratic form that will be addressed in this thesis in chapter 7.
Now, consider an underactuated mechanical system with two DOF and conﬁgu
ration vector q = (q
1
, q
2
)
T
. Assume the inertia matrix of this system only depends
on q
2
, i.e. M = M(q
2
). Then, the system has kinetic symmetry w.r.t. q
1
. The
Lagrangian of this underactuated system can be expressed as the following
L(q, ˙ q) =
1
2
˙ q
T
¸
m
11
(q
2
) m
12
(q
2
)
m
21
(q
2
) m
22
(q
2
)
˙ q −V (q) (3.16)
53
Here is our main result on normal forms for underactuated systems with two DOF
and kinetic symmetry:
Theorem 3.9.1. All underactuated mechanical systems with two degrees of freedom
(q
1
, q
2
) and kinetic symmetry w.r.t. q
1
have a Lagrangian given by (3.16) and can be
(possibly globally) transformed into quadratic nontriangular normal form (3.15) (or
(3.10)) using a change of coordinates in analytically explicit form.
The proof of this theorem relays on the corresponding proofs for two cases where
either q
2
is actuated, or q
2
is unactuated. This provides a natural classiﬁcation of
underactuated systems to two classes as the following.
Deﬁnition 3.9.1. (ClassI,II underactuated systems) We call an underactuated sys
tem with Lagrangian (3.16) ClassI underactuated system, iﬀ q
2
is actuated. We call
it a ClassII underactuated system, iﬀ q
2
is unactuated.
The EulerLagrange equations of motion for this underactuated system are
d
dt
∂L
∂ ˙ q
1
−
∂L
∂q
1
= τ
1
d
dt
∂L
∂ ˙ q
2
−
∂L
∂q
2
= τ
2
(3.17)
where for a Class–I system τ
1
= 0 and for a Class–II system τ
2
= 0. Equation (3.17)
can be rewritten as
m
11
(q
2
)¨ q
1
+ m
12
(q
2
)¨ q
2
+ m
11
(q
2
) ˙ q
1
˙ q
2
+ m
12
(q
2
) ˙ q
2
2
−g
1
(q
1
, q
2
) = τ
1
m
21
(q
2
)¨ q
1
+ m
22
(q
2
)¨ q
2
−
1
2
m
11
(q
2
) ˙ q
2
1
+
1
2
m
22
(q
2
) ˙ q
2
2
−g
2
(q
1
, q
2
) = τ
2
(3.18)
where g
i
(q
1
, q
2
) = −∂V (q)/∂q
i
, i = 1, 2 and
denotes d/dq
2
. The following propo
sition states that all Class–I underactuated systems can be globally and explicitly
transformed into cascade nonlinear systems in strict feedback form.
Proposition 3.9.1. (ClassI normal form) Consider a ClassI underactuated system
with two DOF. Then, the following global change of coordinates obtained from the
Lagrangian of the system
z
1
= q
1
+ γ(q
2
)
z
2
= m
11
(q
2
)p
1
+m
12
(q
2
)p
2
=
∂L
∂ ˙ q
1
ξ
1
= q
2
ξ
2
= p
2
(3.19)
where
γ(q
2
) =
q
2
0
m
−1
11
(θ)m
12
(θ)dθ (3.20)
54
transforms the dynamics of the system into a cascade nonlinear system in strict feed
back form
˙ z
1
= m
−1
11
(ξ
1
)z
2
˙ z
2
= g
1
(z
1
−γ(ξ
1
), ξ
1
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(3.21)
where u is the new control from collocated partial feedback linearization and g
1
(q
1
, q
2
) =
−∂V (q)/∂q
1
.
Proof. By deﬁnition of z
1
and z
2
, we have ˙ z
1
= m
−1
11
(q
2
)z
2
. Noting that z
2
= ∂L/∂ ˙ q
1
,
from the ﬁrst line of (3.17), τ
1
= 0 and we get
˙ z
2
=
∂L
∂q
1
=
∂K
∂q
1
−
∂V
∂q
1
= g
1
(q
1
, q
2
)
The last equality follows from kinetic symmetry property of the system w.r.t. q
1
(i.e.
∂K/∂q
1
= 0).
Remark 3.9.1. After a second change of coordinates
y
1
= z
1
, y
2
= m
−1
11
(ξ
1
)z
2
the dynamics of the system in (3.21) transforms into normal form (3.10).
Corollary 3.9.1. The Acrobot, the InertiaWheel pendulum, and the TORA example
are all Class–I underactuated systems and can be globally transformed into normal
form (3.21) using an explicit global change of coordinates.
Proposition 3.9.2. (Class–II normal form) Consider a ClassII underactuated sys
tem with two DOF. Then, the following change of coordinates obtained from the La
grangian of the system
z
1
= q
1
+ γ(q
2
)
z
2
= m
21
(q
2
)p
1
+m
22
(q
2
)p
2
=
∂L
∂ ˙ q
2
ξ
1
= q
2
ξ
2
= p
2
(3.22)
where
γ(q
2
) =
q
2
0
m
−1
21
(θ)m
22
(θ)dθ (3.23)
is deﬁned over
U = ¦q
2
[ m
21
(q
2
) = 0¦
transforms the dynamics of the system into a cascade nonlinear system in quadratic
nontriangular form where u is the new control from noncollocated partial feedback
linearization.
55
Proof. By deﬁnition of z
1
and z
2
, it follows that ˙ z
1
= z
2
/m
21
(q
2
). From the second
line of (3.16), τ
2
= 0 and we get
˙ z
2
= ∂L/∂q
2
= ∂K/∂q
2
−∂V (q)/∂q
2
(∂K/∂q
2
≡ 0)
= g
2
(q
1
, q
2
) +
1
2
m
11
(q
2
)p
2
1
+ m
21
(q
2
)p
1
p
2
+
1
2
m
22
(q
2
)p
2
2
Hence, after substituting
q
1
= z
1
−γ(q
2
)
p
1
= (z
2
−m
22
(q
2
)p
2
)/m
21
(q
2
)
in the last equation, we obtain
˙ z
2
= g
2
(z
1
−γ(q
2
), q
2
) +
m
11
(q
2
)
2m
2
21
(q
2
)
[z
2
−m
22
(q
2
)p
2
]
2
+
m
21
(q
2
)
m
21
(q
2
)
[z
2
−m
22
(q
2
)p
2
]p
2
+
m
22
(q
2
)
2
p
2
2
which gives the following quadratic nontriangular normal form for ClassII underac
tuated systems
˙ z
1
= m
−1
21
(q
2
)z
2
˙ z
2
= g
2
(z
1
−γ(q
2
), q
2
)
+
m
11
(q
2
)
2m
2
21
(q
2
)
z
2
2
+
m
21
(q
2
)
m
21
(q
2
)
−
m
22
(q
2
)m
11
(q
2
)
2m
2
21
(q
2
)
z
2
p
2
+
m
2
22
(q
2
)
2m
2
21
(q
2
)
m
11
(q
2
) −
m
22
(q
2
)
m
21
(q
2
)
m
21
(q
2
) +
1
2
m
22
(q
2
)
p
2
2
˙ q
2
= p
2
˙ p
2
= u
(3.24)
Corollary 3.9.2. The Pendubot, the CartPole system, the Rotating Pendulum, and
the BeamandBall system are all ClassII underactuated systems with two degrees of
freedom and can be transformed into a cascade nonlinear system in quadratic nontri
angular form over the set U = ¦q
2
[ m
21
(q
2
) = 0¦ using a change of coordinates in
explicit form.
Proof. The proof is by direct substitution of the elements of the inertia matrix of
each system in equation (3.24).
Proposition 3.9.3. (feedforward normal form) Consider a Class–II underactuated
system with two degrees of freedom over the set U = ¦q
2
[ m
21
(q
2
) = 0¦ and assume
the following conditions hold:
i) g
2
(q
1
, q
2
) is independent of q
1
, i.e. D
q
1
D
q
2
V (q) = 0.
ii) m
11
is constant.
iii) ψ(q
2
) = g
2
(q
2
)/m
21
(q
2
) satisﬁes ψ
(0) = 0.
56
Then, applying the change of coordinates
y
1
= z
1
, y
2
= z
2
/m
21
(q
2
)
transforms the normal form (3.24) for ClassII underactuated systems with two de
grees of freedom into the following cascade nonlinear system in feedforward form
˙ z
1
= z
2
˙ z
2
= ψ(q
2
) +
m
22
(q
2
)
2m
21
(q
2
)
−
m
22
(q
2
)
m
2
21
(q
2
)
m
21
(q
2
)
p
2
2
˙ q
2
= p
2
˙ p
2
= u
(3.25)
In addition, the origin for this feedforward system can be globally asymptotically sta
bilized using nested saturations [102].
Proof. The normal form (3.25) follows from the proof of proposition 3.9.2 by direct
calculation and conditions i), ii). The stabilization using nested saturations follows
from condition iii) and [102].
Corollary 3.9.3. The CartPole system can be transformed into a cascade nonlinear
system in feedforward form using the change of coordinates in proposition 3.9.3 and
can be globally asymptotically stabilized to its upright equilibrium point over q
2
∈
(−π/2, π/2) using nested saturations.
Proof. For the cartpole system, g
2
= g
2
(q
2
) and m
11
is constant. In addition,
ψ(q
2
) = g tan(q
2
) (g is the gravity constant) that implies ψ
(0) = g = 0. Thus,
the cartpole system satisﬁes all three conditions in proposition 3.9.3. Also, because
m
21
(q
2
) = a cos(q
2
) (a > 0 is a constant), the stabilization result over q
2
∈ (−π/2, π/2)
follows (see section 5.4 for further details on control design for the cartpole system).
57
Chapter 4
Reduction and Control of
HighOrder Underactuated
Systems
In this chapter, we address reduction of highorder underactuated mechanical systems
with kinetic symmetry. By reduction, we mean control of the original higherorder
underactuated system reduces to control of a lowerorder nonlinear system (we pre
cisely deﬁne reduction later on). It turns out that actuation or lack of actuation of a
subset of conﬁguration variables that somehow represent the “shape” of a mechanical
system plays an important role in reduction of underactuated mechanical systems.
The details regarding the notion of “shape variables” for Lagrangian systems and
“kinetic symmetry” are discussed in the following section.
Our main contribution in this chapter is to obtain structured cascade normal forms
for higherorder underactuated systems in explicit form and present an appropriate
control design method for each normal form. These cascade normal forms have struc
tural properties that allow eﬀective use of existing and recently developed systematic
nonlinear control design methods. Types of cascade normal forms that are explic
itly obtained for underactuated systems include nonlinear systems in strict feedback
form, feedforward form, and nontriangular linear/quadratic form (see section 3.8 for
the deﬁnitions).
4.1 Shape Variables and Kinetic Symmetry
Consider a simple Lagrangian system with conﬁguration vector q ∈ Q. The variables
that appear in the inertia matrix of the system are called shape variables. If a con
ﬁguration variable q
j
does not appear in the inertia matrix i.e. ∂M(q)/∂q
j
= 0, it is
called an external variable. This implies that for an external variable q
j
the following
identity holds
∂K(q, ˙ q)
∂q
j
= 0
58
In other words,a simple Lagrangian system has kinetic symmetry w.r.t. external vari
ables. A more general way to deﬁne this symmetry is via action of a group on a
manifold. Let q
j
∈ G where G is a cyclic group, e.g. R (a translation group), or S
1
(a
rotation group). Decompose the conﬁguration manifold as Q = BG. The action of
G on Q is a mapping Φ : QG →Q. For a ﬁx j ∈ ¦1, . . . , n¦, let α ∈ G and q ∈ Q,
deﬁne Φ(q, α) = (q
1
, . . . , q
j
+ α, . . . , q
n
). Formally, we call q
j
an external variable, if
M(q) = M(Φ(q, α)). Let Q
x
= G
1
. . . G
n
be the conﬁguration manifold of all
external variables where G
i
’s are onedimensional cyclic groups. Then Q = Q
x
Q
s
where Q
s
is called the shape space, i.e. the conﬁguration manifold of the shape vari
ables. The notion of shape variables or internal variables originally appeared in the
control literature by the study of interconnected mechanical systems and multibody
problems [44, 45, 46, 51].
In the classical sense, symmetry in mechanics is deﬁned as the invariance of the
Lagrangian under the action of a group. This is fundamentally diﬀerent from the
invariance of the kinetic energy. The classical symmetry gives raise to the existence
of conserved quantities while the kinetic symmetry does not give raise to the conser
vation of the generalized momentums. (unless both τ
j
and ∂V (q)/∂q
j
( i.e. the jth
gravity term) vanish in which case ∂L/∂ ˙ q
j
is a conserved quantity). The fact that
the generalized momentum p
j
= ∂L/∂ ˙ q
j
is not a conserved quantity in the presence
of the kinetic symmetry plays a vital role in controllability and stabilization of broad
classes of underactuated mechanical systems (see remark 4.2.4).
Throughout this chapter, we consider underactuated mechanical control systems
with kinetic symmetry that have the following Lagrangian
L(q, ˙ q) =
1
2
˙ q
T
M(q
s
) ˙ q −V (q) (4.1)
where q = (q
x
, q
s
) ∈ Q = Q
x
Q
s
is the conﬁguration vector decomposed to external
and shape variables. The forced EulerLagrange equations of motion for this system
is
d
dt
∂L
∂ ˙ q
x
−
∂L
∂q
= F
x
(q)τ
d
dt
∂L
∂ ˙ q
x
−
∂L
∂q
= F
s
(q)τ
(4.2)
where τ ∈ R
m
and F(q) = col(F
x
(q), F
s
(q)) is the force matrix with the underactua
tion property
rank F(q) = m < n = dim(Q)
In analysis and control design for the underactuated system (4.2), a number of
cases arise depending on whether the shape conﬁguration vector q
s
is fullyactuated,
partiallyactuated, or unactuated and the presence or lack of any input couplings due
to the force matrix F(q). This leads to analysis of a ﬁnite number of cases. For each
case, we provide a method for reduction of the given underactuated system and global
transformation of the system into a cascade normal form. Eventually, this leads to
classiﬁcation of underactuated control systems based on four basic properties (to be
59
mentioned later). It turns out that there are eight diﬀerent classes of underactuated
systems and three types of obtained normal forms. Namely, cascade nonlinear systems
in strict feedback form, strict feedforward form, and nontriangular linearquadratic
form.
4.2 Underactuated Systems with Noninteracting
Inputs and Integrable Momentums
In this section, we address reduction of highorder underactuated systems with in
tegrable normalized momentums for the noninteracting input case. There could be
three cases: i) underactuated systems with fullyactuated shape variables, ii) under
actuated systems with unactuated shape variables, and iii) underactuated systems
with partiallyactuated shape variables. We address reduction of cases i) and ii) in
the following. Case iii) is rather similar to Case ii) and will not be discussed in this
section. Later, we present reduction of underactuated systems with nonintegrable
momentums in case iii).
4.2.1 Underactuated Systems with Actuated Shape Variables
First, we consider underactuated mechanical systems in (4.2) with fullyactuated
shape variables and noninteracting inputs i.e. F
x
(q) = 0, F
s
(q) = I
m
.
Theorem 4.2.1. Consider an underactuated system with kinetic symmetry w.r.t. q
x
and fullyactuated shape variables q
s
m
xx
(q
s
)¨ q
x
+ m
xs
(q
s
)¨ q
s
+ h
x
(q, ˙ q) = 0
m
sx
(q
s
)¨ q
x
+ m
ss
(q
s
)¨ q
s
+ h
s
(q, ˙ q) = τ
(4.3)
Let τ = α(q
s
)u + β(q, ˙ q) be the collocated partially linearizing change of control for
(4.3). Assume all the elements of
ω = m
−1
xx
(q
s
)m
xs
(q
s
)dq
s
(4.4)
are exact oneforms and let ω = dγ(q
s
). Then, there exists a global change of coordi
nates (i.e. diﬀeomorphism) obtained from the Lagrangian of the system
q
r
= q
x
+ γ(q
s
) =: Φ(q
x
, q
s
)
p
r
= m
xx
(q
s
) ˙ q
x
+ m
xs
(q
s
) ˙ q
s
=
∂L
∂ ˙ q
x
(4.5)
that transforms the dynamics of the underactuated system in (4.3) into a cascade
60
normal form in strict feedback form
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= g
r
(q
r
, q
s
)
˙ q
s
= p
s
˙ p
s
= u
(4.6)
In addition, the (q
r
, p
r
)subsystem is a Lagrangian system with conﬁguration vector
q
r
that belongs to the reduced manifold Q
r
= Φ(Q
x
Q
s
) with reduced Lagrangian
L
r
(q
r
, ˙ q
r
, q
s
) =
1
2
˙ q
T
r
m
r
(q
s
) ˙ q
r
−V
r
(q
r
, q
s
) (4.7)
that satisﬁes the unforced EulerLagrange equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= 0
with
m
r
(q
s
) := m
xx
(q
s
)
V
r
(q
r
, q
s
) := V (q
r
−γ(q
s
), q
s
)
g
r
(q
r
, q
s
) := −∂V
r
(q
r
, q
s
)/∂q
r
Proof. By deﬁnition of q
r
, p
r
, we have
˙ q
r
= m
−1
xx
(q
s
)p
r
= m
−1
r
(q
s
)p
r
From the ﬁrst line of EulerLagrange equation in (4.2), we get
˙ p
r
=
∂L
∂q
x
=
∂K
∂q
x
−
∂V (q
x
, q
s
)
∂q
x
=: g
x
(q
x
, q
r
)
due to kinetic symmetry w.r.t. q
x
, ∂K/∂q
x
= 0. But
g
x
(q
x
, q
s
) = −
∂V (q
r
−γ(q
s
), q
s
)
∂q
r
∂q
r
∂q
x
= g
r
(q
r
, q
s
) I
n−m
= g
r
(q
r
, q
s
)
and the second line of (4.6) follows. By direct calculation
p
r
=
∂L
r
∂ ˙ q
r
, ˙ p
r
=
∂L
r
∂q
r
which proves the reduced Lagrangian L
r
satisﬁes the unforced EulerLagrange equa
tion.
61
Remark 4.2.1. Stabilization of the reduced system
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= g
r
(q
r
, q
s
)
(4.8)
is addressed in section 4.7 of this chapter.
Remark 4.2.2. After renaming the variables in (4.6) as (z
1
, z
2
) = (q
r
, p
r
) and (ξ
1
, ξ
2
) =
(q
s
, p
s
), one obtains
˙ z = f(z, ξ
1
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.9)
which is a special case of ByrnesIsidori normal form [36] associated to the output
ξ
1
= q
s
which has global uniform relative degree two and the following Lagrangian
zerodynamics
˙ z
1
= m
−1
r
(ξ
1
)z
2
˙ z
2
= g
x
(z
1
−γ(ξ
1
), ξ
1
)
Remark 4.2.3. Based on standard backstepping procedure, control design for the
nonlinear system in (4.3) reduces to control design for its (q
r
, p
r
)subsystem with
control input q
s
. For this reason, we call (q
r
, p
r
)subsystem the reduced system with
reduced conﬁguration vector q
r
. Explicit transformation of the system into the nor
mal form (4.6) reduces control of the original underactuated nonlinear system with
n secondorder subsystems to control of the reduced system with (n − m) second
order subsystems. This is a tremendous reduction in complexity of control design for
(possibly) highorder underactuated systems.
Remark 4.2.4. If the potential energy V (q) is independent of the external variable q
x
,
i.e. ∂V (q
x
, q
s
)/∂q
x
= 0, then g
r
= 0 and the generalized momentum p
r
is a conserved
quantity. Therefore, the nonlinear system in (4.6) is not controllable or stabilizable
to any equilibrium points for initial conditions with p
r
(0) = 0. For example, neither
the Acrobot, nor the Pendubot are controllable/stabilizable in lack of gravity.
The fact that the reduced system with conﬁguration vector q
r
is a simple La
grangian system that satisﬁes the EulerLagrange equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= 0
without any input forces means that the system must be controlled via its potential
force (or energy) that is parameterized by q
s
. Therefore the shape vector q
s
plays the
role of the control input for the dynamics of the reduced system in Q
r
. In addition,
the reduced system is a doubleintegrator gradient system parameterized with q
s
as
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= −
∂V
r
(q
r
, q
s
)
∂q
r
62
which admits a reduced Hamiltonian parameterized by q
s
as the following
H
r
(q
r
, p
r
, q
s
) =
1
2
p
T
r
m
−1
r
(q
s
)p
r
+ V
r
(q
r
, q
s
) (4.10)
that satisﬁes
˙
H
r
=
∂H
r
(q
r
, p
r
, q
s
)
∂q
s
p
s
(4.11)
Based on (4.11) the reduced Hamiltonian H
r
is not necessarily a conserved quantity.
To see this, ﬁrst we need the following deﬁnition.
Deﬁnition 4.2.1. (vector sigmoidal function) σ(x) : R
n
→ R
n
is called a vector
sigmoidal function if it satisﬁes the following properties.
i) ∃L > 0 : ∀x ∈ R
n
, σ(x) ≤ L.
ii) σ(0) = 0 and xσ(x) > 0, ∀x = 0.
iii) y
T
(σ(x) −σ(x + y)) < 0, ∀y = 0 . Examples of vector sigmoidal functions are
σ(x) = (σ
1
(x
1
), . . . , σ
n
(x
n
))
T
where σ
i
are onedimensional sigmoidal functions
like arctan(z) and tanh(z).
Now, consider the following cases:
i) p
s
= 0, or p
s
satisfying ∇
qs
H
r
p
s
= 0 conserves the energy H
r
.
ii) p
s
= −σ(∇
qs
H
r
) decreases the energy of the reduced system.
iii) p
s
= σ(∇
qs
H
r
) increases the energy of the reduced system.
In other words, the energy is not necessarily conserved for the reduced system in the
parameterized Hamiltonian form
˙ q
r
=
∂H
r
(q
r
, p
r
, q
s
)
∂p
r
˙ p
r
= −
∂H
r
(q
r
, p
r
, q
s
)
∂q
r
Remark 4.2.5. The fact that the total energy of the reduced system can be increased or
decreased is useful for swingup control design and stabilization of an underactuated
system to an equilibrium manifold/point, respectively.
4.2.2 Underactuated Systems with Unactuated Shape Vari
ables
In this section, we address reduction of underactuated mechanical systems in (4.2)
that all of their shape variables are unactuated, have equal number of external and
shape variables, and their inputs are noninteracting, i.e. F
x
(q) = I
m
, F
s
(q) = 0.
63
Theorem 4.2.2. Consider an underactuated mechanical system with fullyactuated
external variables q
x
and unactuated shape variable q
s
and assume dim(Q
x
) = dim(Q
s
) =
m. The EulerLagrange equations of motion for this system is as the following
m
xx
(q
s
)¨ q
x
+ m
xs
(q
s
)¨ q
s
+ h
x
(q, ˙ q) = τ
m
sx
(q
s
)¨ q
x
+ m
ss
(q
s
)¨ q
s
+ h
s
(q, ˙ q) = 0
(4.12)
Let τ = α(q
s
)u + β(q, ˙ q) be the noncollocated partially linearizing change of control
for (4.12) over
U = ¦q
s
∈ Q
s
[ det(m
xs
(q
s
)) = 0¦
Assume all the elements of
ω = m
−1
sx
(q
s
)m
ss
(q
s
)dq
s
are exact oneforms over U and let ω = dγ(q
s
). Then, there exists a global change of
coordinates (i.e. diﬀeomorphism) obtained from the Lagrangian of the system
q
r
= q
x
+ γ(q
s
)
p
r
= m
sx
(q
s
) ˙ q
x
+ m
ss
(q
s
) ˙ q
s
=
∂L
∂ ˙ q
s
(4.13)
that transforms the dynamics of the underactuated system (4.12) into a cascade non
linear system in nontriangular quadratic form
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= g
r
(q
r
, q
s
) + Σ(q
s
, p
r
, p
s
)
˙ q
s
= p
s
˙ p
s
= u
(4.14)
where Σ = ∇
qs
K (K is the kinetic energy) is a quadratic form in (p
r
, p
s
)
Σ(q
s
, p
r
, p
s
) =
¸
p
r
p
s
T
Π(q
s
)
¸
p
r
p
s
with a cubic weight matrix Π(q
s
) and
m
r
(q
s
) := m
sx
(q
s
)
g
r
(q
r
, q
s
) := −[∇
qs
V
r
(q
x
, q
s
)]
qx=qr−γ(qs)
In addition, if V (q) = V (q
x
), then g
r
≡ 0 and the (q
r
, p
r
)subsystem is a Lagrangian
system with conﬁguration vector q
r
and reduced Lagrangian
L
r
(q
r
, ˙ q
r
, q
s
) =
1
2
˙ q
T
r
m
r
(q
s
) ˙ q
r
64
that satisﬁes the forced EulerLagrange equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= Σ(q
s
, p
r
, p
s
)
Proof. By deﬁnition of q
r
and p
r
, we have ˙ q
r
= m
−1
r
(q
s
)p
r
. Calculating ˙ p
r
, we get
˙ p
r
=
d
dt
∂L
∂ ˙ q
s
=
∂L
∂q
s
= −
∂V (q
x
, q
s
)
∂q
s
+
∂K
∂q
s
= g
r
(q
r
, q
s
) + Σ(q
s
, p
r
, p
s
)
Noting that ∂L
r
/∂q
r
= 0, one obtains
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= ˙ p
r
= Σ
and the forced EulerLagrange equation in the question follows. It remains to calculate
K(q
s
, p
r
, p
s
) and show it has a quadratic form with a cubic weight matrix. For doing
so, note that
¸
˙ q
x
˙ q
s
= W(q
s
)
¸
p
r
p
s
where W is an n n matrix given by
W(q
s
) =
¸
m
−1
sx
(q
s
) −m
−1
sx
(q
s
)m
ss
(q
s
)
0 I
m
Thus
K(q
s
, p
r
, p
s
) =
1
2
¸
p
r
p
s
T
N(q
s
)
¸
p
r
p
s
with
N(q
s
) = W
T
(q
s
)M(q
s
)W(q
s
)
Setting the ith layer of Π(q
s
) to dN(q
s
)/dq
i
s
(i.e. elementwise derivative of N(q
s
)
w.r.t. the ith component of q
s
) ﬁnishes the proof.
Example 4.2.1. Consider an underactuated system with conﬁguration vector q =
(x
1
, x
2
, θ
1
, θ
2
) that belongs to Q = R R S
1
S
1
. The system has the following
Lagrangian and (x
1
, x
2
) are actuated with external forces F = (F
1
, F
2
)
T
L =
1
2
˙ x
1
˙ x
2
˙
θ
1
˙
θ
2
¸
¸
¸
¸
T
1 0 cos θ
1
0
0 1 0 cos θ
2
cos θ
1
0 1 0
0 cos θ
2
0 1
¸
¸
¸
¸
˙ x
1
˙ x
2
˙
θ
1
˙
θ
2
¸
¸
¸
¸
−cos θ
1
cos θ
2
65
Apparently, q
s
= (θ
1
, θ
2
) is the vector of shape variables and q
x
= (x
1
, x
2
) is the vector
of actuated external variables of this system. We have
m
sx
(q
s
) =
¸
cos θ
1
0
0 cos θ
2
and det(m
sx
) = cos θ
1
cos θ
2
. Thus, over U = R
2
(−π/2, π/2)
2
, m
sx
(q
s
) is invertible
and the system can be partially linearized using a noncollocated partially linearizing
change of control F = α(q
s
)u +β(q
s
, ˙ q). Moreover, the vector of oneforms
ω = m
−1
sx
(q
s
)m
ss
dq
s
=
1
cos θ
1
0
0
1
cos θ
2
¸
¸
¸
¸
dθ
1
dθ
2
has exact elements ω
1
= (1/ cos θ
1
)dθ
1
and ω
2
= (1/ cos θ
2
)dθ
2
which means
γ(θ
1
, θ
2
) = (γ
0
(θ
1
), γ
0
(θ
2
))
T
with
γ
0
(θ) =
θ
0
ds
cos s
= log
1 + tan(θ/2)
1 −tan(θ/2)
, θ ∈ (−π/2, π/2)
Now, based on theorem 4.2.2 after applying the change of coordinates
q
r
=
¸
x
1
+ γ
0
(θ
1
)
x
2
+ γ
0
(θ
2
)
, p
r
=
¸
cos θ
1
0
0 cos θ
2
¸
˙ x
1
˙ x
2
+
¸
˙
θ
1
˙
θ
2
the dynamics of the system transforms into normal form (4.14).
Before presenting our next result, we need the following deﬁnition and lemma.
Deﬁnition 4.2.2. We say a square matrix function m(x) : R
n
→ R
n
2
has diﬀeren
tially symmetric rows, if the i
th
row of the matrix m
i∗
(x) satisﬁes
∂m(x)
∂x
i
=
∂m
i∗
(x)
∂x
(4.15)
for i = 1, . . . , n.
Remark 4.2.6. Apparently, in the scalar case with n = 1, the condition of the dif
ferential symmetry of rows trivially holds. This is already used in reduction of a
planar CartPole system in corollary 3.9.3. In addition, any linear combination of a
constant matrix and diagonal matrices in the form diag(f
1
(x
1
), . . . , f
n
(x
n
)) where f
i
’s
are scalar functions has diﬀerentially symmetric rows as well. The following matrix
is an example of a nondiagonal matrix with diﬀerentially symmetric rows
m(x
1
, x
2
) =
¸
cos x
1
sin x
1
sin x
2
0 cos x
1
cos x
2
66
where x
1
, x
2
∈ R.
Lemma 4.2.1. Suppose m(x) : R
n
→ R
n
2
is a matrix with diﬀerentially symmetric
rows. Let ˙ x(t) = v(t), w(t) ∈ R
n
be functions of time. Then, the following identity
holds
˙ m(x, ˙ x)w(t) = ∇
x
¦v(t)
T
m(x)w(t)¦
x=x(t)
(4.16)
where ˙ m is the elementwise timederivative of the matrix m(x(t)).
Proof. By deﬁnition of ˙ m, we have
˙ m(x, ˙ x)w(t) = col( ˙ m
1∗
w(t), . . . , ˙ m
n∗
w(t))
= col( ˙ x
T
∇
x
m
1∗
(x)w(t), . . . , ˙ x
T
∇
x
m
n∗
(x)w(t))
= col( ˙ x
T
∇
x
1
m
(
x)w(t), . . . , ˙ x
T
∇
xn
m
(
x)w(t))
= ∇
x
¦v(t)
T
m(x)w(t)¦
x=x(t)
Proposition 4.2.1. Assume all the conditions in theorem 4.2.2 hold. In addition,
the underactuated system (4.12) satisﬁes the following conditions
i) m
xx
(q
s
) is constant.
ii) m
sx
(q
s
) has diﬀerentially symmetric rows, i.e.
∂m
sx
(q
s
)
∂q
i
s
=
∂m
i∗
sx
∂q
s
, i = 1, . . . , m
where m
i∗
sx
(q
s
) is the ith row of m
sx
(q
s
).
iii) V (q) = V (q
s
).
Then, applying the change of coordinates
z
1
= q
r
, z
2
= m
−1
r
(q
s
)p
r
(where (q
r
, p
r
) are deﬁned in (4.13)) transforms the original system (4.12) into a
cascade system in feedforward form as the following
˙ z
1
= z
2
˙ z
2
= ψ(q
s
) + p
T
s
Π(q
s
)p
s
˙ q
s
= p
s
˙ p
s
= u
(4.17)
where Π(q
s
) is a cubic matrix and ψ : Q
s
→R
m
is deﬁned as
ψ(q
s
) = −m
−1
r
(q
s
)∇
qs
V (q
s
)
Moreover, if ψ(0) = 0 and ψ(q
s
) has an invertible Jacobian ∇
qs
ψ(q
s
) at q
s
= 0, the
origin for (4.17) (and (4.12)) can be globally asymptotically and locally exponentially
stabilized over U using a state feedback in explicit form as nested saturations.
67
Proof. We prove that under assumptions in the question ˙ p
r
only contains a quadratic
form in p
s
plus g
r
(q
s
) = −∇
qs
V (q
s
) and is independent of p
r
. For doing so, recall
that
Σ(q
s
, p
r
, p
s
) = ∇
qs
K
in equation (4.14). Since m
xx
is constant and
K =
1
2
p
T
x
m
xx
p
x
+ p
T
s
m
sx
(q
s
)p
x
+
1
2
p
T
s
m
ss
(q
s
)p
s
we get
∇
qs
K = ∇
qs
¦p
T
s
m
sx
(q
s
)p
x
¦ +∇
qs
¦
1
2
p
T
s
m
ss
(q
s
)p
s
¦
Clearly, only the ﬁrst term depends on p
x
(and thus p
r
). Based on lemma 4.2.1,
condition ii) implies m
sx
(q
s
) satisﬁes the following property
˙ m
sx
p
x
= ∇
qs
¦p
T
s
m
sx
(q
s
)p
x
¦ (4.18)
Notice that the time derivative in ˙ m
sx
is taken elementwise. Calculating ˙ z
2
, we
obtain
˙ z
2
=
d
dt
¦m
−1
sx
(q
s
)¦p
r
+ m
−1
sx
(q
s
) ˙ p
r
but m
−1
sx
(q
s
)m
sx
(q
s
) = I
m
and after diﬀerentiating both sides w.r.t. time, we get
d
dt
m
−1
sx
(q
s
) = −m
−1
sx
(q
s
) ˙ m
sx
m
−1
sx
(q
s
)
Therefore, ˙ z
2
can be expressed as
˙ z
2
= m
−1
sx
(q
s
)¦− ˙ m
sx
[p
x
+ m
−1
sx
(q
s
)m
ss
(q
s
)p
s
] + ˙ p
r
¦
or
˙ z
2
= m
−1
sx
(q
s
)[− ˙ m
sx
p
x
+∇
qs
¦p
T
s
m
sx
(q
s
)p
x
¦ − ˙ m
sx
m
−1
sx
(q
s
)m
ss
(q
s
)p
s
+ g
r
(q
s
) +∇
qs
¦
1
2
p
T
s
m
ss
(q
s
)p
s
¦]
= m
−1
sx
(q
s
)g
r
(q
s
) +p
T
s
Π(q
s
)p
s
where Π(q
s
) satisﬁes
−m
−1
sx
(q
s
) ˙ m
sx
m
−1
sx
(q
s
)m
ss
(q
s
)p
s
+m
−1
sx
(q
s
)∇
qs
¦
1
2
p
T
s
m
ss
(q
s
)p
s
¦ = p
T
s
Π(q
s
)p
s
and the feedforward normal form in (4.17) follows. The stabilization of feedforward
systems with higherorder perturbations using nested saturations is due Teel [102].
To obtain the stabilizing state feedback law explicitly, deﬁne
z
3
= q
s
, z
4
= p
s
where z
i
∈ R
m
and let
A = [∇
qs
ψ(q
s
)]
qs=0
68
be the invertible matrix in the question. The dynamics of (4.17) can be written as
˙ z
1
= z
2
˙ z
2
= Az
3
+ ϕ(z
3
, z
4
)
˙ z
3
= z
4
˙ z
4
= u
where ϕ(z
3
, z
4
) = z
T
4
Π(z
3
)z
4
+ ψ(z
3
) − Az
3
is quadratic in (z
3
, z
4
). Following [102],
deﬁne the change of coordinates and control
η
1
= A
−1
z
1
+ 2A
−1
z
2
+ 2z
3
+ z
4
η
2
= A
−1
z
2
+ z
3
+ z
4
w = z
3
+ z
4
+ u
we get
˙ η
1
= η
2
+ w + 2A
−1
ϕ(z
3
, z
4
)
˙ η
2
= w
Setting
w = −σ
1
(η
2
+σ
2
(η
1
))
or
u = −z
3
−z
4
+σ
1
(η
2
+σ
2
(η
1
))
globally asymptotically stabilizes the origin for this system where σ
i
’s are satura
tion functions that operate componentwise on an mdimensional vector and have
suﬃciently small thresholds and magnitudes (see [102] for further details).
A restriction of underactuated systems with unactuated shape variables which
are considered in this section is that they have equal number of external and shape
variables. This is in fact not necessary, if one uses a collocated partially lineariz
ing feedback that linearizes the dynamics of external variables, or a noncollocated
nonlinear combination of q
x
, q
s
. The following result provides a global normal form
for a class of underactuated systems with unactuated shape variables. This class
is important particularly due to its application in tracking control for ﬂexiblelink
robots.
Theorem 4.2.3. Consider an underactuated mechanical system with fullyactuated
external variables q
x
∈ R
n−m
and unactuated shape variable q
s
∈ R
m
which is aug
mented with an integrator at the input and satisﬁes the following equations of motion
m
xx
(q
s
)¨ q
x
+m
xs
(q
s
)¨ q
s
+ h
x
(q, ˙ q) = τ
m
sx
(q
s
)¨ q
x
+m
ss
(q
s
)¨ q
s
+ h
s
(q, ˙ q) = F
d
(q
s
, ˙ q
s
)
˙ τ = v
(4.19)
where F
d
(q
s
, ˙ q
s
) is an internal damping force acting on shape variables (e.g. F
d
=
−D(q
s
) ˙ q
s
with D(q
s
) = D
T
(q
s
) ≥ 0). Assume the potential energy of the system only
depends on shape variables, i.e. V (q) = V (q
s
). Suppose all the elements of the vector
ω = m
−1
xx
(q
s
)m
xs
(q
s
)dq
s
69
are exact oneforms and let ω = dγ(q
s
). Then, there exists a nonlinear output
y = h(q) = q
x
+ γ(q
s
)
which has global relative degree 3 w.r.t. v so that after applying the global change of
coordinates
z
1
= q
s
z
2
= p
s
ξ
1
= y = q
x
+ γ(q
s
)
ξ
2
= ˙ y = ˙ q
x
+ m
−1
xx
(q
s
)m
xs
(q
s
) ˙ q
s
ξ
3
= ¨ y = ¨ q
x
+ m
−1
xx
(q
s
)m
xs
(q
s
)¨ q
s
+
d
dt
¦m
−1
xx
(q
s
)m
xs
(q
s
)¦ ˙ q
s
(4.20)
the system in (4.19) transforms into ByrnesIsidori normal form with a tripleintegrator
˙ z
1
= z
2
˙ z
2
= f(z, ξ
1
, ξ
2
, ξ
3
)
˙
ξ
1
= ξ
2
˙
ξ
2
= ξ
3
˙
ξ
3
= u
(4.21)
where
f(z, ξ
1
, ξ
2
, ξ
3
) = M
−1
s
(z
1
)[g(z
1
) + F
d
(z
1
, z
2
) + Σ(z
1
, z
2
, ξ
2
) + m
sx
(z
1
)ξ
3
]
and
M
s
(q
s
) = m
ss
(q
s
) −m
sx
(q
s
)m
−1
xx
(q
s
)m
xs
(q
s
)
g(q
s
) = −∇V (q
s
)
Also, Σ is quadratic in (z
2
, ξ
2
) with a cubic weight matrix Π(z
1
) and
u = m
−1
xx
(q
s
)v + β(q, ˙ q, τ)
is an invertible change of control in v. In addition, the zerodynamics associated with
the output y = h(q) is a simple Lagrangian system with reduced shape Lagrangian
L
s
(q
s
, ˙ q
s
) =
1
2
˙ q
T
s
M
s
(q
s
) ˙ q
s
−V (q
s
)
that satisﬁes the forced EulerLagrange equation
d
dt
∂L
s
∂ ˙ q
s
−
∂L
s
∂q
s
= F
d
(q
s
, ˙ q
s
)
Moreover, if the following conditions hold
i) V (q
s
) is a positive deﬁnite proper function with V (0) = 0.
ii) p
T
s
F
d
(q
s
, p
s
) < 0 for all q
s
∈ R
m
, p
s
= 0.
70
Then, the Lagrangian zerodynamics given by
˙ q
s
= M
−1
s
(q
s
)π
s
˙ π
s
= −∇V (q
s
) + F
d
(q
s
, ˙ q
s
)
is globally minimumphase with a reduced Hamiltonian
H
s
(q
s
, π
s
) =
1
2
π
T
s
M
−1
s
(q
s
)π
s
+ V (q
s
)
which is a valid Lyapunov function for the zerodynamics, i.e. H
s
is smooth positive
deﬁnite and proper function satisfying
˙
H
s
≤ 0 (where π
s
is the generalized momentum
∂L
s
/∂ ˙ q
s
of the reduced system conjugate to q
s
).
Proof. By deﬁnition of ξ
2
, we have
˙
ξ
2
= ξ
3
= m
−1
xx
(q
s
)(τ −h
x
(q
s
, ˙ q)) +
d
dt
¦m
−1
xx
(q
s
)m
xs
(q
s
)¦ ˙ q
s
or simply
ξ
3
= m
−1
xx
(q
s
)τ +β
1
(q
s
, ˙ q)
which means ξ
3
is an invertible change of variable in τ. Diﬀerentiating the last
equation in time gives
˙
ξ
3
= u with a new control u as deﬁned in the question. To
prove the equation of ˙ z
2
in (4.21), let us substitute for ¨ q
x
from the deﬁnition of ξ
3
in
the second line of (4.19) to get
M
s
(q
s
)¨ q
s
+ m
sx
(q
s
)ξ
3
−m
sx
(q
s
)
d
dt
¦m
−1
xx
(q
s
)m
xs
(q
s
)¦ ˙ q
s
+ h
s
= F
d
or
¨ q
s
= M
−1
s
(q
s
)[−m
sx
(q
s
)ξ
3
+ m
sx
(q
s
)
d
dt
¦m
−1
xx
(q
s
)m
xs
(q
s
)¦ ˙ q
s
+ F
d
(q
s
, ˙ q
s
) −h
s
(q, ˙ q)]
that is in the form of second line of (4.21). To obtain the speciﬁc structure of f(z, ξ)
in (4.21), we need to calculate h
s
more explicitly as
h
s
(q
s
, ˙ q) = ˙ m
sx
˙ q
x
+ ˙ m
ss
˙ q
s
−∇
qs
K −g
s
(q
s
)
where g
s
(q
s
) = −∇V (q
s
) and ∇
qs
K is quadratic in (p
x
, p
s
) = ( ˙ q
x
, ˙ q
s
). But
¸
p
x
p
s
=
¸
I −m
−1
xx
(q
s
)m
xs
(q
s
)
0 I
¸
ξ
2
z
2
thus after substituting for (p
x
, p
s
) in ∇
qs
K, we obtain
∇
qs
K =
¸
ξ
2
z
2
Π
K
(q
s
)
¸
ξ
2
z
2
71
where Π
K
(q
s
) is a cubic matrix. Deﬁning
Σ :=
¸
ξ
2
z
2
Π
k
(q
s
)
¸
ξ
2
z
2
+ m
sx
(q
s
)
d
dt
¦m
−1
xx
(q
s
)m
xs
(q
s
)¦ ˙ q
s
− ˙ m
sx
˙ q
x
− ˙ m
ss
˙ q
s
or
Σ :=
¸
ξ
2
z
2
Π
k
(q
s
)
¸
ξ
2
z
2
− ˙ m
sx
ξ
2
+
˙
M
s
˙ q
s
gives the desired structure of f(z, ξ) in (4.21) (the last two terms are both quadratic
in (z
2
, ξ
2
)). To obtain the equations of the zerodynamics, let ξ
1
= ξ
2
= ξ
3
≡ 0 and
notice that
ξ
2
= ˙ q
x
+ m
−1
xx
(q
s
)m
ss
(q
s
) ˙ q
s
= 0
is a holonomic velocity constraint and under such a constraint the new Lagrangian
of the system can be expressed as
L
s
= L(q
s
, ˙ q
x
, ˙ q
s
) +λ
T
1
(q
x
+γ(q
s
)) + λ
T
2
( ˙ q
x
+ m
−1
xx
(q
s
)m
ss
(q
s
) ˙ q
s
)
where λ
1
, λ
2
∈ R
n−m
are the vectors of Lagrangian multipliers. Under the constraints
y = 0 and ˙ q
x
= −m
−1
xx
(q
s
)m
xs
(q
s
) ˙ q
s
, L
s
can be expressed as
L
s
=
1
2
˙ q
T
s
¸
−m
−1
xx
(q
s
)m
xs
(q
s
)
I
m
¸
m
xx
(q
s
) m
xs
(q
s
)
m
sx
(q
s
) m
ss
(q
s
)
¸
−m
−1
xx
(q
s
)m
xs
(q
s
)
I
m
˙ q
s
−V (q
s
)
that after simpliﬁcation takes the form
L
s
(q
s
, ˙ q
s
) =
1
2
˙ q
T
s
(m
ss
(q
s
) −m
sx
(q
s
)m
−1
xx
(q
s
)m
xs
(q
s
)) ˙ q
s
−V (q
s
)
or
L
s
(q
s
, ˙ q
s
) =
1
2
˙ q
T
s
M
s
(q
s
) ˙ q
s
−V (q
s
)
Therefore, the zerodynamics is a simple Lagrangian system with inertia matrix
M
s
(q
s
) and its Lagrangian L
s
satisﬁes the EulerLagrange equation with input damp
ing force F
d
(q
s
, ˙ q
s
). The equation of the zerodynamics in shape space can be equiv
alently written as
M
s
(q
s
)¨ q
s
+ C
s
(q
s
, ˙ q
s
) ˙ q
s
= g(q
s
) + F
d
(q
s
, ˙ q
s
)
where
˙
M
s
= C
s
(q
s
, ˙ q
s
) + C
T
(q
s
, ˙ q
s
). Apparently, H
s
by deﬁnition is smooth proper
positive deﬁnite function. Calculating
˙
H
s
, we get
˙
H
s
= ˙ q
T
s
M
s
(q
s
)¨ q
s
+
1
2
˙ q
T
s
˙
M
s
(q
s
) ˙ q
s
+ ˙ q
T
s
∇V (q
s
)
= −˙ q
T
s
C ˙ q
s
+
1
2
˙ q
T
s
(C + C
T
) ˙ q
s
+ ˙ q
s
g(q
s
) + ˙ q
T
s
∇V (q
s
) + ˙ q
T
s
F
d
(q
s
, ˙ q
s
)
= ˙ q
T
s
F
d
(q
s
, ˙ q
s
) +
1
2
˙ q
T
s
(C
T
−C) ˙ q
s
72
but C
T
−C is a skewsymmetric matrix and for all ˙ q
s
= p
s
= 0
p
T
s
(C
T
−C)p
s
= [p
T
s
(C
T
−C)p
s
]
T
= −p
T
s
(C
T
−C)p
s
= 0
Thus
˙
H
s
= p
T
s
F
d
(q
s
, p
s
) < 0
for all p
s
= 0. Based on LaSalle’s invariance principle [40], all the solutions of
the zerodynamics system asymptotically converge to the largest invariant set in
¦(q
s
, p
s
) [
˙
H
s
= 0¦ = ¦(q
s
, p
s
) [ p
s
= 0¦ which is equal to ¦0¦. Therefore, the ori
gin is globally asymptotically stable for the zerodynamics, i.e. the zerodynamics
is globally minimumphase, and H
s
is a valid Lyapunov function for the Lagrangian
zerodynamics.
Example 4.2.2. (Flexible linkrobots) Consider a ﬂexible onelink robot arm with
a single actuator at the joint θ as shown in Figure 41. Modeling the link as an
w
M J
x
y
θ
α
L L
X
Y
Figure 41: A ﬂexible onelink robot arm.
EulerBernoulli beam and using truncated modal analysis with m modes that have
amplitudes δ = (δ
1
, . . . , δ
m
)
T
[25], the dynamics of this ﬂexible link is an underactu
ated system as the following [25]
¸
m
θθ
(δ) m
θδ
m
δθ
m
δδ
¸
¨
θ
¨
δ
+
¸
h
θ
(δ,
˙
δ,
˙
θ)
h
δ
(δ,
˙
θ)
=
¸
τ
−D
˙
δ
with a scalar control τ. The external variable q
x
= θ is actuated and the m
dimensional shape variable q
s
= δ is unactuated but has natural springs and dampers
with constants k
j
, d
j
> 0 at each deformation mode such that
V (δ) =
1
2
δ
T
kδ, F
d
(
˙
δ) = −D
˙
δ
with k = diag(k
1
, . . . , k
m
) and D = diag(d
1
, . . . , d
m
). This ﬂexible onelink arm is
an example of a highorder underactuated system with kinetic symmetry and many
unactuated shape variables. In section 5.9, based on theorem 4.2.3, we prove that
there exists a nonlinear noncollocated minimumphase output in the form
y = θ + σ(c
T
δ)
73
that has global relative degree 3 w.r.t. the control v of the augmented system with an
integrator ˙ τ = v. Here, c ∈ R
m
is a constant and σ() is a scalar sigmoidal function
(see [71] for details on trajectory tracking control design).
4.3 Underactuated Systems with Input Coupling
In this section, we focus on reduction of a class of underactuated systems with in
put coupling. By input coupling, we mean there exists no permutation of rows of
the force (or input) matrix F(q) with full column rank m that transforms it into
col(F
1
(q), F
2
(q)) such that F
1
≡ 0 and F
2
is an m m invertible matrix. In the dy
namics of aerospace vehicles like aircraft and helicopter, the eﬀect of the body torque
appears in the translational dynamics of the vehicle. This motivated us to consider
eliminating this coupling eﬀect for the class of underactuated systems with kinetic
symmetry and input coupling via a global change of coordinates that decouples the
dynamics of the external variables and shape variables w.r.t. to the control applied
to the shape variables. The VTOL aircraft has a constant inertia matrix, while we
will see in chapter 5 that the inertia matrix of an aircraft or a helicopter depends on
the pitch and roll angles and is not constant.
On the other hand, in the nature, a ﬂying bird, a swimming ﬁsh, and a walking
human being are all examples of mechanical systems that their position is unactuated
and their locomotion is due to the changes in their physical shape (see [74, 9, 49,
104] for simpler examples). Moreover, the body of all three examples contains fully
actuated joints. In other words, a bird, a ﬁsh, and a human are examples of non
ﬂat underactuated systems with actuated shape variables and unactuated external
variables. Roughly speaking, the nonﬂatness property of these systems is due to the
fact that any changes in the shape variables of any of them aﬀects the physical shape
and thus their corresponding inertia matrix. Therefore, for the sake of generality
of our analysis, we ﬁrst consider the class of nonﬂat underactuated systems with
fullyactuated shape variables and input coupling.
Theorem 4.3.1. Consider an underactuated system with kinetic symmetry w.r.t. q
x
and fullyactuated shape variables. Assume the inertia matrix of this system is block
diagonal, i.e. M(q
s
) = diag(m
xx
(q
s
), m
ss
(q
s
)), and the force matrix is independent of
q
x
, i.e. F(q) = F(q
s
). The dynamics of the system can be expressed in the form
m
xx
(q
s
)¨ q
x
+ h
x
(q, ˙ q) = F
x
(q
s
)τ
m
ss
(q
s
)¨ q
s
+h
s
(q, ˙ q) = F
s
(q
s
)τ (4.22)
where τ ∈ R
m
and F
s
(q
s
) is an invertible mm matrix. Due to m
xs
(q
s
) = 0, system
(4.22) is globally partially feedback linearizable using an invertible change of control
τ = F
−1
s
(q
s
)[m
ss
(q
s
)u + h
s
(q, ˙ q)]
74
Suppose all the elements of
ω = m
−1
xx
(q
s
)F
x
(q
s
)F
−1
s
(q
s
)m
ss
(q
s
)dq
s
are exact oneforms and let ω = dγ(q
s
). Then, the following global change of coordi
nates (i.e. diﬀeomorphism)
q
r
= q
x
−γ(q
s
)
p
r
= m
xx
(q
s
)p
x
−F
x
(q
s
)F
−1
s
(q
s
)m
ss
(q
s
)p
s
(4.23)
with (p
x
, p
s
) = ( ˙ q
x
, ˙ q
s
) transforms (4.22) into the following cascade system in nontri
angular quadratic normal form
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= g
r
(q
r
, q
s
) + Σ(q
s
, p
r
, p
s
)
˙ q
s
= p
s
˙ p
s
= u
(4.24)
In addition, the (q
r
, p
r
)subsystem is a Lagrangian system with conﬁguration vector
q
r
and the reduced Lagrangian parameterized by q
s
L
r
(q
r
, ˙ q
r
, q
s
) =
1
2
˙ q
T
r
m
r
(q
s
) ˙ q
r
−V
r
(q
r
, q
s
)
that satisﬁes the following forced EulerLagrange equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= −F
x
(q
s
)F
−1
s
(q
s
)g
s
(q
r
+γ(q
s
), q
s
) + Σ(q
s
, p
r
, p
s
)
where
m
r
(q
s
) := m
xx
(q
s
)
V
r
(q
r
, q
s
) := V (q
r
+γ(q
s
), q
s
)
g
r
(q
r
, q
s
) := [g
x
(q
x
, q
s
) −F
x
(q
s
)F
−1
s
(q
s
)g
s
(q
x
, q
s
)]
qx=qr+γ(qs)
g
x
(q
x
, q
s
) := −∇
qx
V (q
x
, q
s
)
g
s
(q
x
, q
s
) := −∇
qs
V (q
x
, q
s
)
and
Σ(q
s
, p
r
, p
s
) =
¸
p
r
p
s
T
Π(q
s
)
¸
p
r
p
s
is a quadratic form in (p
r
, p
s
) with a cubic weight matrix Π(q
s
).
Proof. Consider the following generalized momentums
π
x
:=
∂L
∂ ˙ q
x
= m
xx
(q
s
) ˙ q
x
, π
s
:=
∂L
∂ ˙ q
s
= m
ss
(q
s
) ˙ q
s
By deﬁnition of q
r
, p
r
, we have ˙ q
r
= m
−1
r
(q
s
)p
r
and
p
r
= π
x
−F
x
(q
s
)F
−1
s
(q
s
)π
s
75
Based on EulerLagrange equation, π
x
, π
s
satisfy
˙ π
x
=
∂L
∂q
x
+ F
x
(q
s
)τ, ˙ π
s
=
∂L
∂q
s
+ F
s
(q
s
)τ
Thus, we have
˙ p
r
= ˙ π
x
−F
x
(q
s
)F
−1
s
(q
s
) ˙ π
s
−
d
dt
¦F
x
(q
s
)F
−1
s
(q
s
)¦π
s
=
∂L
∂q
x
−F
x
(q
s
)F
−1
s
(q
s
)
∂L
∂q
s
−p
T
s
π
F
(q
s
)p
s
where π
F
(q
s
) is a cubic matrix with (n−m) layers that are mm matrices satisfying
the relation
p
T
s
π
F
(q
s
)p
s
=
d
dt
¦F
x
(q
s
)F
−1
s
(q
s
)¦m
xx
(q
s
)p
s
Note that τ is eliminated from the equation of ˙ p
r
. By direct calculation, we have
∂L
∂q
x
= g
x
(q
x
, q
s
),
∂L
∂q
s
= g
s
(q
x
, q
s
) +
∂K
∂q
s
Thus, noting that ∂K/∂q
s
is a vector quadratic form in (p
x
, p
s
) with a cubic weight
matrix as a function of q
s
and setting
g
r
(q
r
, q
s
) =
g
x
(q
x
, q
s
) −F
x
(q
s
)F
−1
s
(q
s
)g
s
(q
x
, q
s
)
qx=qr+γ(qs)
it follows that
˙ p
r
= g
r
(q
r
, q
s
) + Σ(q
s
, p
r
, p
s
)
where
Σ(q
s
, p
r
, p
s
) = −F
x
(q
s
)F
−1
s
(q
s
)
∂K
∂q
s
−p
T
s
π
F
(q
s
)p
s
which proves equation (4.24). To obtain an explicit expression for Σ, let the cubic
matrix π
K
(q
s
) satisfy
F
x
(q
s
)F
−1
s
(q
s
)
∂K
∂q
s
=
¸
p
x
p
s
T
π
K
(q
s
)
¸
p
x
p
s
and note that each layer of π
K
(q
s
) is block diagonal according to the partition of
q = (q
x
, q
s
) (because M(q
s
) is block diagonal). By deﬁnition of p
r
, we get
¸
p
x
p
s
= W(q
s
)
¸
p
r
p
s
where W is an n n matrix given by
W(q
s
) =
¸
m
−1
xx
(q
s
) m
−1
xx
(q
s
)F
x
(q
s
)F
−1
s
(q
s
)m
ss
(q
s
)
0 I
m
76
This means
Π(q
s
) = −W
T
(q
s
)π
K
(q
s
)W(q
s
) −diag(0, π
F
(q
s
))
where the last term is a cubic matrix with block diagonal layers of 0 (as n layers of
0
n−m
) and π
F
(q
s
). The part that the reduced Lagrangian satisﬁes the forced Euler
Lagrange equation given in the question follows from the equation of ˙ p
r
and the
identity
∂L
r
∂q
r
=
¸
∂L
∂q
x
qx=qr+γ(qs)
∂q
x
∂q
r
= [g
x
(q
x
, q
s
)]
qx=qr+γ(qs)
Example 4.3.1. (A Flying Bird) Figure 42 shows the simpliﬁed model of a ﬂying
bird. The position of the center of the body of the bird and the orientation of the body
is denoted by (x, R
0
) ∈ R
3
SO(3). In addition, each wing has its own orientation
R
i
∈ SO(3), i = 1, 2 which is fullyactuated. The orientation of the tail is parame
terized by a pitch angle φ
3
and roll angle θ
3
which both are actuated. Therefore, this
bird is an underactuated system with 14 DOF and 8 control inputs. It is clear that if
φ
θ
x , R
R
R
2
0
1
3
3
Figure 42: A ﬂying bird.
R
0
, R
1
, R
2
, φ
3
, θ
3
are ﬁxed, then the physical shape of a bird does not change under
translation of the center of mass of its body. This means that the vector of external
variables is q
x
= x and the vector of shape variables is q
s
= (R
0
, R
1
, R
2
, φ
3
, θ
3
). Due
to the variable physical shape of the bird, the inertia matrix for this system is not
constant and depends on q
s
= (R
0
, R
1
, R
2
, φ
3
, θ
3
).
The following result provides classes of underactuated systems in theorem 4.3.1
that can be globally transformed into nonlinear systems in feedforward form. In
addition, suﬃcient conditions are given such that these feedforward forms are globally
asymptotically stabilizable using a state feedback in explicit form.
Proposition 4.3.1. Assume all the conditions in theorem 4.3.1 hold and in addition
i) m
r
= m
xx
is constant.
ii) V (q) = k
T
0
q
x
+ V (q
s
) where k
0
is a constant vector.
77
Then, the global change of coordinates in (4.23) transforms the dynamics of the un
deractuated system (4.22) into a cascade nonlinear system in feedforward form as the
following
˙ q
r
= m
−1
r
p
r
˙ p
r
= g
r
(q
s
) + p
T
s
Π(q
s
)p
s
˙ q
s
= p
s
˙ p
s
= u
(4.25)
where Π(q
s
) is a cubic matrix and
g
r
(q
s
) = −k
0
+ F
x
(q
s
)F
−1
s
(q
s
)∇
qs
V (q
s
)
Moreover, if the following conditions are satisﬁed
iii) dim(q
x
) = dim(q
s
).
iv) g
r
(0) = 0 and the Jacobian matrix ∇
qs
g
r
(q
s
) is invertible at q
s
= 0.
Then, the origin (q
r
, p
r
, q
s
, p
s
) = 0 for the nonlinear system in (4.25) can be globally
asymptotically stabilized using a state feedback in explicit form as nested saturations.
Proof. Following the proof of theorem 4.3.1, because m
xx
is constant ∂K/∂q
s
is a
quadratic form in p
s
and Σ takes the following form
Σ = Σ(q
s
, p
s
) = p
T
s
π(q
s
)p
s
which is independent of p
r
. In addition, we have
g
r
(q
r
, q
s
) = g
r
(q
s
) = −k
0
−F
x
(q
s
)F
−1
s
(q
s
)g
s
(q
s
)
and (4.25) holds. The result on stabilization of the origin follows from the method
of nested saturations for feedforward systems due to Teel [102]. (see the proof of
proposition 4.2.1 for further details on control design).
Theorem 4.3.2. Consider a ﬂat underactuated system with conﬁguration vector q =
(q
x
, q
s
) and a force matrix that is independent of q
x
. Assume the potential energy of
the system is independent of q
s
, i.e. V (q) = V (q
x
). The EulerLagrange equations of
motion for this system are as the following
m
xx
¨ q
x
−g
x
(q
x
) = F
r
(q
s
)τ
r
+ F
x
(q
s
)τ
m
ss
¨ q
s
= F
s
(q
s
)τ
(4.26)
where g
x
(q
x
) = −∇
qx
V (q
x
), τ
r
∈ R, F
r
(q
s
) : R
m
→R
n−m
is a unit vector that is onto
over a unit ball in R
n−m
, τ ∈ R
m
, m
xx
and m
ss
are constant, and F
s
(q
s
) is an mm
invertible matrix. Let
τ = F
s
(q
s
)
−1
m
ss
u
be the partially linearizing change of control. Assume all the elements of
ω = m
−1
xx
F
x
(q
s
)F
−1
s
(q
s
)m
ss
dq
s
78
are exact oneforms and let ω = dγ(q
s
). Then, the following global change of coordi
nates
q
r
= q
x
−γ(q
s
)
p
r
= m
xx
p
x
−F
x
(q
s
)F
−1
s
(q
s
)m
ss
p
s
(4.27)
with (p
x
, p
s
) = ( ˙ q
x
, ˙ q
s
) transforms the dynamics of (4.26) into the following form
˙ q
r
= m
−1
r
p
r
˙ p
r
= g
r
(q
r
+ γ(q
s
)) −p
T
s
π
F
(q
s
)p
s
+ F
r
(q
s
)τ
r
˙ q
s
= p
s
˙ p
s
= u
(4.28)
where π
F
(q
s
) is a cubic matrix satisfying
d
dt
¦F
x
(q
s
)F
−1
s
(q
s
)¦m
ss
p
s
= p
T
s
π
F
(q
s
)p
s
In addition, the (q
r
, p
r
)subsystem is a ﬂat Lagrangian system with conﬁguration vec
tor q
r
and reduced Lagrangian
L
r
(q
r
, ˙ q
r
, q
s
) =
1
2
˙ q
T
r
m
r
˙ q
r
−V (q
r
+ γ(q
s
))
satisfying the forced EulerLagrangian equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= −p
T
s
π
F
(q
s
)p
s
+ F
r
(q
s
)τ
r
where m
r
= m
xx
. Moreover, if the following conditions hold:
i) p
T
s
π
F
(q
s
)p
s
= (p
T
s
Qp
s
)F
r
(q
s
) with Q ∈ R
m
2
.
ii) V (q
x
) is a linear function.
Then, the reduced Lagrangian system is a fullyactuated ﬂat system that satisﬁes
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= F
r
(q
s
)˜ τ
r
where
˜ τ
r
= τ
r
−p
T
s
Qp
s
is the new scalar control.
Proof. The proof is by direct calculation.
Example 4.3.2. (VTOL aircraft) The VTOL aircraft discussed in section 3.3.6 sat
isﬁes all the conditions of proposition 4.3.2 and after applying a global change of
coordinates that decouples the position and orientation dynamics of the VTOL air
craft w.r.t. the input torque u
2
in equation (5.41), the reduced dynamics of the
position is a fullyactuated ﬂat mechanical system. The conﬁguration of the VTOL
79
aircraft can be globally asymptotically stabilized to the origin using a smooth static
state feedback (see [70], or section 5.10 for more details on control design procedure).
Later on, we will show a similar result applies to an autonomous helicopter.
Corollary 4.3.1. Assume all the conditions in theorem 4.3.2 hold. In addition,
suppose F
x
, F
s
are constant and τ
r
= 0. Then, the nonlinear system in (4.28) is in
strict feedback form.
Example 4.3.3. (InertiaWheel Pendulum) The InertiaWheel Pendulum satisﬁes all
the conditions of the preceding corollary and can be globally asymptotically stabilized
using a state feedback in explicit form (see section 5.3 for more details).
4.4 Underactuated Systems with Nonintegrable
Momentums
In this section, we consider the class of underactuated systems with kinetic symmetry
that their normalized momentums are notintegrable. By normalized momentums, we
mean
π
x
= ˙ q
x
+ m
−1
xx
(q
s
)m
xs
(q
s
) ˙ q
s
= m
−1
xx
(q
s
)
∂L
∂ ˙ q
x
(4.29)
π
s
= ˙ q
x
+ m
−1
sx
(q
s
)m
ss
(q
s
) ˙ q
s
= m
−1
sx
(q
s
)
∂L
∂ ˙ q
s
(4.30)
where π
x
, π
s
are the normalized momentums conjugate to q
x
, q
s
, respectively. The
normalized momentum π
x
(or π
s
) is called nonintegrable, if h = h(q
x
, q
s
) :
˙
h =
π
x
(or π
s
). It turns out that in this case, underactuated systems can be reduced to
kinematic systems driven by the shape velocity vector p
s
as the control input. First,
we consider underactuated systems with actuated shape variables and nonintegrable
normalized momentums.
Theorem 4.4.1. Consider the underactuated system in (4.3) and suppose all the
assumptions in theorem (4.2.1) hold except for the exactness property of the elements
of ω. In other words, assume given
µ(q
s
) = m
−1
xx
(q
s
)m
xs
(q
s
)
µ(q
s
)p
s
is not integrable. Then, after applying the global change of coordinates
q
r
= q
x
p
r
= m
xx
(q
s
)p
x
+ m
xs
(q
s
)p
s
with (p
x
, p
s
) = ( ˙ q
x
, ˙ q
s
), the dynamics of the system transforms into the following
80
nontriangular normal form
˙ q
r
= m
−1
r
(q
s
)p
r
−µ(q
s
)p
s
˙ p
r
= g
r
(q
r
, q
s
)
˙ q
s
= p
s
˙ p
s
= u
(4.31)
that is aﬃne in the shape velocity p
s
as the control input of (q
r
, p
r
, q
s
)subsystem.
Proof. The equation of ˙ p
r
follows from the proof of theorem (4.2.1) and the ﬁrst
equation in (4.31) is a direct result of the deﬁnition of p
r
.
Remark 4.4.1. Deﬁning x = (q
r
, p
r
, q
s
) and v = p
s
, the dynamics of the (q
r
, p
r
, q
s
)
subsystem can be written as a nonlinear system aﬃne in control
˙ x = f(x) +g(x)v
where g(x) = col(−µ(q
s
), I
m
). This system has m (i.e. number of control inputs) ﬁrst
order diﬀerential equations less than the original system and both the control design
and controllability analysis of this system can be carried out in a lowerorder space
R
3m
instead of the original space R
4m
.
Remark 4.4.2. Recall that in the integrable case for underactuated systems with
actuated shape variables, it is possible to use backstepping procedure due to strict
feedback form of the transformed system. Here, due to nontriangular structure of the
normal form in (4.31) the backstepping procedure is not applicable. Control design
for (4.31) will be treated later in chapter 7.
Example 4.4.1. (threelink planar robot) Consider a planar threelink robot arm
with revolute joints (q
1
, q
2
, q
3
) and two actuators at q
2
, q
3
as shown in Figure 43.
The dynamics of this robot is given in section A.2 (Appendix A). It can be shown
q
q
q
1
2
3
x
y
Figure 43: A triplelink robot arm with two actuators.
that the inertia matrix of this triplelink robot has the following structure
M(q) = M(q
2
, q
3
) =
m
11
(q
2
, q
3
) m
12
(q
2
, q
3
) m
13
(q
2
, q
3
)
m
21
(q
2
, q
3
) m
22
(q
3
) m
23
(q
3
)
m
31
(q
2
, q
3
) m
32
(q
3
) m
33
¸
¸
81
Thus, q
x
= q
1
is the external variable and q
s
= (q
2
, q
3
) is the vector of shape variables.
Moreover, both shape variables are actuated. However, by direct calculation, it can
be shown that the normalized momentum
π
1
= ˙ q
1
+
m
12
(q
2
, q
3
)
m
11
(q
2
, q
3
)
˙ q
2
+
m
13
(q
2
, q
3
)
m
11
(q
2
, q
3
)
˙ q
3
is nonintegrable (see lemma A.2.1). Therefore, based on theorem 4.4.1, a threelink
robot can be transformed into a nontriangular kinematic normal form with p
s
=
( ˙ q
2
, ˙ q
3
) as the input.
Now, we present the case of underactuated systems with actuated external vari
ables and nonintegrable normalized momentums.
Theorem 4.4.2. Consider the underactuated mechanical system in (4.12) and sup
pose all the assumptions in theorem (4.2.2) hold except for the exactness property of
the elements of ω. In other words, assume given
µ(q
s
) = m
−1
sx
(q
s
)m
ss
(q
s
)
µ(q
s
)p
s
is nonintegrable. Then, after applying the change of coordinates
q
r
= q
x
p
r
= m
sx
(q
s
)p
x
+ m
ss
(q
s
)p
s
with (p
x
, p
s
) = ( ˙ q
x
, ˙ q
s
), the dynamics of the system transforms into the following
nontriangular linearquadratic normal form
˙ q
r
= m
−1
r
(q
s
)p
r
−µ(q
s
)p
s
˙ p
r
= g
r
(q
r
, q
s
) + Σ(q
s
, p
r
, p
s
)
˙ q
s
= p
s
˙ p
s
= u
(4.32)
over
U = ¦q
s
∈ Q
s
[ det(m
sx
(q
s
) = 0¦
with p
s
as the control input for the (q
r
, p
r
, q
s
)subsystem where
m
r
(q
s
) := m
sx
(q
s
)
g
r
(q
r
, q
s
) := ∇
qs
V (q
r
, q
s
)
Σ(q
s
, p
r
, p
s
) := ∇
qs
K
Proof. The proof for the equation of ˙ p
r
in (4.32) is the same as in theorem 4.2.2. By
deﬁnition of p
r
, we have
˙ q
x
= m
−1
sx
(q
s
)p
r
−m
−1
sx
(q
s
)m
ss
(q
s
)p
s
= m
−1
r
(q
s
)p
r
−µ(q
s
)p
s
and the result follows. Based on the structure of normal form (4.32), it is clear that
82
the (q
r
, p
r
, q
s
)subsystem is kinematic (i.e. has a control input p
s
that is the shape
velocity vector).
Remark 4.4.3. The normal form in (4.32) is very similar to a normal form previously
found for Lagrangian systems with nonholonomic velocity constraints and classical
symmetry due to Bloch et al. [9]. However, the formulation of theorem 4.4.2 is
fundamentally diﬀerent from the setup studied in [9].
Proposition 4.4.1. Assume all the assumptions in theorem 4.4.2 hold. In addition,
the underactuated system (4.12) satisﬁes the following conditions
i) m
xx
(q
s
) is constant.
ii) m
sx
(q
s
) has diﬀerentially symmetric rows, i.e.
∂m
sx
(q
s
)
∂q
i
s
=
∂m
i∗
sx
∂q
s
, i = 1, . . . , m
where m
i∗
sx
(q
s
) is the ith row of m
sx
(q
s
).
iii) V (q) = V (q
s
).
iv) µ(q
s
) is analytic at q
s
= 0 (elementwise).
Then, applying the change of coordinates
z
1
= q
r
+ µ(0)q
s
, z
2
= m
−1
r
(q
s
)p
r
(where (q
r
, p
r
) are deﬁned in (4.13)) transforms the underactuated system (4.12) into
a cascade system in feedforward form as the following
˙ z
1
= z
2
+ ϕ
1
(q
s
, p
s
)
˙ z
2
= ψ(q
s
) + p
T
s
Π(q
s
)p
s
˙ q
s
= p
s
˙ p
s
= u
(4.33)
where ϕ
1
(q
s
, p
s
) = (µ(0) − µ(q
s
))p
s
is at least quadratic in (q
s
, p
s
), Π(q
s
) is a cubic
matrix, and ψ : Q
s
→R
m
is deﬁned as
ψ(q
s
) = −m
−1
r
(q
s
)∇
qs
V (q
s
)
Moreover, if ψ(0) = 0 and ψ(q
s
) has an invertible Jacobian ∇
qs
ψ(q
s
) at q
s
= 0, the
origin for (4.33) can be globally asymptotically and locally exponentially stabilized over
U using a state feedback in the form of nested saturations.
Proof. Following the proofs of proposition 4.2.1 and theorem 4.4.2, after applying
the change of coordinates
¯ z
1
= q
r
, z
2
= m
−1
r
(q
s
)p
r
83
we obtain
˙
¯ z
1
= z
2
−µ(q
s
)p
s
˙ z
2
= ψ(q
s
) +p
T
s
π(q
s
)p
s
˙ q
s
= p
s
˙ p
s
= u
Noting that µ(q
s
) is elementwise analytic at q
s
= 0 and
µ(q
s
)p
s
= µ(0)p
s
−ϕ
1
(q
s
, p
s
)
where
ϕ
1
(q
s
, p
s
) = O((q
s
, p
s
)
2
)
by setting z
1
= ¯ z
1
+ µ(0)q
s
the result follows. The rest of the proof is rather similar
to the proof of proposition 4.2.1.
Example 4.4.2. (3D CartPole System) Consider an inverted pendulum mounted
on a moving platform via an unactuated 2 DOF joint (θ, φ) with two external forces
as shown in Figure 44. The angles θ, φ denote rotation around x
2
axis and x
1
axis,
respectively. Let (x
1
, x
2
) denote the position of the center of mass of the platform M.
F
F
2
1
x
3
x
1
2
x
M
m
a
φ
θ
l
φ
Figure 44: An inverted pendulum on a moving platform with 4 DOF and 2 controls.
Then, the kinetic energy of the 3D CartPole system (given in section A.7) is in the
form
K =
1
2
˙ x
1
˙ x
2
˙
θ
˙
φ
¸
¸
¸
¸
T
M + m 0 ml cos θ 0
0 M + m −ml sin θ sin φ ml cos θ cos φ
ml cos θ −ml sin θ sin φ ml
2
0
0 ml cos θ cos φ 0 ml
2
cos
2
θ
¸
¸
¸
¸
˙ x
1
˙ x
2
˙
θ
˙
φ
¸
¸
¸
¸
(4.34)
Clearly, the inertia matrix for this system only depends on (θ, φ). Therefore, (θ, φ)
are the shape variables and (x
1
, x
2
) are the external variables of the 3D CartPole
system. The system is an underactuated system with four DOF (x
1
, x
2
, θ, φ) and two
84
actuated external variables (x
1
, x
2
). For the matrix
m
sx
(q
s
) =
¸
ml cos θ −ml sin θ sin φ
0 ml cos θ cos φ
we have
det(m
sx
(q
s
)) = m
2
l
2
cos
2
θ cos φ
Thus, m
sx
is nonsingular over U = R
2
(−π/2, π/2)
2
. The vector of oneforms is
ω = µ(q
s
)p
s
where
µ(q
s
) = m
−1
sx
(q
s
)m
ss
(q
s
) = l
1
cos θ
sin θ sin φ
cos φ
0
cos θ
cos φ
¸
¸
¸
Since the following oneforms are not exact
ω
1
=
1
cos θ
dθ +
sin θ sin φ
cos φ
dφ
ω
2
=
cos θ
cos φ
dφ
Theorem 4.2.2 is not applicable and the normalized momentum π
s
of the 3D CartPole
system is nonintegrable. On the other hand, the dynamics of the 3D CartPole system
satisﬁes all four conditions of proposition 4.4.1. Therefore, it can be transformed into
a feedforward system over U.
Proposition 4.4.2. The 3D CartPole system (in example 4.4.2) can be transformed
into a nonlinear system in feedforward form using the change of coordinates (in 4.4.1)
z
1
= q
x
+ lq
s
, z
2
= m
−1
sx
(q
s
)p
r
In addition, the origin for the system can be globally asymptotically and locally expo
nentially stabilized using nested saturations.
Proof. Transformation into a feedforward system is already proven in example 4.4.2.
To prove the stabilization using nested saturations, we need to prove
ψ(q
s
) = −m
−1
sx
(q
s
)∇
qs
V (q
s
) =
g
cos φ
¸
sin θ
sin φ
has an invertible Jacobian at q
s
= 0. Calculating this Jacobian, we have
∂ψ
∂(θ, φ)
= g
cos θ
cos φ
sin θ sin φ
cos
2
φ
0
1
cos
2
φ
¸
¸
¸
(θ,φ)=0
= gI
2×2
= 0
85
which is clearly nonsingular and the stabilization result follows.
4.5 Momentum Decomposition for Underactuated
Systems
In this section, we consider reduction of underactuated systems with nonintegrable
normalized momentums. We introduce a method to decompose a nonintegrable
normalized momentum as an integrable momentum term called locked momentum
and a nonintegrable term called error momentum. We call this procedure Momentum
Decomposition. This decomposition is unique for a ﬁx choice of a locked conﬁguration
(to be deﬁned later). This procedure is described in the following.
Consider an underactuated mechanical system with kinetic symmetry and con
ﬁguration vector (q
x
, q
s
) where q
x
and q
s
denote the vector of external and shape
variables, respectively. The normalized momentums for this system are given by
π
x
:= m
−1
xx
(q
s
)
∂L
∂ ˙ q
x
= ˙ q
x
+ m
−1
xx
(q
s
)m
xs
(q
s
) ˙ q
s
(4.35)
π
s
:= m
−1
sx
(q
s
)
∂L
∂ ˙ q
s
= ˙ q
x
+ m
−1
sx
(q
s
)m
ss
(q
s
) ˙ q
s
(4.36)
which both can be expressed as the following
π = ˙ q
x
+ µ(q
s
) ˙ q
s
(4.37)
Assume π is nonintegrable, i.e. h :
˙
h = π. Our goal is to ﬁnd an integrable
approximation of π which allows us to apply our previous reduction procedures for
underactuated systems with integrable normalized momentums to the present case.
We refer to µ(q
s
) as the shape inertia matrix. The following theorem demonstrates
how every nonintegrable normalized momentum can be decomposed as an explicitly
integrable momentum plus a nonintegrable momentum term.
Theorem 4.5.1. (Momentum Decomposition) Let π = ˙ q
x
+µ(q
s
) ˙ q
s
be a nonintegrable
normalized momentum of a mechanical control system with kinetic symmetry w.r.t.
q
x
(η(q
s
) is a d m matrix). Denote the shape vector by q
s
= (q
1
s
, . . . , q
m
s
)
T
and
the normalized momentum by π = (π
1
, . . . , π
d
)
T
. Fix ¯ q
s
and deﬁne the locked shape
inertia matrix µ
l
(q
s
) by its elements
µ
l
ij
(q
j
s
) = [µ
ij
(q
s
)]
q
k
s
=¯ q
k
s
;k=1,...,m;k=j
(4.38)
and the locked momentum as
π
l
= ˙ q
x
+ µ
l
(q
s
) ˙ q
s
(4.39)
Then, the locked momentum is integrable in an explicit form
q
r
= q
x
+ γ(q
s
) (4.40)
86
i.e. ˙ q
r
= π
l
, where γ(q
s
) = (γ
1
(q
s
), . . . , γ
d
(q
s
))
T
and
γ
i
(q
s
) = Σ
m
j=1
q
j
s
0
µ
l
ij
(θ
j
)dθ
j
, i = 1, . . . , d (4.41)
In addition, denoting
µ
e
(q
s
) := µ(q
s
) −µ
l
(q
s
) (4.42)
the normalized error momentum can be deﬁned as
π
e
:= µ
e
(q
s
) ˙ q
s
(4.43)
and the nonintegrable momentum π can be uniquely decomposed as
π = π
l
+ π
e
(4.44)
where π
l
is integrable and π
e
is nonintegrable. Moreover, π
e
is independent of (q
x
, ˙ q
x
)
and vanishes at q
s
= ¯ q
s
.
Proof. The ith element of π can be expressed as
π
i
= ˙ q
i
x
+ Σ
m
j=1
µ
ij
(q
s
) ˙ q
j
s
= ˙ q
i
x
+ Σ
m
j=1
[µ
l
ij
(q
j
s
) + µ
e
ij
(q
s
)] ˙ q
j
s
= ˙ q
i
x
+ Σ
m
j=1
µ
l
ij
(q
j
s
) ˙ q
j
s
+ Σ
m
j=1
µ
e
ij
(q
s
) ˙ q
j
s
or
π = π
l
+ π
e
But the ith element of π
l
is
˙ q
i
x
+ Σ
m
j=1
µ
l
ij
(q
j
s
) ˙ q
j
s
= ˙ q
i
x
+ ˙ γ
i
Thus, setting q
r
= q
x
+ γ(q
s
), ˙ q
r
= π
l
and π
l
is explicitly integrable. Since π
e
=
η
e
(q
s
) ˙ q
s
, it is independent of (q
x
, ˙ q
x
). Also, by deﬁnition of µ
l
(q
s
), µ
l
(¯ q
s
) = µ(¯ q
s
) and
therefore both µ
e
(q
s
) and π
e
vanish at q
s
= ¯ q
s
.
In the following, we apply the momentum decomposition method introduced in
theorem 4.5.1 to underactuated systems with kinetic symmetry for two cases: i)
underactuated systems with actuated shape variables, and ii) underactuated systems
with unactuated shape variables.
Theorem 4.5.2. Consider the following underactuated system with kinetic symmetry
and actuated shape variables
m
xx
(q
s
)¨ q
x
+ m
xs
(q
s
)¨ q
s
+ h
x
(q, ˙ q) = 0
m
sx
(q
s
)¨ q
x
+ m
ss
(q
s
)¨ q
s
+ h
s
(q, ˙ q) = τ
(4.45)
Let τ = α(q
s
)u +β(q, ˙ q) be the global collocated partially linearizing change of control
for (4.45). Assume the following normalized momentum with µ
x
(q
s
) = m
−1
xx
(q
s
)m
xs
(q
s
)
87
is nonintegrable
π
x
= ˙ q
x
+µ
x
(q
s
) ˙ q
s
Let
π
x
= π
l
x
+ π
e
x
be the momentum decomposition of π
x
as an integrable locked momentum π
l
x
= ˙ q
x
+
µ
l
x
(q
s
) ˙ q
s
associated with the locked shape q
s
= ¯ q
s
and a nonintegrable normalized
error momentum π
e
x
. Then, there exists a function γ(q
s
) in explicit form satisfying
∇
qs
γ(q
s
) = µ
l
x
(q
s
) such that the global change of coordinates
q
r
= q
x
+ γ(q
s
)
p
r
= m
xx
(q
s
)( ˙ q
x
+ µ
l
x
(q
s
) ˙ q
s
)
(4.46)
transforms the dynamics of the underactuated system in (4.45) into a cascade non
linear system in strict feedback form
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= g
r
(q
r
, q
s
) + δ
˙ q
s
= p
s
˙ p
s
= u
(4.47)
with a perturbation δ = −˙ p
e
where p
e
= m
r
(q
s
)π
e
x
is the error momentum and
m
r
(q
s
) := m
xx
(q
s
)
V
r
(q
r
, q
s
) := V (q
x
, q
s
)[
qx=qr−γ(qs)
g
r
(q
r
, q
x
) := −∇
qr
V
r
(q
r
, q
s
)
The perturbation δ is in the form
δ = m
r
(q
s
)µ
e
x
(q
s
)u + p
T
s
Π
δ
(q
s
)p
s
which is aﬃne in u, quadratic in p
s
with cubic weight matrix Π
δ
(q
s
), and vanishes
uniformly in u at (q
s
, p
s
) = (¯ q
s
, 0). In addition, the reduced system is a Lagrangian
system with reduced parameterized Lagrangian
L
r
(q
r
, ˙ q
r
, q
s
) =
1
2
˙ q
T
r
m
r
(q
s
) ˙ q
r
−V
r
(q
r
, q
s
)
that satisﬁes the forced EulerLagrange equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= δ (4.48)
Proof. By deﬁnition of q
r
, p
r
, and m
r
(q
s
), ˙ q
r
= m
−1
r
(q
s
)p
r
. Due to kinetic symmetry,
we have
∂L
∂q
x
= −∇
qx
V (q
x
, q
s
)
88
which means in new coordinates
∂L
r
∂q
r
= g
r
(q
r
, q
s
)
On the other hand, by deﬁnition of π
x
and based on momentum decomposition the
orem, we have
∂L
∂ ˙ q
x
= m
r
(q
s
)π
x
= m
r
(q
s
)(π
l
x
+ π
e
x
) = p
r
+ p
e
Hence
˙ p
r
+ ˙ p
e
=
d
dt
∂L
∂ ˙ q
x
=
∂L
∂q
x
= g
r
(q
r
, q
s
)
or
˙ p
r
= g
r
(q
r
, q
s
) − ˙ p
e
The forced EulerLagrange equation with input δ = −˙ p
e
follows from the last equa
tion.
Remark 4.5.1. Theorem 4.5.2 shows that control design for the original underactuated
system with a nonintegrable normalized momentum reduces to control design for
system (4.47) which is in strict feedback form plus a nonlinear perturbation δ. One
possible way to stabilize the nonlinear system in (4.47) is to design a controller for
the unperturbed system which reduces to control of the (q
r
, p
r
)subsystem combined
with a backstepping stage. Then, perform a Lyapunovbased robustness analysis
w.r.t. the nonlinear perturbation δ which vanishes at a desired equilibrium point
with shape conﬁguration and velocity (q
s
, p
s
) = (¯ q
s
, 0).
Remark 4.5.2. If the normalized momentum π
x
is integrable, then π
e
≡ 0 which
means δ, p
e
≡ 0 and thus theorem 4.5.2 reduces to theorem 4.2.1 stated earlier for
underactuated systems with actuated shape variables and integrable normalized mo
mentums.
Remark 4.5.3. The main diﬀerence between normal forms (4.47) and (4.31) for under
actuated systems with nonintegrable normalized momentums is that in the former
one, the perturbation term δ appears in the equation of ˙ p
r
and a reduced Lagrangian
system with conﬁguration vector q
r
and input force δ is identiﬁable (as in theorem
4.5.2), but in the latter case, the perturbation µ(q
s
)p
s
appears in the equation of ˙ q
r
and no Lagrangian reduced system can be identiﬁed.
Example 4.5.1. (threelink planar robot with actuated shape variables) Consider
the threelink planar robot arm with two actuated shape variables (q
2
, q
3
) as shown
in Figure 45 (this system was also discussed earlier in example 4.4.1). The dynamics
of the actuated shape variables of this system can be globally linearized using a
collocated partiallylinearizing change of control as the following
¨ q
2
= u
2
, ¨ q
3
= u
3
The normalized momentum of this system conjugate to q
1
is nonintegrable (see lemma
89
A.2.1). Note that for this threelink robot
µ(q
s
) = m
−1
11
(q
2
, q
3
)
m
12
(q
2
, q
3
) m
13
(q
2
, q
3
)
After locking q
2
, q
3
at (¯ q
2
, ¯ q
3
) = (0, 0), the locked approximation of π
1
can be calcu
q
q
q
1
2
3
x
y
Figure 45: A triplelink robot arm with actuated shape variables.
lated as
π
l
1
= ˙ q
1
+
m
12
(q
2
, 0)
m
11
(q
2
, 0)
˙ q
2
+
m
13
(0, q
3
)
m
11
(0, q
3
)
˙ q
3
which gives explicit forms for γ
1
and γ
2
as the following
γ
1
(q
2
) =
q
2
0
m
12
(s, 0)
m
11
(s, 0)
ds, γ
2
(q
3
) =
q
3
0
m
12
(0, s)
m
11
(0, s)
ds
After applying the change of coordinates
q
r
= q
1
+ γ(q
2
) + γ(q
3
)
p
r
= m
11
(q
2
, q
3
)[p
1
+
m
12
(q
2
, 0)
m
11
(q
2
, 0)
p
2
+
m
13
(0, q
3
)
m
11
(0, q
3
)
p
3
]
the system transforms into normal form (4.47). Calculating the error momentum
p
e
= m
r
(q
s
)π
e
1
, we get
p
e
= (m
12
(q
2
, q
3
) −m
12
(q
2
, 0)) ˙ q
2
+ (m
13
(q
2
, q
3
) −m
13
(0, q
3
)) ˙ q
3
and thus δ = −˙ p
e
is given by
δ = −(m
12
(q
2
, q
3
) −m
12
(q
2
, 0))u
2
−(m
13
(q
2
, q
3
) −m
13
(0, q
3
))u
3
+ λ(q
2
, ˙ q
2
, q
3
, ˙ q
3
)
where δ can be made exponentially vanishing by applying a linear control u
3
=
K
3
(q
3
, p
3
) = −c
1
q
3
−c
2
p
3
with c
1
, c
2
> 0. Keeping this in mind, the dynamics of the
reduced system for the threelink robot can be rewritten as
˙ q
r
= m
−1
r
(q
2
, 0)p
r
+ ϕ
1
˙ p
r
= g
r
(q
r
, q
2
, 0) + ϕ
2
90
where the nonlinear perturbations ϕ
1
, ϕ
2
are given by
ϕ
1
= (m
−1
r
(q
2
, q
3
) −m
r
(q
2
, 0))p
r
, ϕ
2
= (g
r
(q
r
, q
2
, q
3
) −g
r
(q
r
, q
2
, 0)) + δ
Both ϕ
1
and ϕ
2
vanish exponentially as (q
3
, p
3
) converges exponentially fast to zero.
In other words, control of the threelink planar robot which is a sixth order nonlinear
system reduces to stabilization of the secondorder unperturbed reduced system
˙ q
r
= m
−1
r
(q
2
, 0)p
r
˙ p
r
= g
r
(q
r
, q
2
, 0)
But this normal form is exactly the same as the reduced system of an Acrobot with
conﬁguration vector (q
1
, q
2
) and the following inertia matrix and potential energy
M
Acro
(q
2
) =
¸
m
11
(q
2
, 0) m
12
(q
2
, 0)
m
21
(q
2
, 0) m
22
(0)
, V
Acro
(q
r
, q
2
) = [V (q
1
, q
2
, 0)]
q
1
=qr−γ
1
(q
2
)
The shape variable of this Acrobot is q
2
which is actuated. Since the Acrobot can
be globally asymptotically stabilized to its upright equilibrium point using a smooth
state feedback u
2
= K
2
(q
r
, p
r
, q
2
, p
2
) [68] (under minor conditions), after a rather
elementary perturbation analysis it follows that the origin for a threelink planar
robot can be globally asymptotically stabilized using (static) state feedback laws
u
2
= K
2
(q
r
, p
r
, q
2
, p
2
) and u
3
= K
3
(q
3
, p
3
).
Remark 4.5.4. The result of example 4.5.1 is in agreement with our intuition that if
one sets the joint angle q
3
identically to zero, a threelink planar robot graphically
looks the same as an Acrobot with an actuated shape variable q
2
. Keep in mind that
this choice is not unique and one could have started by exponential stabilization of
(q
2
, p
2
)subsystem to zero and then stabilize an Acrobot with an external variable q
1
and shape variable q
3
.
Next, we consider the case of underactuated systems with partiallyactuated shape
variables and nonintegrable momentums.
Theorem 4.5.3. Consider the following underactuated system with kinetic symmetry
and partiallyactuated shape variables q
s
= col(q
s
1
, q
s
2
) ∈ Q
s
1
Q
s
2
m
xx
(q
s
)¨ q
x
+ m
xs
1
(q
s
)¨ q
s
1
+ m
xs
2
(q
s
)¨ q
s
2
+ h
x
(q, ˙ q) = τ
1
m
s
1
x
(q
s
)¨ q
x
+ m
s
1
s
1
(q
s
)¨ q
s
1
+ m
s
1
s
2
(q
s
)¨ q
s
2
+ h
s
1
(q, ˙ q) = 0
m
s
2
x
(q
s
)¨ q
x
+ m
s
2
s
1
(q
s
)¨ q
s
1
+ m
s
2
s
2
(q
s
)¨ q
s
2
+ h
s
2
(q, ˙ q) = τ
2
(4.49)
Assume dim(Q
x
) = dim(Q
s
1
) and dim(Q
s
2
) ≥ 0. Let τ = α(q
s
)u + β(q, ˙ q) be the
noncollocated partially linearizing change of control for (4.49) over
U = ¦q
s
∈ Q
s
[ det(m
s
1
x
(q
s
)) = 0¦
which linearizes the dynamics of the shape variables as ¨ q
s
= u. Assume the following
91
normalized momentum is nonintegrable
π
s
1
= ˙ q
x
+µ
s
1
(q
s
) ˙ q
s
with
µ
s
1
(q
s
) = m
−1
s
1
x
(q
s
)
m
s
1
s
1
(q
s
) m
s
1
s
2
(q
s
)
Let
π
s
1
= π
l
s
1
+ π
e
s
1
be the momentum decomposition of π
s
1
as an integrable locked momentum π
l
s
1
=
˙ q
x
+µ
l
s
1
(q
s
) ˙ q
s
associated with the locked shape q
s
= ¯ q
s
and a nonintegrable normalized
error momentum π
e
s
1
. Then, there exists a function γ(q
s
) in explicit form satisfying
∇
qs
γ(q
s
) = µ
l
s
1
(q
s
) such that the following change of coordinates over U
q
r
= q
x
+ γ(q
s
)
p
r
= m
s
1
x
(q
s
)( ˙ q
x
+ µ
l
s
1
(q
s
) ˙ q
s
)
(4.50)
transforms the dynamics of the underactuated system in (4.49) into a cascade non
linear system in nontriangular quadratic normal form
˙ q
r
= m
−1
r
(q
s
1
, q
s
2
)p
r
˙ p
r
= g
r
(q
r
, q
s
1
, q
s
2
) + Σ(q
s
1
, q
s
2
, p
r
, p
s
) + δ
˙ q
s
1
= p
s
1
˙ p
s
1
= u
1
˙ q
s
2
= p
s
2
˙ p
s
2
= u
2
(4.51)
with a perturbation δ = −˙ p
e
where p
e
= m
r
(q
s
)π
e
s
1
is the error momentum and
m
r
(q
s
) := m
s
1
x
(q
s
)
V
r
(q
r
, q
s
) := V (q
x
, q
s
)[
qx=qr−γ(qs)
g
r
(q
r
, q
x
) := −∇
qs
1
V
r
(q
r
, q
s
1
, q
s
2
)
Σ(q
s
1
, q
s
2
, p
r
, p
s
) := ∇
qs
1
K
The perturbation δ is in the form
δ = m
r
(q
s
)µ
e
s
1
(q
s
)u +p
T
s
Π
δ
(q
s
)p
s
which is aﬃne in u, quadratic in p
s
with a cubic weigh matrix Π
δ
(q
s
) , and vanishes
uniformly in u at (q
s
, p
s
) = (¯ q
s
, 0). In addition, if V (q) = V (q
x
), the reduced system
is a Lagrangian system itself with reduced parameterized Lagrangian
L
r
(q
r
, ˙ q
r
, q
s
) =
1
2
˙ q
T
r
m
r
(q
s
) ˙ q
r
92
that satisﬁes the forced EulerLagrange equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= Σ(q
s
, p
r
, p
s
) + δ (4.52)
Proof. By deﬁnition of q
r
and p
r
, ˙ q
r
= m
−1
s
1
x
(q
s
)p
r
. From the second line of (4.49),
we have
d
dt
∂L
∂ ˙ q
s
1
=
∂L
∂q
s
1
= −∇
qs
1
V (q
x
, q
s
) +∇
qs
1
K
which after substituting q
x
= q
r
−γ(q
s
) and ˙ q
x
= m
−1
s
1
x
(q
s
)p
r
−µ
l
s
1
(q
s
)p
s
, one obtains
d
dt
∂L
∂ ˙ q
s
1
= g
r
(q
r
, q
s
) + Σ(q
s
, p
r
, p
s
)
Based on momentum decomposition theorem
∂L
∂ ˙ q
s
1
= m
r
(q
s
)(π
l
s
1
+ π
e
s
1
) = p
r
+ p
e
Hence ˙ p
r
= g
r
(q
r
, q
s
) +Σ(q
s
, p
r
, p
s
) − ˙ p
e
. The forced EulerLagrange equation follows
from the last equation in a straightforward fashion.
Remark 4.5.5. Assuming ¯ q
s
2
= 0, the normal form in (4.51) can be rewritten as the
following
˙ q
r
= m
−1
r
(q
s
1
, 0)p
r
+ ϕ
1
˙ p
r
= g
r
(q
r
, q
s
1
, 0) + Σ(q
s
1
, 0, p
r
, p
s
1
, 0) + ϕ
2
˙ q
s
1
= p
s
1
˙ p
s
1
= u
1
˙ q
s
2
= p
s
2
˙ p
s
2
= u
2
(4.53)
where
ϕ
1
= (m
−1
r
(q
s
1
, q
s
2
) −m
r
(q
s
1
, 0))p
r
ϕ
2
= g
r
(q
r
, q
s
1
, q
s
2
) −g
r
(q
r
, q
s
1
, 0) + Σ(q
s
1
, q
s
2
, p
r
, p
s
1
, p
s
2
) −Σ(q
s
1
, 0, p
r
, p
s
1
, 0) + δ
where ϕ
1
and the ﬁrst two terms of ϕ
2
vanish at (q
s
2
, p
s
2
) = 0. If δ vanishes at
(q
s
2
, ˙ q
s
2
, u
2
) = 0, then both nonlinear perturbations ϕ
1
, ϕ
2
can be made exponentially
vanishing over compact domains by applying the linear control u
2
= K
2
(q
s
2
, p
s
2
) =
−c
1
q
s
2
−c
2
p
s
2
with c
1
, c
2
> 0. This reduces the control design for the overall system
in (4.51) to control of its (q
r
, p
r
, q
s
1
)subsystem with control input p
s
1
˙ q
r
= m
−1
r
(q
s
1
, 0)p
r
˙ p
r
= g
r
(q
r
, q
s
1
, 0) + Σ(q
s
1
, 0, p
r
, p
s
1
, 0)
˙ q
s
1
= p
s
1
(4.54)
Later, we show that under certain conditions, stabilization of the last nonlinear system
93
reduces to stabilization of the following reduced system
˙ q
r
= m
−1
r
(q
s
1
, 0)p
r
˙ p
r
= g
r
(q
r
, q
s
1
, 0)
(4.55)
with control input q
s
1
. Since q
r
and q
s
1
have the same dimensions, this stabilization
problem can be addressed (see section 4.7 of this chapter).
Example 4.5.2. (threelink planar robot) In this example, we consider a threelink
planar robot as shown in Figure 46 with two controls. Both the external variable θ
1
and one of the two shape variables θ
2
, θ
3
are actuated. Notice that this is diﬀerent
than the case discussed in example 4.5.1 where both shape variables are actuated.
The inertia matrix and dynamics of this robot are given in section A.2. The inertia
x
y
l
L
1
1
θ
θ
3
2
1
θ
, m ,
1
I
1
τ
τ
3
2
τ
1
Figure 46: A planar threelink robot arm.
matrix of this robot has the following structure
M = M(θ
2
, θ
3
) =
m
11
(θ
2
, θ
3
) m
12
(θ
2
, θ
3
) m
13
(θ
2
, θ
3
)
m
21
(θ
2
, θ
3
) m
22
(θ
3
) m
23
(θ
3
)
m
31
(θ
2
, θ
3
) m
32
(θ
3
) m
33
¸
¸
All of the normalized momentums of this threelink planar robot are nonintegrable
(see lemma A.2.1). In the following, we consider two cases: i) θ
2
is unactuated, and
ii) θ
3
is unactuated. In each case, based on theorem 4.5.3 and using momentum
decomposition procedure we obtain reduced normal forms of the system.
Case (i): θ
2
is unactuated. Set (q
x
, q
s
1
, q
s
2
) = (θ
1
, θ
2
, θ
3
). Assuming m
12
(θ
2
, θ
3
) >
0 for all θ
2
, θ
3
, using a global noncollocated partially linearizing feedback law, the
dynamics of the shape variables can be transformed into
˙
θ
2
= ω
2
˙ ω
2
= u
1
˙
θ
3
= ω
3
˙ ω
3
= u
2
94
Then, calculating the normalized momentum conjugate to θ
2
, we get
π
θ
2
=
˙
θ
1
+
m
22
(θ
3
)
m
21
(θ
2
, θ
3
)
˙
θ
2
+
m
23
(θ
3
)
m
21
(θ
2
, θ
3
)
˙
θ
3
Thus
µ
θ
2
(θ
2
, θ
3
) =
¸
m
22
(θ
3
)
m
21
(θ
2
, θ
3
)
m
23
(θ
3
)
m
21
(θ
2
, θ
3
)
Now, according to the locked shape (
¯
θ
2
,
¯
θ
3
) = (0, 0), the locked shape inertia matrix
µ
l
θ
2
is given by
µ
l
θ
2
(θ
2
, θ
3
) =
¸
m
22
(0)
m
21
(θ
2
, 0)
m
23
(θ
3
)
m
21
(0, θ
3
)
Deﬁning γ
1
and γ
2
as the integral of the elements of µ
l
θ
2
γ
1
(θ
2
) =
θ
2
0
m
22
(0)
m
21
(s, 0)
ds, γ
2
(θ
3
) =
θ
3
0
m
23
(0)
m
21
(0, s)
ds,
and applying the global change of coordinates
θ
r
= θ
1
+ γ
1
(θ
2
) + γ
2
(θ
3
)
ω
r
= m
12
(θ
2
, θ
3
)(ω
1
+
m
22
(0)
m
21
(θ
2
, 0)
ω
2
+
m
23
(θ
3
)
m
21
(0, θ
3
)
ω
3
)
the dynamics of the threelink robot with unactuated shape variable θ
2
transforms
into the following nontriangular quadratic normal form
˙
θ
r
= m
12
(θ
2
, θ
3
)ω
r
˙ ω
r
= −g
r
(θ
r
, θ
2
, θ
3
) + Σ(θ
2
, θ
3
, ω
r
, ω
2
, ω
3
) − ˙ p
e
˙
θ
2
= ω
2
˙ ω
2
= u
1
˙
θ
3
= ω
3
˙ ω
3
= u
2
(4.56)
where the error momentum p
e
is given by
p
e
= (
m
22
(θ
3
)
m
21
(θ
2
, θ
3
)
−
m
22
(0)
m
21
(θ
2
, 0)
)ω
2
+ (
m
23
(θ
3
)
m
21
(θ
2
, θ
3
)
−
m
23
(θ
3
)
m
21
(0, θ
3
)
)ω
3
Clearly, p
e
vanishes at (θ
3
, ω
3
) = (0, 0). This implies the perturbation δ = −˙ p
e
vanishes at (θ
3
, ω
3
, u
2
) = (0, 0, 0) as well. Following the procedure in remark 4.5.5,
stabilization of (4.56) reduces to stabilization of its fourthorder subsystem as
˙
θ
r
= m
12
(θ
2
, 0)ω
r
˙ ω
r
= −g
r
(θ
r
, θ
2
, 0) + Σ(θ
2
, 0, ω
r
, ω
2
, 0)
˙
θ
2
= ω
2
˙ ω
2
= u
1
(4.57)
95
which is exactly the normal form for the Pendubot with inertia matrix and potential
energy
M
Pen
(q
1
) =
¸
m
11
(θ
2
, 0) m
12
(θ
2
, 0)
m
21
(θ
2
, 0) m
22
(0)
, V
Pen
(θ
1
, θ
2
) = V (θ
1
, θ
2
, 0)
and an unactuated shape variable θ
2
. Later, we show that the Pendubot can be
stabilized using Fixed Point Controllers [67].
Case (ii): θ
3
is unactuated. Take (q
x
, q
s
1
, q
s
2
) = (θ
1
, θ
3
, θ
2
). After a similar argu
ment to Case (i), it can be shown that the control of the overall system reduces to
control of the Pendubot with inertia matrix and potential energy
M
Pen
(q
1
) =
¸
m
11
(0, θ
3
) m
13
(0, θ
3
)
m
31
(0, θ
3
) m
33
, V
Pen
(θ
1
, θ
3
) = V (θ
1
, 0, θ
3
)
Notice that this Pendubot has an unactuated shape variable θ
3
. In this case, the per
turbations ϕ
1
, ϕ
2
in (4.53) vanish exponentially over a compact domain by applying
the linear feedback u
2
= K
2
(θ
2
, ω
2
) = −c
1
θ
2
−c
2
ω
2
with c
1
, c
2
> 0.
Remark 4.5.6. In conclusion, based on Figure 47, control of the middle robot reduces
to control of the Acrobot in example 4.5.1 and control of the other two robots reduces
to control of the Pendubot as discussed in Cases (i) and (ii) of example 4.5.2. This
fact can be intuitively understood by the following graphical procedure without any
calculations.
Figure 47: Three possible actuation conﬁguration for a planar threelink robot arm.
Graphical Reduction Procedure for an nlink underactuated planar robot:
1) Set j = n.
2) If j is the last actuated joint, QUIT.
3) If joint j is actuated, lock the joint angle θ
j
= 0 and merge the jth link and the
(j −1)th link as one. Set j := j −1 and GOTO step 2.
4) If joint j is unactuated, set j := j −1 and GOTO step 2.
96
This algorithm ends up with the Acrobot or the Pendubot as the ﬁnal underactuated
system with a single actuator and two remaining links.
Proposition 4.5.1. Consider the 3D CartPole system with conﬁguration vector
(q
x
, q
s
) = (x
1
, x
2
, θ, φ) and unactuated shape variables (θ, φ). Then, the following
change of coordinates over U = R
2
(−π/2, π/2)
2
z
1
= x + γ(θ, φ)
z
2
= ˙ x + (l
˙
θ/ cos θ, l
˙
φ/cosφ)
T
(4.58)
with γ(θ, φ) = (γ
0
(θ), γ
0
(φ))
T
and γ
0
deﬁned as
γ
0
(θ) =
θ
0
l
cos s
ds = l log
1 + tan(θ/2)
1 −tan(θ/2)
, θ ∈ (−π/2, π/2)
transforms the dynamics of the 3D CartPole System into a cascade nonlinear system
in feedforward form
˙ z
1
= z
2
˙ z
2
= ψ(q
s
) + p
T
s
Π(q
s
)p
s
+ ϕ(q
s
)u
˙ q
s
= p
s
˙ p
s
= u
(4.59)
where
ψ(q
s
) = −m
−1
sx
(q
s
)∇
qs
V (q
s
)
has a nonsingular Jacobian at q
s
= 0, Π(q
s
) is a cubic matrix, and ϕ(q
s
) = −µ
e
s
(q
s
)
satisﬁes ϕ(0) = 0
2×2
and σ
max
(ϕ(q
s
)) = O([q
s
[
2
). In addition, the origin for (4.59)
can be globally asymptotically and locally exponentially stabilized using a state feedback
in the form of nested saturations.
Proof. We use the method of momentum decomposition to reduce the dynamics
of the 3D CartPole system. This system has equal number of external and shape
variables. Thus, following the notation of theorem 4.5.3, dim(Q
s
2
) = 0 and the
external and shape conﬁguration vectors of the 3D CartPole system are q
x
= (x
1
, x
2
)
and q
s
= q
s
1
= (θ, φ), respectively. From the kinetic energy of the 3D CartPole
system in (4.34), the block matrices m
sx
(q
s
) and m
ss
(q
s
) for this system can be
identiﬁed as
m
sx
(q
s
) =
¸
ml cos θ −ml sin θ sin φ
0 ml cos θ cos φ
, m
ss
(q
s
) =
¸
ml
2
0
0 ml
2
cos
2
θ
In example 4.4.2, it was shown that the normalized momentum
π
s
= ˙ q
x
+µ
s
(q
s
) ˙ q
s
97
is nonintegrable over U = R
2
(−π/2, π/2)
2
where
µ
s
(q
s
) = m
−1
sx
(q
s
)m
ss
(q
s
) =
l
cos θ
l sin θ sin φ
cos φ
0
l cos θ
cos φ
¸
¸
¸
Hence, after choosing the lock conﬁguration (
¯
θ,
¯
φ) = (0, 0) and setting φ = 0 and
θ = 0, respectively, in the ﬁrst and second columns of µ
s
, the normalized locked
shape inertia and error inertia for the system can be obtained as
µ
l
s
(θ, φ) =
l
cos θ
0
0
l
cos φ
¸
¸
¸
, µ
e
s
(θ, φ) =
0
l sin θ sin φ
cos φ
0
l(cos θ −1)
cos φ
¸
¸
¸
Take
γ
0
(z) =
z
0
l
cos s
ds , z ∈ (π/2, π/2)
Then, γ(θ, φ) = (γ
0
(θ), γ
0
(φ))
T
satisﬁes ∇
qs
γ(q
s
) = µ
l
s
(q
s
). The change of coordinates
in (4.50) can be explicitly calculated as
q
r
= x + γ(θ, φ)
p
r
= m
sx
(θ, φ)( ˙ x + µ
l
s
(θ, φ) ˙ q
s
)
(4.60)
After simpliﬁcation, we obtain
q
1
r
= x
1
+ γ
0
(θ)
p
1
r
= ml cos(θ) ˙ x
1
−ml sin(θ) sin(φ) ˙ x
2
+ ml
2
˙
θ −ml
2
sin(θ) tan(φ)
˙
φ
q
2
r
= x
2
+ γ
0
(φ)
p
2
r
= ml cos(θ) cos(φ) ˙ x
2
+ ml
2
cos(θ)
˙
φ
(4.61)
where q
r
= (q
1
r
, q
2
r
)
T
is the conﬁguration vector of the reduced system and p
r
=
(p
1
r
, p
2
r
)
T
is its conjugate momentum. The nonlinear change of coordinates in (4.61)
transforms the dynamics of the 3D CartPole system into
˙ q
r
= m
−1
r
(θ, φ)p
r
˙ p
r
= g
r
(θ, φ) + Σ(q
s
, p
r
, p
s
) + δ
˙
θ = ω
1
˙ ω
1
= u
1
˙
φ = ω
2
˙ ω
2
= u
2
(4.62)
where m
r
= m
sx
and
g
r
(θ, φ) = −∇
qs
V (θ, φ) =
¸
mgl sin θ cos φ
mgl cos θ sin φ
98
Notice that m
xx
is constant for this system and Σ is linear in ˙ q
x
, thus
Σ(q
s
, p
r
, p
s
) = ∇
qs
K = ∇
qs
¦ ˙ q
T
s
m
sx
(q
s
) ˙ q
x
¦ +∇
qs
¦
1
2
˙ q
T
s
m
ss
(q
s
) ˙ q
s
¦
Following the line of proposition 4.4.1, apply a second change of coordinates as the
following
z
1
= q
r
z
2
= m
−1
r
(θ, φ)p
r
(4.63)
We prove that the last change of coordinates transforms the dynamics of the 3D
CartPole system into the feedforward normal form in (4.59). First, observe that
z
1
= x + γ(θ, φ)
z
2
= ˙ x + µ
l
s
(θ, φ) ˙ q
s
(4.64)
and therefore ˙ z
1
= z
2
. In addition, we have
∂L
∂ ˙ q
s
= m
sx
(q
s
)( ˙ q
x
+ µ
s
(q
s
) ˙ q
s
)
= m
sx
(q
s
)( ˙ q
x
+ µ
l
s
(q
s
) ˙ q
s
+ µ
e
s
(q
s
) ˙ q
s
)
= m
sx
(q
s
)z
2
+ p
e
where
p
e
= m
sx
(q
s
)µ
e
s
(q
s
) ˙ q
s
But from EulerLagrange equation
d
dt
∂L
∂ ˙ q
s
=
∂L
∂q
s
= g
r
(q
s
) + Σ
and we obtain
m
sx
(q
s
) ˙ z
2
= g
r
(q
s
) + Σ − ˙ m
sx
z
2
− ˙ p
e
= g
r
(q
s
) +∇
qs
¦
1
2
˙ q
T
s
m
ss
(q
s
) ˙ q
s
¦
+ ∇
qs
¦ ˙ q
T
s
m
sx
(q
s
) ˙ q
x
¦ − ˙ m
sx
˙ q
x
− ˙ m
sx
µ
l
s
(q
s
) ˙ q
s
− ˙ p
e
But similar to condition ii) of proposition 4.4.1, m
sx
(q
s
) block of the inertia matrix
of the 3D CartPole system has diﬀerentially symmetric rows that implies
∇
qs
¦ ˙ q
T
s
m
sx
(q
s
) ˙ q
x
¦ = ˙ m
sx
˙ q
x
Therefore, we obtain the following equations for the zsubsystem
˙ z
1
= z
2
˙ z
2
= ψ(q
s
) +
˜
Σ(q
s
, p
s
) + ϕ(q
s
)u
(4.65)
99
where
ψ(q
s
) = m
−1
sx
(q
s
)g
r
(q
s
)
˜
Σ(q
s
, p
s
) = m
−1
sx
(q
s
)[∇
qs
¦
1
2
˙ q
T
s
m
ss
(q
s
) ˙ q
s
¦ − ˙ m
sx
µ
l
s
(q
s
) ˙ q
s
−
d
dt
¦m
sx
(q
s
)µ
e
s
(q
s
)¦ ˙ q
s
]
ϕ(q
s
) = −µ
e
s
(q
s
)
and
˜
Σ(q
s
, p
s
) = p
T
s
Π(q
s
)p
s
is quadratic in p
s
. Notice that from the explicit expression
of µ
e
s
(q
s
), ϕ(0) = 0
2×2
and σ
max
(ϕ(q
s
)) = O([(θ, φ)[
2
). The stabilization result using
nested saturations [102] follows from the fact that
˜
Σ is quadratic and [∇
qs
ψ(q
s
)]
qs=0
=
gI
2×2
is nonsingular (where g is the gravity constant).
4.6 Classiﬁcation of Underactuated Systems
In this section, we classify underactuated systems based on possession or lack of some
or all of the following four properties:
(a) Actuated shape variables
(b) Noninteracting inputs (i.e. lack of input coupling)
(c) Integrable normalized momentums
(d) Extra required conditions (as described in the deﬁnition of each class in the
following).
This leads at most to sixteen possible classes. However, due to redundancy and the
fact that some of these cases never have appeared in applications, we only focus on
a subset of these sixteen classes that frequently appear in control of underactuated
mechanical systems. This subset includes eight diﬀerent classes.
Note: In case an underactuated system is ﬂat (i.e. has constant inertia matrix), we
call the variables that appear in the force matrix F(q) shape variables.
Table 4.1 provides deﬁnition and examples of diﬀerent classes of underactuated
systems. The following remarks have to be taken into account regarding Table 4.1:
– a “Yes” or a “No” means the corresponding property holds or does not hold,
respectively.
– “Q”, “L”, and “LQ” in front of “Nontriangular” refer to “Quadratic”, “Linear”,
and “LinearQuadratic” structure of the normal form with respect to p
s
(i.e.
shape velocity).
– a “Yes” for property (d) means that certain extra conditions are required that
can be found in the description of the corresponding class.
100
– a controlled VTOL aircraft means the dynamics of the VTOL in closed loop
with a feedback u
1
= u
1
(x, y, ˙ x, ˙ y) and a single remaining control u
2
.
Table 4.1: Classiﬁcation of Underactuated Systems
Class (a) (b) (c) (d) Normal Form Control Design example(s)
I Yes Yes Yes No Strict Feedback Backstepping
Acrobot (2D,1C), TORA,
InertiaWheel Pendulum
IIa No Yes Yes No Nontriangular Q FixedPoint Law
Rotating Pendulum
Pendubot, BeamnBall
IIb No Yes Yes Yes Nontriangular L FixedPoint Law ≈Flexiblelink Robots
III No Yes Yes Yes Feedforward Nested Saturations 2D CartPole (2D,1C)
IVa Yes No Yes No Nontriangular Q FixedPoint Law Controlled VTOL (3D,1C)
IVb Yes No Yes Yes Nontriangular Q FixedPoint Law Controlled VTOL (3D,1C)
V Yes No Yes Yes Strict Feedback Backstepping
VTOL aircraft (3D,2C))
≈Aircraft (6D,4C)
≈Helicopter (6D,4C)
VIa Yes Yes No No Nontriangular L FixedPoint Law Threelink arm (3D,2C)
VIb Yes Yes No Yes Feedback+∆ Backstepping Threelink arm (3D,2C)
VIIa No Yes No No Nontriangular LQ FixedPoint Law Threelink arm (3D,2C)
VIIb No Yes No Yes Nontriangular+∆ FixedPoint Law Threelink arm (3D,2C)
VIII No Yes No Yes Feedforward Nested Saturations 3D CartPole (4D,2C)
In the following, we rewrite the normal form corresponding to each class of under
actuated systems in table 4.1 by partitioning the state vector as col(z, ξ) ∈ R
n
R
r
.
Throughout this section, (z
1
, z
2
, ξ
1
, ξ
2
) denotes (q
r
, p
r
, q
s
, p
s
) unless otherwise is stated.
Furthermore, the term f
0
is due to the potential energy of the system. Here is the
description of each class:
• ClassI: underactuated systems with actuated shape variables, integrable mo
mentums, and noninteracting inputs. From equation (4.6), the normal form of
ClassI underactuated systems can be rewritten as the following
˙ z
1
= N(ξ
1
)z
2
˙ z
2
= f
0
(z
1
, ξ
1
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.66)
where N(ξ
1
) is a symmetric and positive deﬁnite matrix. System (4.66) is in
strict feedback form with a vector ﬁeld nonaﬃne in ξ
1
.
• ClassIIa: underactuated systems with unactuated shape variables, integrable
normalized momentums, noninteracting inputs, and equal number of exter
nal and shape variables. From equation (4.14), the normal form of ClassIIa
101
underactuated systems can be rewritten as the following
˙ z
1
= z
2
˙ z
2
= f
0
(z
1
, ξ
1
) + z
T
2
g
11
(ξ
1
)z
2
+ z
T
2
g
12
(ξ
1
)ξ
2
+ ξ
T
2
g
22
(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.67)
where z
2
= m
−1
r
(q
s
) and g
ij
’s are cubic matrices with the property g
12
= g
T
21
.
System (4.67) is in nontriangular quadratic form with a vector ﬁeld nonaﬃne
in (ξ
1
, ξ
2
).
• ClassIIb: a subclass of ClassIIa underactuated systems with unequal number
of external and shape variables. From equation (4.21), the normal form of Class
IIb underactuated systems can be rewritten as the following
˙ z
1
= z
2
˙ z
2
= f
0
(z
1
) + f
d
(z
1
, z
2
) + z
T
2
g
11
(ξ
1
)z
2
+ z
T
2
g
12
(ξ
1
)ξ
2
+ ξ
T
2
g
22
(ξ
1
)ξ
2
+g
3
(z
1
)ξ
3
˙
ξ
1
= ξ
2
˙
ξ
2
= ξ
3
˙
ξ
3
= u
(4.68)
where z, ξ are deﬁned in (4.20) and F
d
is a force due to dissipation satisfying
z
T
2
f
d
(z
1
, z
2
) < 0, ∀z
1
, ∀z
2
= 0. The g
ij
’s are cubic matrices with the property
g
12
= g
T
21
. System (4.68) is in nontriangular form with a vector ﬁeld nonaﬃne
in (ξ
1
, ξ
2
) and aﬃne in ξ
3
.
• ClassIII: a subclass of ClassIIa underactuated systems satisfying conditions
of proposition 4.2.1. From equation (4.17) the normal form of ClassIII under
actuated systems can be rewritten as
˙ z
1
= z
2
˙ z
2
= f
0
(ξ
1
) + ξ
T
2
g(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.69)
where z
2
= m
−1
r
(q
s
)p
r
and g(ξ) is a cubic matrix. System (4.69) is in strict
feedforward form with a vector ﬁeld nonaﬃne in (ξ
1
, ξ
2
).
• ClassIVa: underactuated systems with actuated shape variables, integrable
momentums, and input coupling. From equation (4.24), taking z
2
= m
−1
r
(q
s
)p
r
,
the normal form for ClassIVa underactuated systems can be written as
˙ z
1
= z
2
˙ z
2
= f
0
(z
1
, ξ
1
) + z
T
2
g
11
(ξ
1
)z
2
+ z
T
2
g
12
(ξ
1
)ξ
2
+ ξ
T
2
g
22
(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.70)
102
The g
ij
’s are cubic matrices with the property g
12
= g
T
21
. System (4.70) is in
nontriangular quadratic form with a vector ﬁeld nonaﬃne in (ξ
1
, ξ
2
). Notice
that normal forms (4.70) and (4.67) represent the same class of cascade nonlinear
systems, but two diﬀerent classes of underactuated systems.
• ClassIVb: a subclass of ClassIVa underactuated systems with constant m
xx
and V (q) = V (q
s
). From equation (4.25), the normal form for ClassIVb un
deractuated systems can be written as
˙ z
1
= z
2
˙ z
2
= f
0
(ξ
1
) + ξ
T
2
g(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.71)
where z
2
= m
−1
r
p
r
and g(ξ
1
) is a cubic matrix. System (4.71) is in strict feed
forward form with a vector ﬁeld nonaﬃne in (ξ
1
, ξ
2
).
• ClassVa: underactuated systems with actuated shape variables, integrable
momentums, input coupling, and an extra control in the dynamics of the exter
nal variables with the property that their m
xx
is constant and V (q) = V (q
x
).
From equation (4.28), the normal form for ClassVa underactuated systems can
be written as
˙ z
1
= z
2
˙ z
2
= f
0
(z
1
, ξ
1
) + g
0
(ξ
1
)T + ξ
T
2
g(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.72)
where z
2
= m
−1
r
p
r
, g(ξ
1
) is a cubic matrix, g
0
(ξ
1
) : R
p
→R
n
is deﬁned by
g
0
(ξ
1
) :=
1
m
xx
R(ξ
1
)e, (4.73)
e is a ﬁxed unit vector in R
n
and R(ξ
1
) ∈ SO(n) is a rotation matrix in R
n
. The
control T ∈ R (or thrust) is the magnitude of a force applied to the dynamics
of the vector of external variables, q
x
∈ R
n
, in the direction e. Assuming that
T = K(z, ξ
1
), system (4.72) is in nontriangular quadratic normal form with a
vector ﬁeld nonaﬃne in (ξ
1
, ξ
2
).
• ClassVb: a subclass of ClassVa underactuated systems with the property
that there exists a positive deﬁnite and symmetric matrix Q(ξ
1
) ∈ R
p×p
such
that
ξ
T
2
g(ξ
1
)ξ
2
= [ξ
T
2
Q(ξ
1
)ξ
2
]g
0
(ξ
1
) (4.74)
103
and V (q
x
) is a linear function. From theorem 4.3.2 and equation (4.72), the
normal form for ClassVb underactuated systems can be given as
˙ z
1
= z
2
˙ z
2
= f
0
+ g
0
(ξ
1
)
¯
T
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.75)
where f
0
∈ R
n
is a constant vector and
¯
T in a new control deﬁned as
¯
T = T + ξ
T
2
Q(ξ
1
)ξ
2
Assuming that the state feedback
¯
T = K(z, ξ
1
) globally asymptotically stabi
lizes the equilibrium z = z
0
for the zsubsystem of (4.75), the dynamics of (4.75)
is in strict feedback form with a vector ﬁeld nonaﬃne in ξ
1
.
• Class VIa: underactuated systems with actuated shape variables, nonintegrable
normalized momentums, and noninteracting inputs. From equation (4.31), the
normal form for ClassVIa underactuated systems can be expressed as
˙ z
1
= N(ξ
1
)z
2
+ g
0
(ξ
1
)ξ
2
˙ z
2
= f
0
(z
1
, ξ
1
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.76)
System (4.76) is in nontriangular linear normal form with a vector ﬁeld aﬃne
in ξ
2
. Note that h = h(ξ
1
) : R
p
→R
n
: Dh(ξ
1
) = g
0
(ξ
1
).
• Class VIb: a subclass of ClassVIa with the property that there exists a parti
tion of q
s
= col(q
s
1
, q
s
2
) with an ndimensional vector q
s
1
such that the perturba
tion δ in (4.47) vanishes identically at (q
s
2
, p
s
2
, u
2
) = 0 where ˙ q
s
2
= p
s
2
, ˙ p
s
2
= u
2
.
Let us partition ξ
i
and u as col(η
i
, µ
i
), i = 1, 2 and col(u
1
, u
2
), respectively.
Then, from equation (4.47), the normal form for ClassVIb underactuated sys
tems can be obtained as
˙ z
1
= N(η
1
)z
2
+ ϕ
1
(z
2
, η
1
, µ
1
)
˙ z
2
= f
0
(z
1
, η
1
) + ϕ
2
(z
1
, η, µ, u)
˙ η
1
= η
2
˙ η
2
= u
1
˙ µ
1
= µ
2
˙ µ
2
= u
2
(4.77)
where N(η
1
) = m
−1
r
(q
s
1
, 0) and the perturbation terms ϕ
1
, ϕ
2
vanish identically
at (µ
1
, µ
2
, u
2
) = 0. The (z, η)subsystem of (4.77) is a perturbation of a strict
feedback form.
104
• Class VIIa: underactuated systems with unactuated shape variables, non
integrable normalized momentums, and noninteracting inputs. From equation
(4.32), the normal form for ClassVIIa underactuated systems can be given as
˙ z
1
= z
2
+ g
0
(ξ
1
)ξ
2
˙ z
2
= f
0
(z
1
, ξ
1
) + z
T
2
g
11
(ξ
1
)z
2
+ z
T
2
g
12
(ξ
1
)ξ
2
+ ξ
T
2
g
22
(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
1
= u
(4.78)
where z
2
= m
−1
r
(q
s
)p
r
and the g
ij
’s are cubic matrices satisfying g
12
= g
T
21
.
System (4.78) is in nontriangular linearquadratic normal form with a vector
ﬁeld nonaﬃne in (ξ
1
, ξ
2
).
• Class VIIb: underactuated systems with partially unactuated shape variables,
nonintegrable normalized momentums, and noninteracting inputs. In addi
tion, there exists a partition of q
s
= col(q
s
1
, q
s
2
) with an ndimensional vector
q
s
1
such that the perturbation δ in (4.51) vanishes identically at (q
s
2
, p
s
2
, u
2
) = 0.
Following the notation used in the deﬁnition of ClassVIb, based on equation
(4.51), the normal form for ClassVIIb underactuated systems can be written
as
˙ z
1
= z
2
+ ϕ
1
(z
2
, η
1
, µ
1
)
˙ z
2
= f
0
(z
1
, η
1
) + z
T
2
g
11
(η
1
)z
2
+z
T
2
g
12
(η
1
)η
2
+η
T
2
g
22
(η
1
)η
2
+ϕ
2
(z, η, µ, u)
˙ η
1
= η
2
˙ η
2
= u
1
˙ µ
1
= µ
2
˙ µ
2
= u
2
(4.79)
where the g
ij
’s are cubic matrices and ϕ
1
, ϕ
2
vanish identically at (µ
1
, µ
2
, u
2
) =
0. The (z, η)subsystem of (4.79) is a perturbation of a nontriangular quadratic
normal form (see the normal form of ClassIIa).
• Class VIII: a subclass of ClassVIIa underactuated systems satisfying the con
ditions of proposition 4.4.1. From equation (4.33), the normal form for Class
VIII underactuated systems can be expressed as
˙ z
1
= z
2
+ϕ
1
(ξ
1
, ξ
2
)
˙ z
2
= f
0
(ξ
1
) + ξ
T
2
g(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
1
= u
(4.80)
where
z
1
= q
r
+ µ(0)q
s
, z
2
= m
−1
r
(q
s
)p
r
and ϕ
1
∼ O([(ξ
1
, ξ
2
)[
2
) as deﬁned in proposition (4.4.1). System (4.80) is in
strict feedforward form with a vector ﬁeld nonaﬃne in (ξ
1
, ξ
2
).
Let us make the following deﬁnition.
105
Deﬁnition 4.6.1. (core reduced system) Deﬁne the core reduced system as the dy
namics of the zsubsystem without the following types of terms:
i) terms containing ξ
2
,
ii) terms containing z
2
in the equation of ˙ z
2
,
iii) highorder perturbation terms ϕ
i
.
The following corollary creates a connection in terms of the dynamics of the z
subsystem among all the above classes of underactuated system.
Corollary 4.6.1. According to deﬁnition 4.6.1, for all the aforementioned classes of
underactuated mechanical systems, the core reduced system is in the normal form
˙ z
1
= N(v)z
2
˙ z
2
= f(z
1
, v)
(4.81)
where v is the control input for (4.81) and N(v) is a positive deﬁnite symmetric
matrix. (Note that for several classes N(v) = I).
Global asymptotic stabilization of nonlinear systems in normal form (4.81) is
addressed in theorem 4.7.1 for the case where z
1
, z
2
, v ∈ R
n
.
Remark 4.6.1. Based on proposition 5.11.4. The normal form of the ﬁrstlevel ap
proximate model of a helicopter is in the form
˙ z = f(z) + g
1
(ξ
1
)T + g
2
(ξ
1
, ξ
2
, ) +g
3
(ξ
1
, )u
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(4.82)
where z = col(x, v) ∈ R
6
, ξ
1
= col(φ, θ, ψ) (i.e. three Euler angles), = (
1
,
2
,
3
)
T
∈
R
3
, and g
2
, g
3
are linear with respect to the matrix
E() =
0
2
0
1
0
3
0 0 0
¸
¸
;
1
> 0,
2
< 0, ,
3
> 0
Thus, g
2
, g
3
vanish identically at = (0, 0, 0). The vector ﬁeld of (4.82) is quadratic
in ξ
2
and aﬃne in control u. Under the assumption that = (0, 0, 0), (4.82) reduces
to the normal form of a ClassVIb underactuated system in (4.75).
4.7 Nonlinear Control of the Core Reduced Sys
tem
The fundamental property of the three basic cascade normal forms for underactu
ated systems is that the unforced and unperturbed reduced system for all classes of
106
underactuated systems in section 4.6 is the same and is given by
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= g
r
(q
r
, q
s
)
(4.83)
We refer to (4.83) as the Core Reduced System (also, see deﬁnition 4.6.1). This section
is devoted to stabilization of the core reduced system for all underactuated systems
discussed in this chapter. The main property of the core reduced system is that it
is a simple Lagrangian system without any external force. Thus, the shape vector q
s
plays the role of the control input for the core reduced system. Here is our main result
on global asymptotic stabilization of the core reduced system (4.83). (For simplicity
of the notation, in the following theorem the subscript “r” is dropped).
Theorem 4.7.1. Consider the following nonlinear system nonaﬃne in control
˙ q = N(u)p
˙ p = g(q, u)
(4.84)
where q, p, u ∈ R
n
, g(q, u) : R
n
R
n
→ R
n
is a smooth function with g(0, 0) = 0,
N(u) is an invertible matrix for all u, and M(u) = N
−1
(u) is a positive deﬁnite and
symmetric inertia matrix. Suppose there exists an isolated root u = α(q) of g(q, u) = 0
with the property α(0) = 0 such that
det
∂g
∂u
(q, α(q))
= 0
Set
ψ(q, v) := [M(u)g(q, u)]
u=α(q)+v
Then, for all q ∈ R
n
, w = ψ(q, v) is a local diﬀeomorphism around a neighborhood
of v = 0. Assume there exists an open ball B
r
(0) around w = 0 and a function
β : R
n
R
n
→R
n
such that
ψ(q, β(q, w)) = w, ∀w ∈ B
r
(0) ⊂ R
n
uniformly in q. Let σ(x) : R
n
→R
n
denote the vector sigmoidal function
σ(x) =
x
(1 +x
2
)
1
2
Then, for all c
1
, c
2
∈ (0, r/2], the static state feedback u = K(q, p) deﬁned as the
following
u = K(q, p) := α(q) +β(q, w),
w = K
b
(q, p) := −c
1
σ(q) −c
2
σ(p);
(4.85)
globally asymptotically stabilizes the origin (q, p) = (0, 0) for the nonlinear system in
(4.84).
107
Proof. Calculating ∇
v
ψ(q, v) at v = 0 (or u = α(q)), we get
∇
v
ψ(q, v) = M(α(q)) [∇
v
g(q, α(q) + v)]
v=0
By assumption, both M(α(q)) and ∇
u
g(q, α(q)) are invertible matrices. This means
that ∇
v
ψ(q, v) is invertible at v = 0 and thus ψ(q, v) is a local diﬀeomorphism for all
q. Consider the invertible change of control w = ψ(q, v) with an inverse v = β(q, w)
deﬁned over R
n
B
r
(0) so that ψ(q, β(q, w)) = w. Then, the dynamics of (4.84) can
be written as
˙ q = N(u)p
˙ p = N(u)w
(4.86)
with u = α(q) + β(q, w). Fix c
1
, c
2
∈ (0, r/2]. We have
w = K
b
(q, p) ≤ c
1
σ(q) + c
2
σ(p) < c
1
+ c
2
≤ r
and for all (q, p), K
b
(q, p) ∈ B
r
(0). Applying the state feedback w = K
b
(q, p) in the
question, the closedloop system in (4.86) takes the following form
˙ q = N(u)p
˙ p = −c
1
N(u)σ(q) −c
2
N(u)σ(p)
(4.87)
with u = K(q, p) (deﬁned in (4.85)). Consider the following positive deﬁnite Lya
punov function candidate
H(q, p) = c
1
φ(q) +
1
2
p
2
where the positive semideﬁnite potential function φ(q) : R
n
→R
≥0
is deﬁned as
φ(q) = (1 +q
2
)
1
2
−1
and satisﬁes the property (∇φ(q))
T
= σ(q). Calculating
˙
H along the solutions of the
closedloop system (4.87), we obtain
˙
H =
∂H(q, p)
∂q
˙ q +
∂H(q, p)
∂p
˙ p
= c
1
σ(q)
T
N(u)p −c
1
p
T
N(u)σ(q) −c
2
p
T
N(u)σ(p), (N
T
= N)
= −c
2
p
T
N(u)σ(p) < 0, ∀p = 0
The last inequality is due to positive deﬁniteness of the term
p
T
N(u)σ(p) =
p
T
N(u)p
(1 +p
2
)
1
2
≥ 0
for p = 0 and all u, particularly, u = K(q, p). Based on LaSalle’s invariance principle
[40], all the solutions of the closedloop system converge to the largest invariant set
Ω in ¦(q, p) :
˙
H = 0¦ = ¦(q, 0)¦. But for system (4.87), Ω = ¦(0, 0)¦. Therefore,
108
the origin (q, p) = (0, 0) is globally asymptotically stable for (4.87) (and (4.84)).
Furthermore, H(q, p) is a smooth positive deﬁnite and proper Lyapunov function for
the system.
Deﬁnition 4.7.1. (bisaturated PD controller) We call w = K
b
(q, p) in equation
(4.85), a bisaturated PD controller.
Corollary 4.7.1. Assume all the conditions in theorem 4.7.1 hold and let
w = k
b
(q, p) := −c
0
σ(q + p), c
0
∈ (0, r]
Then, (q, p) = (0, 0) is GAS for (4.84).
Proof. We prove that the following smooth and positive deﬁnite function
H(q, p) = c
0
φ(q + p) +
1
2
p
2
is a valid Lyapunov function for the closedloop system
˙ q = N(u)p
˙ p = −c
0
N(u)σ(q + p)
(4.88)
We have
˙
H = c
0
σ(q + p)
T
( ˙ q + ˙ p) −c
0
p
T
N(u)σ(q + p)
= −c
0
σ(q + p)
2
Thus, all the solutions converge to the largest invariant set Ω in ¦(q, p) : q + p = 0¦.
By inspection, Ω = ¦(0, 0)¦ and the origin is GAS for (4.88) and thus (4.84).
The following result shows that under further structure in the dynamics of the
reduced system it is possible to ﬁnd a static state feedback in closedform that globally
asymptotically stabilizes the origin for the reduced system.
Theorem 4.7.2. Consider the reduced system in (4.83) and assume that dim(Q
r
) =
dim(Q
s
) = d and the reduced inertia matrix is diagonal, i.e.
m
r
(q
s
) = diag(m
r1
(q
s
), . . . , m
rd
(q
s
))
In addition, assume that (4.83) is in the following form
˙ q
j
r
= m
−1
rj
(q
j
s
)p
j
r
˙ p
j
r
= g
rj
(q
j
r
, q
j
s
)
(4.89)
for j = 1, . . . , d. Let q
j
s
= α
j
(q
j
r
) be an isolate root of g
rj
(q
r
, q
s
) for j = 1, . . . , d, i.e.
g
rj
(q
r
, α
j
(q
j
r
)) = 0,
¸
∂g
rj
∂q
j
s
q
j
s
=α
j
(q
j
r
)
> 0
109
Then, the following state feedback
q
s
= α(q
r
) −σ(q
r
+ p
r
)
globally asymptotically stabilizes the equilibrium point (q
r
, p
r
) = (0, 0) of (4.89) where
α(q
r
) = (α
1
(q
1
r
), . . . , α
d
(q
d
r
))
T
, σ(z) = (σ
1
(z
1
), . . . , σ
n
(z
i
))
and all σ
i
’s are one dimensional sigmoidal functions.
The proof of this theorem relays on the following theorem for the onedimensional
case.
Theorem 4.7.3. Consider the following system
˙ q = p,
˙ p = f(p) + g(q, u),
(4.90)
where f and g are C
1
functions with f(0) = 0 and g(0, 0) = 0, f is decreasing
(f
≤ 0), and q, p, u ∈ R. Suppose that zero is not a critical value for g(q, u) and
∂g(q, u)
∂u
= 0
on the manifold
M = Ker(g) = ¦(q, u) ∈ R
2
: g(q, u) = 0¦
Then, g(q, u) has an isolated root α(q) such that
g(q, α(q)) = 0
and there exists a continuously diﬀerentiable state feedback law in the following form
u = α(q) −σ(q + p), (4.91)
that globally asymptotically stabilizes the origin for (4.90) (σ is a sigmoidal function).
Proof. Because zero is not a critical value for g, M = g
−1
(0) = Ker(g) is a one
dimensional diﬀerentiable manifold in R
2
. In other words, M is a twodimensional
curve in plane that divides R
2
into two connected regions:
A
+
= ¦(q, u) ∈ R
2
: u > α(q)¦
and
A
−
= ¦(q, u) ∈ R
2
: u < α(q)¦.
But g
u
(q, u) = 0 on M, based on implicit mapping theorem it follows that g(q, u) = 0
has an isolated root u = α(q) that is a C
1
function and M can be parameterized as
M = ¦(q, α(q)) : q ∈ R¦.
110
Also, on M either g
u
(q, u) > 0 or g
u
(q, u) < 0 (otherwise, at some point on M
g
u
(q, u) = 0). Without loss of generality assume
∂g(q, u)
∂u
[u = α(q) > 0
by continuity of g
u
(q, u) there exists a tubular neighborhood of M
N
δ
(M) = ¦(q, u) ∈ R
2
: [u −α(q)[ < δ, δ > 0¦
such that g
u
(q, u) > 0 on N
δ
(M). Suppose [σ(s)[ < δ, ∀s where σ is a sigmoidal
function and δ > 0. Then, given
u = k(q, p) = α(q) −σ(q + p),
(q, k(q, p)) ∈ N
δ
(M), ∀q, p and therefore
g
u
(q, k(q, p)) > 0
for all q, p. Now, consider the closed loop system (4.90) with control input u = k(q, p)
as the following
˙ q = p,
˙ p = f(p) + g(q, α(q) −σ(q + p)),
(4.92)
we prove that the origin (q, p) = (0, 0) is globally asymptotically stable for (4.92).
Deﬁne
h(q, p) = −f(p) −g(q, α(q) −σ(q + p)),
and
H(q) =
q
0
h(s, 0)ds
Noting that sh(s, 0) > 0, ∀s = 0, H(q) is positive semideﬁnite with H(0) = 0 and
H(q) > 0, ∀q = 0. Consider the following Lyapunov function candidate for (4.92)
V (q, p) = H(q) +
1
2
p
2
,
we have
˙
V = p(h(q, 0) −h(q, p)),
but
h
p
(q, p) = −f
(p) +σ
(q + p) g
u
(q, k(q, p)) > 0,
therefore,
˙
V ≤ 0, ∀q, p. But in
˙
V (q, p) = 0 or ¦(q, p) : p = 0¦ the only invariant set
is the origin (q, p) = (0, 0). From LaSalle’s invariance principle, it follows that the
origin is globally asymptotically stable for (4.92) and the result follows.
111
Chapter 5
Applications to Robotics and
Aerospace Vehicles
In this chapter, detailed control design procedures are provided for several challeng
ing examples of underactuated mechanical systems with kinetic symmetry. In each
case, distinctive comparisons are made between our approach and available control
methods (if any) that clarify the advantages of the reduction procedures introduced
in chapter 4.
All the examples presented in this section are physically meaningful systems. The
examples are chosen either based on the importance of their applications in robotics
and aerospace, or due to their challenging nature and signiﬁcance as benchmark
systems in nonlinear control theory. The systems of the former class are namely a
ﬂexiblelink robot, a multilink underactuated arm, the VTOL aircraft, and an au
tonomous helicopter. The latter category include the Acrobot, the InertiaWheel
Pendulum, the TORA system, the Pendubot, the CartPole system, the Rotating
Pendulum, and the BeamandBall system. The aforementioned examples of under
actuated systems include both highorder and loworder systems. In this section, the
examples appear from loworder systems to highorder systems.
Our control design approach for each example consists of three steps:
i) Reduction to a cascade nonlinear system with structural properties.
ii) Control of the loworder nonlinear part of the cascade system.
iii) Reconstruction of the controller for the overall system using the controller of
the nonlinear part obtained in step ii).
For nonlinear systems in strict feedback form, step iii) is known as the standard
backstepping procedure. The third step is a rather routine procedure and the main
diﬃculty in control design for an underactuated system is passing through steps i)
and ii). These three steps are precisely demonstrated for each example.
Some of our main results include global asymptotic stabilization of the Acrobot,
the CartPole system, the InertiaWheel Pendulum, a threelink planar pendulum
with actuated shape variables, and the VTOL aircraft. In addition, (almost) global
112
exponential output tracking/stabilization of feasible trajectories of an autonomous
helicopter is achieved. The closest available result to this work is global asymptotic
output tracking in [31].
Among the aforementioned nonlinear control problems, some of them are known to
be open in the literature. As an example, global stabilization of the Acrobot (Figure
51) using a smooth static state feedback in analytically explicit form has been an
open problem in nonlinear control for near a decade. One of our contributions is
to address this problem. More precisely, we present a nonlinear state feedback in
closedform that globally asymptotically stabilizes any arbitrary equilibrium point of
the Acrobot including the upright position. The main available result on asymptotic
stabilization of the Acrobot to the upright position can be found in [13, 86]. The
method in [13] consists of two stages: i) a slow swingup of the Acrobot using a
passivitybased approach that brings it close to the upright position, ii) switching
to a balancing linear controller that locally asymptotically stabilizes the pendulum
around its upright equilibrium point. The main drawback of this twostage method is
the poor performance of both swingup and balancing controllers in terms of time (i.e.
both controllers are slow). This is due to the fact that a gainscheduled controllers has
to keep the state of the system close to a set of equilibrium points. However, since the
vector ﬁeld of a closedloop system vanishes at the set of equilibrium points, the state
moves slowly close to this set of equilibrium points. This is the fundamental weakness
of the gain scheduling method. The nonlinear state feedback that we use does not
require a swingup stage and has a large region of attraction (i.e. R
4
), therefore the
nonlinear controller does not need to keep the state close to the set of equilibrium
points. As a result, from the simulation results, the speed of convergence to the
state to zero, is signiﬁcantly better than a switchingbased controller that uses a slow
swingup and balancing. Traditionally, a rather similar twostage switchingbased
approach has been applied to control of the Pendubot [92, 90], the CartPole system
[94, 28], and the Rotating Pendulum [112, 6]. From now on, we avoid repeating the
weaknesses of the passivitybased swingup methods and the gainscheduling approach
in nonlinear control, for each of these examples throughout this section.
Another contribution is that we introduce aggressive swingup control design
methods for benchmarktype underactuated systems that involve invertedpendulums.
Namely, the CartPole system, and the Rotating Pendulum. Formally, a swingup
problem is equivalent to stabilization of a system to a relative equilibrium point.
By aggressive swingup control, we mean (almost) exponential stabilization of an
invertedpendulum system to an equilibrium manifold corresponding to the upright
position of the pendulum.
5.1 The Acrobot
The Acrobot is a twolink revolute planar robot with one actuator at the elbow, as
shown in Figure 51. In this section, we provide a detailed control design procedure
for global asymptotic stabilization of the Acrobot to any of its arbitrary equilibrium
points, particularly, the upright equilibrium point. The inertia matrix for the Acrobot
113
q
q
l
1
L
1
m I
m I
2 2
1 1
1
2
x
y
Figure 51: The Acrobot
is given by (see section A.3)
m
11
(q
2
) = a + b cos(q
2
)
m
12
(q
2
) = c + (b/2) cos(q
2
)
m
22
(q
2
) = c
with
a = m
1
l
2
1
+ m
2
(L
2
1
+ l
2
2
) + I
1
+I
2
b = 2m
2
L
1
l
2
c = m
2
l
2
2
+ I
2
where q
i
, m
i
, L
i
, l
i
, and I
i
denote the angle, the mass, the length, the length of the
center of mass, and the inertia of the ith link, respectively. Clearly, q
2
is the shape
variable of the Acrobot. Since this shape variable is actuated the Acrobot is a ClassI
underactuated system. Also, the potential energy for the Acrobot is as the following
V (q
1
, q
2
) = (m
1
l
1
+ m
2
L
1
)g
0
cos(q
1
) + m
2
l
2
g
0
cos(q
1
+q
2
)
which depends on both q
1
, q
2
. Thus, the Acrobot does not possess symmetry in the
classical sense, but it possesses kinetic symmetry with respect to q
1
. This shows
that kinetic symmetry is a less restrictive property than classical symmetry as the
invariance of the Lagrangian under symmetry group actions.
Based on theorem 4.2.1 (or proposition 3.9.1), the following global change of
coordinates
q
r
= q
1
+ γ(q
2
)
p
r
= m
11
(q
2
)p
1
+ m
12
(q
2
)p
2
transforms the dynamics of the Acrobot to a cascade nonlinear system in strict feed
back form
˙ q
r
= p
r
/m
11
(q
2
)
˙ p
r
= g
r
(q
r
, q
2
)
˙ q
2
= p
2
˙ p
2
= u
(5.1)
114
where g
r
(q
r
, q
2
) = −∇
q
1
V (q
1
, q
2
)[
q
1
=qr−γ(q
2
)
is given by
g
r
(q
r
, q
2
) = (m
1
l
1
+ m
2
L
1
)g
0
sin(q
r
−γ(q
2
)) + m
2
l
2
g
0
sin(q
r
−γ(q
2
) + q
2
) (5.2)
and the function γ(q
2
) can be explicitly calculated as the following
γ(q
2
) =
q
2
0
m
12
(s)
m
11
(s)
ds =
q
2
2
+
(2c −a)
√
a
2
−b
2
arctan(
a −b
a + b
tan(
q
2
2
)) (5.3)
for q
2
∈ [−π, π]. The function γ(q
2
) is plotted in Figure 52. Observe that γ(q
2
) is
a nonlinear function for (2c − a) = 0. The special case of 2c − a = 0 that gives a
linear change of coordinates had been previously known. Stabilization of the Acrobot
−8 −6 −4 −2 0 2 4 6 8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rad
g
a
m
m
a
Figure 52: The function γ(q
2
) for a = 20/3, b = 4, c = 4/3.
reduces to stabilization of the following secondorder nonlinear system nonaﬃne in
control
˙ q
r
= m
−1
11
(v)p
r
˙ p
r
= g
r
(q
r
, v)
(5.4)
with control input v = q
2
. This stabilization problem is extremely complicated even
for a secondorder system due to the highly nonlinear way that the control v appears
in the dynamics of (5.4). Here is our main result for the Acrobot:
Proposition 5.1.1. Assume the parameters of the Acrobot satisfy either of the fol
lowing conditions
i) A < B.
ii) B < ρA, a −2c > 0, and c −b/2 ≥ 0.
115
where
A = m
1
l
1
+m
2
L
1
B = m
2
l
2
ρ =
ψ(π)
1 −ψ(π)
<
1
2
ψ(q
2
) =
m
12
(q
2
)
m
11
(q
2
)
Then, there exists a state feedback in the form
v = K
r
(q
r
, p
r
) = α(q
r
) −σ(q
r
+p
r
)
that globally asymptotically stabilizes the origin for the reduced nonlinear system of
the Acrobot in (5.4) where α(q
r
) is a smooth function that satisﬁes
g
r
(q
r
, α(q
r
)) = 0
and σ() is a scalar sigmoidal function. In addition, there exists a state feedback in
explicit form
u = K(q
r
, p
r
, q
s
, p
s
)
that globally asymptotically stabilizes the origin (q
r
, p
r
, q
s
, p
s
) = 0 for the composite
dynamics of the Acrobot in (5.1).
Proof. Based on theorems 4.7.2 and 4.7.3, if g
r
(q
r
, v) = 0 has an isolated root
v = α(q
r
) that satisﬁes
det
∂g
r
(q
r
, v)
∂v
v=α(qr)
= 0
then, there exists a static state feedback in the explicit form
v = K
r
(q
r
, p
r
) = α(q
r
) ±σ(q
r
+p
r
)
which globally asymptotically stabilizes any equilibrium point (¯ q
r
, 0) of the core
reduced system (5.4) (the sign ± in state feedback v = K
r
depends on whether
∇
v
g
r
(q
r
, α(q
r
)) < 0, or > 0, respectively). In order to ﬁnd the conditions under which
the state feedback K
r
exists, it suﬃces to determine a set of suﬃcient conditions that
guarantees the following equations have no solutions in v
g
r
(q
r
, v) = 0
∇
v
g
r
(q
r
, v) = 0
(5.5)
116
This set of nonlinear equations can be rewritten as
Asin(q
r
−γ(v)) + Bsin(q
r
−γ(v) + v) = 0
A
m
12
(v)
m
11
(v)
cos(q
r
−γ(v)) + B(
m
12
(v)
m
11
(v)
−1) cos(q
r
−γ(v) + v) = 0
(5.6)
where
A = (m
1
l
1
+ m
2
L
1
), B = m
2
l
2
Setting q
1
= q
r
− γ(v), from the property sin
2
(q
1
+ v) + cos
2
(q
1
+ v) = 1, it follows
that
Asin(q
1
)
B
2
+
Am
12
cos(q
1
)
B(m
11
−m
12
)
2
= 1
Keep in mind that a > b and c > b/2 imply m
11
(q
2
), m
12
(q
2
) > 0 for all q
2
. Solving
for sin
2
(q
1
) from the last equation, we obtain
sin
2
(q
1
) =
B
2
(m
11
−m
12
)
2
−A
2
m
2
12
A
2
(m
11
−m
12
)
2
−A
2
m
2
12
(5.7)
Thus, if A < B the right hand side of (5.7) is greater than 1 and the left hand side of
(5.7) is less than or equal to 1. This means the equations in (5.5) have no solutions
in v. On the other hand, (5.7) can be rewritten as
sin
2
(q
1
) =
[B(m
11
−m
12
) −Am
12
][B(m
11
−m
12
) + Am
12
]
A
2
m
11
(m
11
−2m
12
)
By assumption, m
12
(q
2
) ≥ c −b/2 ≥ 0. Assuming A ≥ B, we get
B(m
11
−m
12
) + Am
12
= (A −B)m
12
+ Bm
11
> 0
Also, m
11
−2m
12
= a −2c > 0. Thus, by taking
B(m
11
−m
12
) −Am
12
< 0
the right hand side of the last equation is negative, while the left hand side is positive.
In other words, the equations in (5.5) have no real solutions in v. The last condition
can be expressed as
B
A + B
<
m
12
(v)
m
11
(v)
=: ψ(v) (5.8)
It is elementary to show that ψ(v) takes its minimum at v = ±π which is equal to
ψ(π) =
c −b/2
a −b
Note that ψ(q
2
) ≤ m
12
/(a−2c+2m
12
) < 1/2 for all q
2
. Therefore, if B < ψ(π)(A+B),
117
or
B <
ψ(π)
1 −ψ(π)
A
where ρ :=
ψ(π)
1−ψ(π)
< 1, equation (5.7) has no real solution in q
1
(and thus in v). This
ﬁnishes the proof of the ﬁrst part. The existence and closedform expression for the
state feedback
u = K(q
r
, p
r
, q
s
, p
s
)
follows from the backstepping procedure.
Remark 5.1.1. Assume both links of the Acrobot have the same uniform mass density.
If m
1
= m
2
and l
2
> 3l
1
, then A < B and condition i) in proposition 5.1.1 is satisﬁed.
Also, if l
2
> 1.5l
1
, then c − b/2 > 0. Let us take m
1
= λm
2
and l
2
= 2l
1
. Then,
a − 2c > 0, and ψ(π) = 5/2(λ + 1). This gives ρ = 5/(2λ − 3) and A, B can be
calculated as the following
A = m
1
l
1
+ m
2
L
1
= (λ + 2)m
2
l
1
, B = 2m
2
l
1
Therefore, the condition B < ρ(λ)A is satisﬁed for all λ > λ
∗
= 1.5 and condition ii)
of proposition 5.1.1 holds.
Remark 5.1.2. To ﬁnd the isolated root v = α(q
r
) of g
r
(q
r
, v) = 0, let us diﬀerentiate
the last equation with respect to parameter s ∈ R. We get
∇
qr
g
r
(q
r
, v)
dq
r
ds
+∇
v
g
r
(q
r
, v)
dv
ds
= 0
or
dq
r
ds
= 1
dv
ds
= −
∇
qr
g
r
(q
r
, v)
∇
v
g
r
(q
r
, v)
The solution of this ODE with zero initial conditions both positive and negative in
time gives a parameterization (q
r
, α(q
r
)) of the isolated root of g
r
(q
r
, v) = 0.
For the Acrobot, it is possible to solve the equation g
r
(q
r
, v) = 0 in q
r
explicitly as
a function of v. This gives a numeric look up table of pairs (q
r
, v) as the solution of
equation g
r
(q
r
, v) = 0. To show this, notice that g
r
(q
r
, v) = 0 is equivalent to
Asin(q
1
) + Bsin(q
1
+ v) = 0
or
(A + Bcos(v)) sin(q
1
) +Bsin(v) cos(q
1
) = 0
Hence, the last equation can be solved in q
1
as
q
1
= −arctan
Bsin(v)
A + Bcos(v)
118
−4 −3 −2 −1 0 1 2 3 4
−40
−30
−20
−10
0
10
20
30
40
q
r
(rad)
α
(
q
r
)
(
r
a
d
)
Figure 53: The function α(q
r
) for m
1
= m
2
= 1 and l
1
= l
2
= 1.
but q
1
= q
r
−γ(v), thus
q
r
= γ(v) −arctan
Bsin(v)
A + Bcos(v)
=: φ(v)
In other words, solving the equation g
r
(q
r
, v) = 0 in v is equivalent to inverting the
function φ(v) which is available in closedform. This inversion can be numerically done
using a piecewise linear approximation, splines, neural networks, or other convenient
curve ﬁtting approaches. In this work, we use a piecewise linear approximation
method constructed from a look up table of pairs (φ(v), v) as the solution of equation
g
r
(q
r
, v) = 0. Figures 54 through 59 show simulation results for the Acrobot for
the choice of parameters m
1
= m
2
= 1, l
1
= l
2
= 1, and I
1
= I
2
= 1/3 (the mass
densities of both links are assumed to be uniform and equal). The simulation results
demonstrate eﬀective stabilization of the upright position for the Acrobot from initial
conditions with large deviations of q
1
, q
2
from zero.
119
0 1 2 3 4
−0.5
0
0.5
1
1.5
time (sec)
q
1
(
r
a
d
)
0 20 40 60 80 100
−0.4
−0.2
0
0.2
0.4
time (sec)
q
1
(
r
a
d
)
0 1 2 3 4
−8
−6
−4
−2
0
2
4
time (sec)
q
2
(
r
a
d
)
0 20 40 60 80 100
−8
−6
−4
−2
0
time (sec)
q
2
(
r
a
d
)
Figure 54: Trajectory of (q
1
, q
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π/3, 0, 0) and control input p
2
.
0 1 2 3 4
−15
−10
−5
0
5
time (sec)
p
1
(
r
a
d
/
s
e
c
)
0 20 40 60 80 100
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
time (sec)
p
1
(
r
a
d
)
0 1 2 3 4
−10
0
10
20
30
40
50
time (sec)
p
2
(
r
a
d
)
0 20 40 60 80 100
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
time (sec)
p
2
(
r
a
d
/
s
e
c
)
Figure 55: Trajectory of (p
1
, p
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π/3, 0, 0) and control input p
2
.
120
0 1 2 3 4 5
−1
0
1
2
3
4
time (sec)
q
1
(
r
a
d
)
0 20 40 60 80 100
−0.6
−0.4
−0.2
0
0.2
0.4
time (sec)
q
1
(
r
a
d
)
0 1 2 3 4 5
−15
−10
−5
0
5
10
15
20
time (sec)
q
2
(
r
a
d
)
0 20 40 60 80 100
−12
−10
−8
−6
−4
−2
0
time (sec)
q
2
(
r
a
d
)
Figure 56: Trajectory of (q
1
, q
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π, 0, 0) and control input p
2
.
0 1 2 3 4 5
−60
−40
−20
0
20
40
60
time (sec)
p
1
(
r
a
d
/
s
e
c
)
0 20 40 60 80 100
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
time (sec)
p
1
(
r
a
d
)
0 1 2 3 4 5
−50
0
50
100
150
200
time (sec)
p
2
(
r
a
d
)
0 20 40 60 80 100
−0.5
0
0.5
1
1.5
2
time (sec)
p
2
(
r
a
d
/
s
e
c
)
Figure 57: Trajectory of (p
1
, p
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(π, 0, 0) and control input p
2
.
121
0 0.05 0.1 0.15 0.2
−0.03
−0.025
−0.02
−0.015
−0.01
−0.005
0
time (sec)
q
1
(
r
a
d
)
0 20 40 60 80 100
−0.35
−0.3
−0.25
−0.2
−0.15
−0.1
−0.05
0
time (sec)
q
1
(
r
a
d
)
0 0.05 0.1 0.15 0.2
3
3.05
3.1
3.15
3.2
time (sec)
q
2
(
r
a
d
)
0 20 40 60 80 100
0
0.5
1
1.5
2
2.5
3
3.5
time (sec)
q
2
(
r
a
d
)
Figure 58: Trajectory of (q
1
, q
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(0, π, 0) and control input p
2
.
0 0.05 0.1 0.15 0.2
−0.3
−0.25
−0.2
−0.15
−0.1
−0.05
time (sec)
p
1
(
r
a
d
/
s
e
c
)
0 20 40 60 80 100
−0.2
−0.15
−0.1
−0.05
0
0.05
time (sec)
p
1
(
r
a
d
)
0 0.05 0.1 0.15 0.2
−1.2
−1
−0.8
−0.6
−0.4
−0.2
time (sec)
p
2
(
r
a
d
)
0 20 40 60 80 100
−0.4
−0.3
−0.2
−0.1
0
time (sec)
p
2
(
r
a
d
/
s
e
c
)
Figure 59: Trajectory of (p
1
, p
2
) for the Acrobot with initial condition (q
1
, q
2
, p
1
) =
(0, π, 0) and control input p
2
.
122
5.2 The TORA System
The TORA
1
system depicted in Figure 510 consists of a translational oscillating plat
form which is controlled via a rotational eccentric mass. The TORA system has been
m
m
r
q
k
1
2
2
τ
I
1
2
q
1
Figure 510: The TORA system.
mainly a benchmark example for passivitybased approaches in control of nonlinear
systems and stabilization methods for nonlinear cascade systems [109, 37, 80, 38, 39].
Global stabilization of the TORA system using state feedback and backstepping pro
cedure has been known due to Wan et al. [109]. More recently, global output tracking
for the TORA system was addressed in [39]. Here, we provide a new globally asymp
totically stabilizing state feedback for the TORA system with a physically meaningful
Lyapunov function. The advantage of using a Lagrangianbased change of coordinates
is that the reduced system (i.e. zerodynamics) has a meaningful Lagrangian struc
ture which simpliﬁes the control design. It turns out that for the TORA system, the
reduced system is an undamped onedimensional massspring system.
The Lagrangian of the TORA system is as the following
L(q, ˙ q) =
1
2
m
1
˙ q
2
1
+
1
2
m
2
[( ˙ q
1
+ r cos(q
2
) ˙ q
2
)
2
+ (r sin(q
2
) ˙ q
2
)
2
]
+
1
2
(m
2
r
2
+ I
2
) ˙ q
2
2
−V (q
1
, q
2
)
(5.9)
where the potential energy is equal to
V (q
1
, q
2
) =
1
2
k
1
q
2
1
+ m
2
gr cos(q
2
)
The Lagrangian of the TORA system can be rewritten as
L(q, ˙ q) =
1
2
¸
˙ q
1
˙ q
2
T
¸
m
1
+ m
2
m
2
r cos(q
2
)
m
2
r cos(q
2
) m
2
r
2
+ I
2
¸
˙ q
1
˙ q
2
−V (q
1
, q
2
) (5.10)
From the equation of the Lagrangian in (5.10), it is clear that q
2
is the shape variable
1
Translational Oscillator with Rotational Actuator
123
and q
1
is the external variable of the TORA system. Since q
2
is actuated, the TORA
system is a ClassI underactuated system that can be globally transformed into a
cascade system in strict feedback form. The EulerLagrange equations of motion for
the TORA system are given by
(m
1
+m
2
)¨ q
1
+ m
2
r cos(q
2
)¨ q
2
−m
2
r sin(q
2
) ˙ q
2
2
+k
1
q
1
= 0
m
2
r cos(q
2
)¨ q
1
+ (m
2
r
2
+ I
2
)¨ q
2
+m
2
gr sin(q
2
) = τ
(5.11)
After collocated partial feedback linearization using a change of control
τ = α(q
2
)u + β(q, ˙ q)
with
α(q
2
) = (m
2
r
2
2
+ I
2
) −
(m
2
r cos(q
2
))
2
(m
1
+ m
2
)
> 0 , ∀q
2
∈ [−π, π]
the dynamics of the shape variable q
2
reduces to
˙ q
2
= p
2
, ˙ p
2
= u
The normalized momentum conjugate to q
1
is
p
r
=
∂L
∂ ˙ q
1
= (m
1
+ m
2
) ˙ q
1
+m
2
r cos(q
2
) ˙ q
2
Observe that both p
r
and the normalized momentum π
r
= m
−1
11
p
r
are trivially inte
grable. The function γ(q
2
) can be calculated explicitly as
γ(q
2
) =
q
2
0
m
12
(s)
m
11
ds =
m
2
r sin(q
2
)
m
1
+ m
2
Thus, based on theorem 4.2.1, the global change of coordinates
q
r
= q
1
+ (m
2
r sin(q
2
))/(m
1
+ m
2
)
p
r
= (m
1
+ m
2
)p
1
+ m
2
r cos(q
2
)p
2
(5.12)
transforms the dynamics of the TORA system into a cascade nonlinear system in
strict feedback form
˙ q
r
= (m
1
+ m
2
)
−1
p
r
˙ p
r
= −k
1
q
r
+ k
1
γ(q
2
)
˙ q
2
= p
2
˙ p
2
= u
(5.13)
The reduced system with Lagrangian
L
r
(q
r
, ˙ q
r
) =
1
2
(m
1
+ m
2
) ˙ q
2
r
−
1
2
k
1
q
2
r
124
satisﬁes the onedimensional forced EulerLagrange equation
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= k
1
γ(q
2
)
with a nonlinear input force F
r
(q
2
) = k
1
γ(q
2
) which depends on the shape variable q
2
.
If q
2
≡ 0, the reduced system is an oscillator with frequency ω
0
=
k
1
/(m
1
+ m
2
)
(rad/sec). Using backstepping procedure, stabilization of the overall system reduces
to the stabilization of the reduced undamped massspring system with reduced La
grangian L
r
(q
r
, ˙ q
r
).
Proposition 5.2.1. Consider the TORA system in cascade form (5.13). Then, the
following statements hold:
i) There exists a bounded state feedback
q
2
= K
r
(q
r
, p
r
) = −σ(c
1
q
r
+ c
2
p
r
) , c
1
, c
2
> 0 (5.14)
that globally asymptotically stabilizes the origin (q
r
, p
r
) = 0 for reduced system
of the TORA system in (5.13) where σ denotes a sigmoidal function satisfying
[σ()[ ≤ π/2.
ii) The reduced system has a welldeﬁned new Hamiltonian function
˜
H
r
(q
r
, p
r
) that
is decreasing by time, i.e.
˙
˜
H
r
≤ 0. In addition,
˜
H
r
(q
r
, p
r
) is a valid Lyapunov
function for the reduced system. In other words, the closedloop reduced system
with controller K
r
(q
r
, p
r
) is a passive Hamiltonian system.
iii) There exists a state feedback in explicit form
u = K(q
r
, p
r
, q
2
, p
2
)
that globally asymptotically stabilizes the origin for the composite system in
(5.13).
Proof. We prove parts i) and ii) simultaneously and then part iii) follows from
the ﬁrst two parts. First, observe that if σ(x) is a sigmoidal function satisfying
[σ(x)[ ≤ π/2 for all x, then sin(σ(x)) and thus γ(σ(s)) is a sigmoidal function as well.
Deﬁne the amended potential function
ψ(q
r
) =
qr
0
k
1
γ(σ(c
1
s))ds =
qr
0
k
1
m
2
r sin(σ(c
1
s))
m
1
+m
2
ds
and notice that ψ(q
r
) is positive deﬁnite function. Now, consider the following new
Hamiltonian
˜
H
r
as a candidate Lyapunov function
˜
H
r
(q
r
, p
r
) = H
r
(q
r
, p
r
) + ψ(q
r
)
125
where H
r
is the (original) Hamiltonian of the reduced system given by
H
r
(q
r
, p
r
) =
1
2
(m
1
+ m
2
)
−1
p
2
r
+
1
2
k
1
q
2
r
Since the input force is not identically zero, the Hamiltonian H
r
is not a conserved
quantity. Calculating
˙
˜
H
r
along the solutions of the closedloop reduced system with
controller K
r
(q
r
, p
r
) gives
˙
˜
H
r
=
˙
H
r
+ ˙ q
r
ψ
(q
r
)
= (m
1
+ m
2
)p
r
[−γ(σ(c
1
q
r
+ c
2
p
r
)) + γ(σ(c
1
q
r
))]
= (m
1
+ m
2
)p
r
[¯ σ(c
1
q
r
) − ¯ σ(c
1
q
r
+ c
2
p
r
)] ≤ 0
where ¯ σ(x) = γ(σ(x)) is a sigmoidal function. From the last equation,
˙
˜
H
r
< 0
for all p
r
= 0 and based on LaSalle’s invariance principle, all the solutions of the
closedloop system converge to the largest invariant set Ω with p
r
= 0. In this case,
Ω = ¦(0, 0)¦ and the origin (q
r
, p
r
) = (0, 0) is globally asymptotically stable for the
reduced system. Apparently,
˜
H(q
r
, p
r
) is a valid Hamiltonian for the closedloop
system with conﬁguration q
r
, a new potential energy
˜
V (q
r
) =
1
2
k
1
q
2
r
+ ψ(q
r
)
and a new nonlinear damping force
F
d
(q
r
, p
r
) = γ(σ(c
1
q
r
)) −γ(σ(c
1
q
r
+ c
2
p
r
))
which satisﬁes p
r
F
d
(q
r
, p
r
) ≤ 0.
Remark 5.2.1. The controller obtained in [109] is equivalent to using the following
state feedback
q
2
= K
r
(p
r
) = −σ(c
2
p
r
)
which is a globally asymptotically stabilizing feedback for the reduced system. This
feedback is a special case of the state feedback in part i) of proposition 5.2.1 with
c
1
= 0. In this special case, the amended potential ψ(q
r
) is identically zero.
Figure 511 shows the simulation results for the TORA system with the choice of
parameters m
1
= 10, m
2
= 1, k
1
= 5, r = 1 and I = 1. The controller K
r
is in the
form
K
r
(q
r
, p
r
) = −a tanh(c
1
q
r
+c
2
p
r
)
with a = π/2 and c
1
= c
2
= 1. Figure 511 (c), clearly shows that the reduced system
demonstrates a solution close to the trajectories of position and velocity corresponding
to a damped massspring system with a linear damping force.
126
0 10 20 30 40 50 60
−1.5
−1
−0.5
0
0.5
1
time (sec)
q
1
,
p
1
q
1
p
1
0 10 20 30 40 50 60
−10
−8
−6
−4
−2
0
2
4
6
8
time (sec)
q
2
,
p
2
q
2
p
2
(a) (b)
0 10 20 30 40 50 60
−8
−6
−4
−2
0
2
4
6
8
time (sec)
q
r
,
p
r
q
r
p
r
0 10 20 30 40 50 60
−3
−2
−1
0
1
2
3
4
time (sec)
u
(
c
o
n
t
r
o
l)
(c) (d)
Figure 511: State trajectories and control input of the TORA system for the initial
condition (q
1
, q
2
, p
1
, p
2
) = (1, 0, 0, 0): (a) (q
1
, p
1
), (b) (q
2
, p
2
), (c) (q
r
, p
r
), and (d) u.
5.3 The InertiaWheel Pendulum
The InertiaWheel Pendulum is a planar inverted pendulum with a revolving wheel
at the end, as shown in Figure 512. The wheel is actuated and the joint of the pen
dulum at the base is unactuated. The inertialwheel pendulum was ﬁrst introduced
by Spong et al. in [93] where a supervisory hybrid/switching control strategy is ap
plied to asymptotic stabilization of the inertiawheel pendulum around its upright
equilibrium point. First, a passivitybased controller [28] swings up the pendulum.
Then, a balancing controller that is obtained by Jacobian linearization or (local) exact
feedback linearization stabilizes the pendulum around its upright position.
Here, we show that the inertiawheel pendulum is an example of a ClassI underac
tuated system. Based on theorem 4.2.1, the dynamics of the inertia wheel pendulum
can be transformed into a cascade nonlinear system in strict feedback form; using a
global change of coordinates in an explicit form. Then, global asymptotic stabiliza
tion of the upright equilibrium point can be achieved using the standard backstepping
procedure [36, 57, 80].
127
q
q
I
1
2
2
1
L , I
1
τ
Figure 512: The InertiaWheel Pendulum
The Lagrangian of the inertia wheel pendulum is as the following
L(q, ˙ q) =
1
2
˙ q
T
M ˙ q −V (q
1
)
where q = (q
1
, q
2
)
T
∈ R
2
denotes the conﬁguration vector of the IWP and M is a
constant inertia matrix with elements
m
11
= m
1
l
2
1
+ m
2
L
2
1
+I
1
+ I
2
m
12
= m
21
= m
22
= I
2
The potential energy of the IWP [93] is given by
V (q
1
) = (m
1
l
1
+ m
2
L
1
)g
0
cos(q
1
) =: m
0
cos(q
1
)
The EulerLagrange equations of motion for the inertiawheel pendulum are as the
following
d
dt
(
∂L
∂ ˙ q
1
) −
∂L
∂q
1
= 0
d
dt
(
∂L
∂ ˙ q
2
) −
∂L
∂q
2
= τ
or
m
11
¨ q
1
+ m
12
¨ q
2
= g
1
(q
1
)
m
21
¨ q
1
+ m
22
¨ q
2
= τ
(5.15)
where g
1
(q
1
) = m
0
sin(q
1
).
Due to the invariance of the kinetic energy of the inertiawheel pendulum under the
group action (q
1
, q
2
) →(q
1
+α, q
2
+β) for (α, β) ∈ R
2
, the inertia matrix is constant
and all the Christoﬀel Symbols associated with M vanish. Thus, the inertiawheel
pendulum is a ﬂat underactuated mechanical system with two degrees of freedom
and kinetic symmetry w.r.t. both degrees of freedom. As a consequence, the inertia
128
wheel pendulum is a special case of underactuated systems with symmetry properties
considered in [68] and can be globally asymptotically stabilized.
Based on theorem 4.2.1, the global change of coordinates
z
1
= ∂L/∂ ˙ q
1
= m
11
˙ q
1
+ m
12
˙ q
2
z
2
= q
1
z
3
= ˙ q
2
transforms the dynamics of the inertiawheel pendulum into a strict feedback system
˙ z
1
= m
0
sin(z
2
)
˙ z
2
= (z
1
−m
12
z
3
)/m
11
˙ z
3
= u
(5.16)
where the new control u and the torque τ are related through the expression
τ = (m
22
−m
21
m
12
/m
11
)u + (m
21
m
0
/m
11
) sin(q
1
)
Note that q
2
does not play any important role in the dynamics of the IWP and is
ignored as an state variable (also, q
2
does not appear in L(q, ˙ q)).
Remark 5.3.1. Clearly, the nonlinear system in (5.16) is exact feedback linearizable
with the choice of the output y = z
1
over [z
2
[ < π/2 (i.e. [q
1
[ < π/2). This result has
been previously obtained in [93] using a standard method from [36].
Here is our main result for the inertia wheel pendulum:
Proposition 5.3.1. There exists a nonlinear state feedback law (in explicit form) that
globally asymptotically stabilizes (z
1
, z
2
, z
3
) = 0 (i.e. the upright equilibrium point)
for the Inertia Wheel Pendulum in (5.16).
Proof. Consider the scalar nonlinear system
˙ z
1
= m
0
sin(z
2
)
with control input z
2
. The following state feedback
z
2
= k
1
(z
1
) := c
0
σ
1
(c
1
z
1
) ; −π/2 < c
0
< 0, c
1
> 0
(where σ
1
(s) = tanh(s) is a sigmoidal function) globally asymptotically stabilizes
z
1
= 0 and V (z
1
) = z
2
1
is a valid Lyapunov function for the z
1
subsystem in (5.16).
This is due to the fact that
σ(z
1
) = −m
0
sin(c
0
σ
1
(c
1
z
1
))
is a sigmoidal function that satisﬁes σ(0) = 0, σ
(0) > 0, and sσ(s) > 0, ∀s = 0. Now,
deﬁne the following change of coordinates and control
µ
1
= z
2
−k
1
(z
1
), µ
2
= ˙ µ
1
, v = ˙ µ
2
129
which can be explicitly expressed as
µ
1
= z
2
−k
1
(z
1
)
µ
2
= (z
1
−m
12
z
3
)/m
11
−
˙
k
1
v = (m
0
/m
11
) sin(z
2
) −(m
12
/m
11
)u −
¨
k
1
(5.17)
where by the assumption that σ
1
(s) = tanh(s), we have
˙
k
1
= c
0
c
1
m
0
sin(z
2
)(1 −σ
1
(c
1
z
1
)
2
)
¨
k
1
= c
0
c
1
m
0
(1 −σ
1
(c
1
z
1
)
2
)(cos(z
2
)(z
1
/m
11
−m
12
z
3
/m
11
) −2c
1
m
0
σ
1
(c
1
z
1
) sin(z
2
)
2
)
In new coordinates, the dynamics of the inertia wheel pendulum in (5.16) can be
written in the following cascade form
˙ z
1
= m
0
sin(k
1
(z
1
) + µ
1
)
˙ µ
1
= µ
2
˙ µ
2
= v
(5.18)
The state feedback
v = −c
2
µ
1
−c
3
µ
2
; c
2
, c
3
> 0 (5.19)
globally exponentially stabilizes (µ
1
, µ
2
) = 0 for the µsubsystem of (5.18). Due to
the fact that for any exponentially vanishing function µ
1
(t), the solution of the z
1

subsystem in (5.18) is uniformly bounded (the proof for this boundedness is rather
elementary and is omitted), the origin (z
1
, µ
1
, µ
2
) = 0 for the closed loop cascade
system in (5.18) is globally asymptotically and locally exponentially stable due to a
theorem by Sontag [83]. The overall nonlinear state feedback u = k(z
1
, z
2
, z
3
) can be
calculated as the following
u =
m
11
m
12
(c
2
µ
1
+ c
3
µ
2
+
m
0
m
11
sin(z
2
) −
¨
k
1
)
which is an explicit function of the state (z
1
, z
2
, z
3
).
Remark 5.3.2. Apparently, using standard backstepping procedure in [36] (chapter
9), a Lyapunov function V (z) and a globally stabilizing feedback law for (5.16) can
be obtained by explicit use of V
1
(z
1
) = z
2
1
and propagation of the time derivative of
V
1
along the solutions of (5.16). However, this only complicates the calculation of the
ﬁnal stabilizing state feedback law and there is no need to do so in this special case.
Now, we present simulation results for global stabilization of the inertiawheel
pendulum to its upright equilibrium point. For comparison purposes, we used the
same numeric values for the parameters of the dynamics of the IWP as in [93]. These
parameters are as follows: m
11
= 4.83 10
−3
, m
12
= m
21
= m
22
= 32 10
−6
,
¯ m = 38.7 10
−3
, g
0
= 9.8, and m
0
= ¯ mg
0
.
The state trajectories and the control input of the inertiawheel pendulum for
the downward and vertical initial positions of the pendulum are shown in Figures
513 and 514, respectively. These simulation results demonstrate that the nonlinear
controller aggressively stabilizes the pendulum to its upright position. The values of
130
the controller parameters for the simulation are c
0
= −π/10, c
1
= .03, c
2
= 16, c
3
= 8
and the maximal applied torque was τ
max
= 0.6 Nm.
0 1 2 3 4 5 6
−8
−6
−4
−2
0
2
4
sec
r
a
d
, r
a
d
/s
e
c
q
1
dq
1
/ dt
−0.5 0 0.5 1 1.5 2 2.5 3 3.5
−7
−6
−5
−4
−3
−2
−1
0
1
q
1
d
q
1
/ d
t
(a) (b)
0 1 2 3 4 5 6
0
1000
2000
3000
4000
5000
6000
sec
d
q
2
/ d
t (
r
a
d
/s
e
c
)
0 1 2 3 4 5 6
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
sec
ta
u
(
N
m
)
(
c
o
n
tr
o
l)
(c) (d)
Figure 513: For the initial condition (π, 0, 0), (a) shows the state trajectories for
(q
1
, ˙ q
1
), (b) shows the state trajectory in (q
1
, ˙ q
1
) plane, (c) shows ˙ q
2
, and (d) shows
the control input τ.
131
0 1 2 3 4 5 6
−3
−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
sec
r
a
d
, r
a
d
/s
e
c
q
1
dq
1
/ dt
−0.4 −0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
−3
−2.5
−2
−1.5
−1
−0.5
0
0.5
q
1
d
q
1
/ d
t
(a) (b)
0 1 2 3 4 5 6
0
1000
2000
3000
4000
5000
6000
7000
sec
d
q
2
/ d
t (
r
a
d
/s
e
c
)
0 1 2 3 4 5 6
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
sec
ta
u
(
N
m
)
(
c
o
n
tr
o
l)
(c) (d)
Figure 514: For the initial condition (π/2, 0, 0), (a) shows the state trajectories for
(q
1
, ˙ q
1
), (b) shows the state trajectory in (q
1
, ˙ q
1
) plane, (c) shows ˙ q
2
, and (d) shows
the control input τ.
5.4 The CartPole System
The CartPole system consists of a cart and an invertedpendulum on it, as shown
in Figure 515. A horizontal force F as the control input is applied along q
1
and the
joint q
2
is unactuated. The inertia matrix of the cartpole system is given by (see
section A.5)
m
11
= m
1
+ m
2
m
12
(q
2
) = m
21
(q
2
) = m
2
l
2
cos q
2
m
22
= m
2
l
2
2
+ I
2
where m
1
is the mass of the cart and m
2
, l
2
, I
2
are the mass, length of the center of
mass, and the inertia of the pendulum, respectively. Also, the potential energy of the
cartpole system is in the form
V (q
2
) = m
2
gl
2
cos(q
2
)
132
m
F
m , l , I
1
2 2 2
q
q
1
2
Figure 515: The CartPole system.
The EulerLagrange equations of motion for the cartpole system can be expressed as
the following
(m
1
+ m
2
)¨ q
1
+ m
2
l
2
cos(q
2
)¨ q
2
−m
2
l
2
sin(q
2
) ˙ q
2
2
= F
m
2
l
2
cos(q
2
)¨ q
1
+ (m
2
l
2
2
+ I
2
)¨ q
2
−m
2
gl
2
sin(q
2
) = 0
Apparently, the inertia matrix only depends on q
2
and thus q
1
is the external variable
and q
2
is the shape variable of the cartpole system. Since the shape variable q
2
is
unactuated, the cartpole system falls within the category of Class–II underactuated
systems.
5.4.1 Global Stabilization to an Equilibrium Point
In this section, we introduce a state feedback that globally asymptotically stabilizes
the upright equilibrium point of the cartpole system in the upper halfplane. First,
we need to perform noncollocated feedback linearization on the dynamics of the cart
pole system. Notice that
(m
12
(q
2
) −
m
11
m
22
m
21
(q
2
)
)¨ q
2
+ (m
1
+ m
2
)g tan(q
2
) −m
2
l
2
sin(q
2
) ˙ q
2
2
= F
and the coeﬃcient of ¨ q
2
is negative thus applying the feedback law
F = (m
12
(q
2
) −
m
11
m
22
m
21
(q
2
)
)u + (m
1
+ m
2
)g tan(q
2
) −m
2
l
2
sin(q
2
) ˙ q
2
2
(5.20)
reduces the dynamics of q
2
to a double integrator as
˙ q
2
= p
2
, ˙ p
2
= u
where u is the new control. Based on theorem 4.2.2, the following change of coordi
nates
q
r
= q
1
+ γ(q
2
)
p
r
= m
21
(q
2
)p
1
+ m
22
p
2
133
where
γ(q
2
) =
q
2
0
m
22
m
21
(θ)
dθ =
(m
2
l
2
2
+I
2
)
m
2
l
2
log
1 + tan(q
2
/2)
1 −tan(q
2
/2)
, q
2
∈ (−π/2, π/2)
transforms the dynamics of the cartpole system into
˙ q
r
=
1
m
21
(q
s
)
p
r
˙ p
r
= g
r
(q
s
) +
1
m
21
(q
s
)
dm
21
(q
s
)
dq
s
p
r
p
s
−
m
22
m
21
(q
s
)
dm
21
(q
s
)
dq
s
p
2
s
˙ q
s
= p
s
˙ p
s
= u
where q
s
= q
2
and g
r
(q
s
) = m
2
gl
2
sin(q
s
). Clearly, the equation of ˙ p
r
is a quadratic
form is (p
r
, p
s
) and the reason that the coeﬃcient of p
2
r
is identically zero is that m
11
is
constant for the cartpole system. This property guarantees that the cartpole system
is a nonlinear system in strict feedforward form (see [102]). The (strict) feedforward
property of the cartpole system has been used for semiglobal stabilization of the
system above the halfplane by Teel [101] as an application of nonlinear smallgain
theorem. In [56], Mazenc and Praly constructed a Lyapunov function for the cartpole
system (after extremely lengthy and complicated algebraic calculations) that proves
existence of a globally stabilizing state feedback law for the cartpole system above
the halfplane. Here is our main result for the cartpole system:
Proposition 5.4.1. There exists an state feedback law in the form of nested sat
urations that globally asymptotically stabilizes the origin (q
1
, p
1
, q
2
, p
2
) = 0 for the
cartpole system over the upper halfplane.
Proof. The proof relays on the aforementioned change of coordinates and further
simpliﬁcation of the reduced system and a famous result due to Teel in [101]. The
equation of ˙ p
r
can be simpliﬁed as
˙ p
r
= −tan(q
s
)p
r
p
s
+m
22
tan(q
s
)p
2
s
+ m
2
gl
2
sin(q
s
)
Applying the following change of coordinates
z
1
= q
r
z
2
= p
r
/m
21
(q
s
)
we get
˙ z
1
= z
2
˙ z
2
= tan(q
s
)(g +
m
2
l
2
2
+ I
2
m
2
l
2
cos(q
s
)
p
2
s
)
To map the upper halfplane to R, we use another change of coordinates and control
134
as the following
ξ
1
= tan(q
s
)
ξ
2
= (1 + tan(q
s
)
2
)p
s
v = (1 + tan(q
s
)
2
)u + 2 tan(q
s
)(1 + tan(q
s
)
2
)p
s
This transforms the dynamics of the cartpole system into
˙ z
1
= z
2
˙ z
2
= ξ
1
(k
1
+k
2
ξ
2
2
(1 + ξ
2
1
)
3
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(5.21)
where k
1
, k
2
are the following positive constants
k
1
= g, k
2
=
m
2
l
2
2
+ I
2
m
2
l
2
Observing that [ξ
1
/(1 + ξ
2
1
)
3
2
[ < 1, the nonlinear system in (5.21) is in strict feedfor
ward form and satisﬁes all the conditions of Teel’s theorem in [102]. Therefore, the
origin (q, p) = 0 for the cartpole system can be globally asymptotically stabilized
using small nested saturations in explicit form. The control law can be obtained as
the following. First, based on [102], apply the change of coordinates and control
η
1
=
1
k
1
z
1
+
2
k
1
z
2
+ 2ξ
1
+ ξ
2
η
2
=
1
k
1
z
2
+ξ
1
+ξ
2
w = ξ
1
+ξ
2
+ v
Then, the state feedback law
w = −σ
2
(η
2
+σ
1
(η
1
))
or
v = −ξ
1
−ξ
2
−σ
2
(η
2
+ σ
1
(η
1
))
globally asymptotically stabilizes the origin for the cartpole system where σ
i
’s are
saturation functions with appropriate small thresholds and magnitudes.
Figure 516 shows the simulation results for the cartpole system for the ini
tial state (q
1
, p
1
, q
2
, p
2
) = (5, 0, π/3, 0). Clearly, the controller with σ
i
(x) = sat
1
(x)
stabilizes the inverted pendulum in its upright position at q
1
= 0 after a rather
short time. The parameters are chosen as m
1
= m
2
= 1, l
2
= 0.75, g = 9.8 and
sat
(x) = sgn(x) min¦1, [x/[¦. However, based on Teel’s theorem the satu
ration functions must have suﬃciently small thresholds and magnitudes. To see
how the controller designed based on small nested saturations performs, we used
135
σ
i
= sat
0.1
(), i = 1, 2. The thresholds and magnitudes of saturation functions are
reduced by a factor of ten compared to the previous case, but the same initial con
dition (5, 0, π/3, 0) is chosen. The results are shown in Figure 517. Clearly, the
controller has a poor performance in terms of the settling time and maximum peak
of the position. The settling time increases by an approximate factor of 10 (which
is expected) and the maximum peak of the position has increases by a factor of 4
(which is better than expected). In addition, the convergence rate of the position is
almost linear. This reveals a large deterioration of the performance of the controller
for nested saturations with relatively small magnitudes.
0 5 10 15 20 25
−20
−10
0
10
20
30
40
50
time (sec)
p
o
s
/
v
e
lo
c
it
y
q
1
p
1
0 5 10 15 20 25
−1.5
−1
−0.5
0
0.5
1
1.5
time (sec)
a
n
g
le
/
a
n
g
u
la
r
v
e
lo
c
it
y
q
2
p
2
0 5 10 15 20 25
−40
−20
0
20
40
60
time (sec)
F
o
r
c
e
0 5 10 15 20 25
−6
−4
−2
0
2
4
time (sec)
C
o
n
t
r
o
l
u
(a) (b)
Figure 516: (a) State trajectory of the cartpole system for the initial condition
(5, 0, π/3, 0) and σ
i
() = sat
1
(), (b) The force F and control input u.
0 50 100 150 200 250
−50
0
50
100
150
200
time (sec)
p
o
s
/
v
e
lo
c
it
y
q
1
p
1
0 50 100 150 200 250
−1
−0.5
0
0.5
1
1.5
time (sec)
a
n
g
le
/
a
n
g
u
la
r
v
e
lo
c
it
y
q
2
p
2
0 50 100 150 200 250
−20
−10
0
10
20
30
40
50
time (sec)
F
o
r
c
e
0 50 100 150 200 250
−4
−3
−2
−1
0
1
2
time (sec)
C
o
n
t
r
o
l
u
(a) (b)
Figure 517: (a) State trajectory of the cartpole system for the initial condition
(5, 0, π/3, 0) and σ
i
() = sat
0.1
(), (b) The force F and control input u.
Remark 5.4.1. It is needless to mention that the controller obtained here for global
asymptotic stabilization of the cartpole system over the upper halfplane is much
simpler than the one given by Mazenc and Praly [56].
136
In chapter 7, we introduce a diﬀerent approach to stabilization of cartpole sys
tem that is more general than the one presented here. This new method applies to
nonlinear systems that are not necessarily in strict feedforward/feedback forms.
5.4.2 Aggressive Swingup of the Inverted Pendulum
In this section, we provide a state feedback law for aggressive swingup of an inverted
pendulum on a cart in a single action without swinging the pendulum. In the previous
section, we showed that using noncollocated feedback linearization the dynamics of
q
2
can be reduced to a double integrator
˙ q
2
= p
2
, ˙ p
2
= u
a globally exponentially stabilizing (i.e. aggressive) state feedback that asymptotically
stabilizes (q
2
, p
2
) = 0 is given by
u = −c
1
q
2
−c
2
p
2
where c
1
, c
2
> 0 are constants. The main diﬃculty in stabilizing the angle of the
pendulum from initial angle of q
2
= π to ﬁnal angle of q
2
= 0 is that at some moment
of time, the state trajectory of q
2
passes through the point q
2
= π/2. Based on
(5.20), this means F → ∞ due to the division by cos(q
2
). To avoid this singularity,
we saturate the control input F at a maximum level F
max
that is relatively large, i.e.
F
max
1. In other words, we apply the following approximate control
F
app
(q
2
, p
2
) = F
max
sat(F/F
max
)
where
F(q
2
, p
2
) = (
m
11
m
22
m
21
(q
2
)
−m
12
(q
2
))(c
1
q
2
+ c
2
p
2
) + (m
1
+ m
2
)g tan(q
2
) −m
2
l
2
sin(q
2
)p
2
2
The diﬀerence between F and F
app
F −F
app
= φ(F)
can be expressed in terms of the famous deadzone nonlinearity φ shown in Figure
518. The function φ(F) is zero over the interval [−F
max
, F
max
] and linear with unit
slope outside of this interval. The dynamics of the closed loop system with control
F = F
app
(q
2
, p
2
) can be expressed as
˙ q
2
= p
2
˙ p
2
= −c
1
q
2
−c
2
p
2
+
m
21
(q
2
)
D(q
2
)
φ(F(q
2
, p
2
))
(5.22)
where D(q
2
) = det(M(q
2
)) > 0. Apparently, whenever [F[ < F
max
(i.e. the controller
does not saturate) φ(F) = 0 and (q
2
, p
2
) = (0, 0) is exponentially stable as long as the
137
Φ( ) F
F
F F
max max
Figure 518: The deadzone nonlinearity.
solution passes through an area that the controller saturates, i.e. [F
app
(q
2
, p
2
)[ = F
max
.
The nonlinear feedback F(q
2
, p
2
) has a discontinuity along the lines q
2
= ±π/2 in the
(q
2
, p
2
)plane. This discontinuity property is closely related to the function p
2
= α(q
2
)
deﬁned by
α(q
2
) = −
c
1
c
2
q
2
−
m
11
m
2
gl
2
sin(q
2
)
c
2
D(q
2
)
In fact, F(q
2
, p
2
) is continuous at isolated points p
2
= α(±π/2) on the lines q
2
= ±π/2.
The following result determines the area in which the controller F
app
(q
2
, p
2
) saturates.
Proposition 5.4.2. Consider the neighborhood
N
= ¦(q
2
, p
2
) ∈ I R [ [q
2
−π/2[ ≤ , [p
2
[ ≤ n
0
π, n
0
> 0¦
for some 0 < ≤ π/2 and I = [0, π]. Then, there exist F
∗
(, n
0
) > 0 in the form
F
∗
(, n
0
) =
a
1
+a
2
n
0
sin()
+ a
3
n
2
0
, a
1
, a
2
, a
3
> 0
and a neighborhood U
⊂ N
of q
2
= π/2 such that for all F
max
≥ F
∗
, the controller
saturates (i.e. [F
app
(q
2
, p
2
)[ = F
max
) over (q
2
, p
2
) ⊂ U
and remains unsaturated over
I [−n
0
π, n
0
π] ` N
.
Proof. F(q
2
, p
2
) can be rewritten as
F =
c
2
D(q
2
)(p
2
−α(q
2
))
m
2
l
2
cos(q
2
)
−m
2
l
2
sin(q
2
)p
2
2
which is discontinuous along the lines q
2
= ±π/2 and p
2
= α(±π/2). We have
[F[ ≤
D
0
[c
1
q
2
+ c
2
p
2
[ +m
11
m
2
gl
2
m
2
l
2
1
[ cos(q
2
)[
+ m
2
l
2
p
2
2
≤
D
0
(c
1
+c
2
n
0
)π + (m
1
+ m
2
)m
2
gl
2
m
2
l
2
1
[ cos(q
2
)[
+ m
2
l
2
n
2
0
π
2
138
where D
0
> 0 is a constant satisfying 0 < D(q
2
) ≤ D
0
:= m
11
m
22
and the equality is
achieved at q
2
= π/2. Since cos(q
2
) takes its maximum at the boundary of N
where
q
2
= π/2 ±, for all (q
2
, p
2
) ∈ N
, D(q
2
) < D
0
and we get
[F[ <
D
0
(c
1
+ c
2
n
0
)π + (m
1
+m
2
)m
2
gl
2
m
2
l
2
sin()
+ m
2
l
2
n
2
0
π
2
Hence, deﬁning
F
∗
(, n
0
) =
D
0
(c
1
+ c
2
n
0
)π + (m
1
+ m
2
)m
2
gl
2
m
2
l
2
sin()
+ m
2
l
2
n
2
0
π
2
the controller F
app
does not saturate outside of the region N
. On the other hand,
for p
2
= α(π/2), F(q
2
, p
2
) → ∞. By deﬁnition, for any F
max
≥ F
∗
, there exists a
neighborhood U
⊂ N
of the line q
2
= π/2 where F(q
2
, p
2
) ≥ F
max
and therefore the
controller saturates in the region U
. Over the region N
` U
the controller might or
might not saturate. N
is an upper bound (w.r.t. ⊂) for the maximal region U
that
the controller saturates.
The values of F
∗
for some typical choices of , n
0
are given in Table 5.1. Using some
standard phase plane arguments, one can show that F
app
semiglobally asymptotically
stabilizes (q
2
, p
2
) = 0 for a suﬃciently large F
max
≥ F
∗
(, n
0
) and the solution expo
nentially converges to the origin over the region ¦(q
2
, p
2
) ∈ R
2
[q
2
∈ [−π/2+, π/2−]¦.
Due to ˙ q
2
= p
2
, q
2
is strictly decreasing in the region q
2
p
2
< 0 and therefore the so
lution passes through the saturation region N
once and exponentially converges to
the origin under the condition that c
1
, c
2
> 0 are suﬃciently large. The simulation
Table 5.1: The Values of F
∗
(, n
0
).
(n0, ) π/20 π/10 π/5 π/4 π/2
1 454 234 127 107 78
2 637 337 192 164 125
results for this controller are shown in Figure 519. The parameters are chosen as
F
max
= 100 and c
1
= c
2
= 4, g = 9.8. The pendulum starts from the downward
position at q
2
= π and the angle of the pendulum (almost) exponentially converges
to zero. The convergence is not truly exponential, due to the fact that F(q
2
, p
2
) is
saturated for a relatively short period of time. A sample path in (q
2
, p
2
)plane for the
initial condition starting at (q
2
, p
2
) = (π, 0) is shown in Figure 519 (b).
5.4.3 Switchingbased Controller
For the inverted initial condition (0, 0, π, 0) (i.e q
2
= π), one can use the controller
F
app
to bring the pendulum to the upper halfplane area and switch (say whenever
[q
2
[ ≤ π/3) to the globally stabilizing controller that makes use of nested satura
139
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
−3
−2
−1
0
1
2
3
4
time (sec)
a
n
g
le
/
a
n
g
u
la
r
v
e
lo
c
it
y
q
2
p
2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
−100
−50
0
50
100
time (sec)
c
o
n
t
r
o
l
f
o
r
c
e
−4 −3 −2 −1 0 1 2 3 4
−4
−3
−2
−1
0
1
2
3
4
q
2
p
2
(a) (b)
Figure 519: (a) Aggressive swingup of an inverted pendulum on a cart from the
downward position (q
2
, p
2
) = (π, 0), (b) The path of the pendulum in (q
2
, p
2
)plane.
tions. This switchingbased controller semiglobally asymptotically stabilizes the up
right equilibrium point of the cartpole system over the whole plane with q
2
∈ [−π, π].
The semiglobal nature of this stabilization is due to bounded initial angular velocity
and speed required for the former controller that brings the pendulum to the up
per half plane. The simulation results for the overall stabilization of the cartpole
system for the initial condition (0, 0, π, 0) where the pendulum is in the downward
position are shown in Figure 520. This ﬁgure demonstrates that the switchingbased
controller aggressively stabilizes the origin for the cartpole system.
5.5 The Rotating Pendulum
The Rotating Pendulum consists of an inverted pendulum on a rotating arm, as shown
in Figure 521. The elements of the inertia matrix for the rotating pendulum are given
by (see section A.6)
m
11
(q
2
) = I
1
+ m
1
l
2
1
+ m
2
(L
2
1
+l
2
2
sin
2
(q
2
))
m
12
(q
2
) = m
21
(q
2
) = m
2
L
1
l
2
cos(q
2
)
m
22
= I
2
+ m
2
l
2
2
The potential energy for this system is
V (q
1
, q
2
) = m
2
gl
2
cos(q
2
)
Clearly, the inertia matrix of the rotating pendulum only depends on q
2
. Therefore,
q
x
= q
1
is the external variable and q
s
= q
2
is the shape variable of the rotating
pendulum. Due to lack of actuation of the shape variable, the rotating pendulum is a
Class–II underactuated mechanical system. The EulerLagrange equations of motion
140
0 5 10 15 20 25
−15
−10
−5
0
5
10
time (sec)
p
o
s
/
v
e
lo
c
it
y
q
1
p
1
0 5 10 15 20 25
−3
−2
−1
0
1
2
3
4
time (sec)
a
n
g
le
/
a
n
g
u
la
r
v
e
lo
c
it
y
q
2
p
2
0 5 10 15 20 25
−100
−50
0
50
100
time (sec)
F
o
r
c
e
−0.5 0 0.5 1 1.5 2 2.5 3 3.5
−2.5
−2
−1.5
−1
−0.5
0
0.5
q
2
p
2
(a) (b)
Figure 520: (a) State trajectory of the cartpole system for the initial condition
(0, 0, π, 0), (b) The input force F and the trajectory in (q
2
, p
2
)plane.
x
y
z
1
2
q
m , L , I
m , L , I
2
1 1
2
q
1
2
Figure 521: The Rotating Pendulum
for the rotating pendulum are given by
m
11
(q
2
)¨ q
1
+m
12
(q
2
)¨ q
2
+ 2m
2
l
2
2
sin(q
2
) cos(q
2
) ˙ q
1
˙ q
2
−m
2
l
2
L
1
sin(q
2
) ˙ q
2
2
= τ
m
21
(q
2
)¨ q
1
+m
22
(q
2
)¨ q
2
−m
2
l
2
2
sin(q
2
) cos(q
2
) ˙ q
2
1
−m
2
gl
2
sin(q
2
) = 0
Noting that the second line of this equation can be simpliﬁed as
¨ q
1
+
m
22
m
21
(q
2
)
¨ q
2
−
l
2
L
1
sin(q
2
) ˙ q
2
1
−
g
L
1
tan(q
2
) = 0
and applying the change of control
τ = (m
12
(q
2
) −
m
11
(q
2
)m
22
m
21
(q
2
)
)u + 2m
2
l
2
2
sin(q
2
) cos(q
2
) ˙ q
1
˙ q
2
−m
2
l
2
L
1
sin(q
2
) ˙ q
2
2
+ m
11
(q
2
)
l
2
L
1
sin(q
2
) + m
11
(q
2
)
g
L
1
tan(q
2
)
141
(which is welldeﬁned for q
2
= π/2) reduces the dynamics of the shape variable q
2
to
˙ q
2
= p
2
˙ p
2
= u
Based on theorem 4.2.2, we apply the following change of coordinates
q
r
= q
1
+ γ(q
2
)
p
r
= m
21
(q
2
)p
1
+ m
22
p
2
or
z
1
= q
r
z
2
= p
r
/m
21
(q
2
)
where
γ(q
2
) =
q
2
0
m
22
m
21
(θ)
dθ =
(m
2
l
2
2
+I
2
)
m
2
l
2
L
1
log
1 + tan(q
2
/2)
1 −tan(q
2
/2)
, q
2
∈ (−π/2, π/2)
that transforms the dynamics of the rotating pendulum into
˙ z
1
= z
2
˙ z
2
=
g
L
1
tan(q
2
) +
l
2
L
1
sin(q
2
)(z
2
−
m
22
m
21
(q
2
)
p
2
)
2
+
m
22
m
21
(q
2
)
tan(q
2
)p
2
2
˙ q
2
= p
2
˙ p
2
= u
(5.23)
The nonlinear system in (5.23) is neither in feedforward form (due to the term
l
2
sin(q
2
)z
2
2
/L
1
in the equation of ˙ z
2
which depends on z
2
), nor in strict feedback
form. Therefore, direct feedforwarding or backstepping procedures are not applicable
to the rotating pendulum. Based on a recent work [10], global/semiglobal asymptotic
stabilization of the rotating pendulum to its upright equilibrium point is considered
an open problem. In chapter 7, we will address this problem based on a theory that
is developed in this thesis for asymptotic stabilization of cascade nonlinear systems in
nontriangular normal forms. To remove the constraint q
2
∈ (−π/2, π/2), we apply the
following invertible global change of coordinates and control (similar to the cartpole
system)
ξ
1
= tan(q
2
)
ξ
2
= (1 + tan(q
2
)
2
)p
2
v = (1 + tan(q
2
)
2
)(u + 2 tan(q
2
)p
2
)
This transforms the dynamics of the rotating pendulum in (5.23) into
˙ z
1
= z
2
˙ z
2
= ξ
1
(
g
L
1
+
l
2
L
1
1
1 + ξ
2
1
(z
2
−
m
22
m
2
l
2
L
1
ξ
2
1 + ξ
2
1
)
2
+
m
22
m
2
l
2
L
1
ξ
2
2
(1 + ξ
2
1
)
3
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(5.24)
142
Notice that in the equation of ˙ z
2
the quantity in the parentheses is strictly positive
deﬁnite. This equation can be simpliﬁed further as
˙ z
1
= z
2
˙ z
2
= ξ
1
(k
1
+ k
2
z
2
2
(1 +ξ
2
1
)
1
2
−k
3
z
2
ξ
2
(1 + ξ
2
1
)
+ k
4
ξ
2
2
(1 + ξ
2
1
)
3
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(5.25)
where
k
1
=
g
L
1
, k
2
=
l
2
L
1
, k
3
=
2m
22
m
2
L
2
1
, k
4
=
m
22
m
2
l
2
L
1
(1 +
m
22
m
2
L
2
1
)
are all positive constants. Observe that k
i
/k
1
, i = 2, 3, 4 all vanish as l
2
/g → 0
(i.e. in relatively high gravity). We use this property later for semiglobal asymptotic
stabilization of the rotating pendulum in its upright position.
5.5.1 Aggressive Swingup of the Rotating Pendulum
In this section, we introduce a state feedback that aggressively swings up an inverted
pendulum mounted on a rotating arm in a single action without swinging it back and
forth (i.e. a passivitybased swingup as in [6, 112]). The procedure is very similar to
the one applied to the cartpole system. Thus, we omit the details. Simply, one has
to set
u = −c
1
q
2
−c
2
p
2
, c
1
, c
2
> 0
which gives the following input torque
τ =
D(q
2
)
m
2
l
2
L
1
cos(q
2
)
(c
1
q
2
+ c
2
p
2
) + 2m
2
l
2
2
sin(q
2
) cos(q
2
) ˙ q
1
˙ q
2
−m
2
l
2
L
1
sin(q
2
) ˙ q
2
2
+ m
11
(q
2
)
l
2
L
1
sin(q
2
) + m
11
(q
2
)
g
L
1
tan(q
2
)
where D(q
2
) = det(M(q
2
)). Due to division by cos(q
2
) in the last equation as q
2
→
π/2, τ(q
2
, p
2
, p
1
) → ∞ for almost every p
2
. To avoid this singularity, we apply the
saturated torque as the following
τ
app
(q
2
, p
2
, p
1
) = τ
max
sat(τ/τ
max
)
where τ
max
> 0 is the maximum allowed torque (to be speciﬁed later). One obtains
the following result which is very similar to proposition 5.4.2 for the cartpole system.
Proposition 5.5.1. Consider a neighborhood N
= ¦(q
2
, p
2
, p
1
) ∈ I R
2
[[q
2
−π/2[ ≤
, [p
2
[ ≤ n
0
π, [p
1
[ < n
1
π, n
0
≥ n
1
> 0¦ for some 0 < ≤ π/2 and I = [0, π]. Then,
143
there exist τ
∗
(, n
0
, n
1
) > 0 in the form
τ
∗
(, n
0
, n
1
) =
a
1
+ a
2
n
0
sin()
+ a
3
n
2
0
+ a
4
n
0
n
1
+ a
5
, a
i
> 0, i = 1, . . . , 5
and a neighborhood U
⊂ N
of q
2
= π/2 such that for all τ
max
≥ τ
∗
, τ
app
(q
2
, p
2
, p
1
)
saturates (i.e. [τ
app
(q
2
, p
2
)[ = τ
max
) over (q
2
, p
2
) ⊂ U
and it remains unsaturated
over I [−n
0
π, n
0
π] ` N
.
Proof. Following the line of proof of proposition 5.4.2, let us deﬁne
D
0
= D(π/2), M
0
= m
11
(π/2)
and note that for all q
2
∈ [−π, π], D(q
2
) ≤ D
0
and m
11
(q
2
) ≤ M
0
. For 0 < ≤ π/2
and q
2
= π/2 −, we have
[τ[ <
D
0
(c
1
+ n
0
c
2
)π + M
0
m
2
gl
2
m
2
l
2
L
1
sin()
+ 2m
2
l
2
2
n
0
n
1
π
2
+ m
2
l
2
L
1
n
2
0
π
2
+
M
0
l
2
L
1
Thus, we deﬁne τ
∗
(, n
0
) as the right hand side of the last equation. In other words,
we have
a
1
=
D
0
c
1
+ M
0
m
2
gl
2
m
2
l
2
L
1
, a
2
=
D
0
c
2
m
2
l
2
L
1
, a
3
= m
2
l
2
L
1
π
2
, a
4
= 2m
2
l
2
2
π
2
, a
5
=
M
0
l
2
L
1
For the choice of parameters m
1
= 1, l
1
= .5, m
2
= 1, l
2
= .75, g = 9.8, n
1
= 1,
the values of τ
∗
(, n
0
, 1) are shown in Table 5.2. The simulation results for almost
Table 5.2: The Values of τ
∗
(, n
0
, 1).
(n0, ) π/20 π/10 π/5 π/4 π/2
1 1471 755 406 341 247
2 1885 981 541 459 340
exponential swingup of the rotating pendulum from the downward position (i.e.
initial condition (0, 0, π, 0)) to (q
2
, p
2
) = (0, 0) are shown in Figure 522. Apparently,
the overall solution (q
2
(t), p
2
(t)) is not exponentially vanishing but it has a relatively
long exponentially vanishing tail–thus the name almost exponential. The maximal
torque is chosen to be τ
max
= 200.
5.6 The Pendubot
The Pendubot is a twolink revolute planar robot with one actuator at the shoulder,
as shown in Figure 523. In this section, we provide an explicit global change of
coordinates that transforms the dynamics of the Pendubot into a cascade nonlinear
144
0 0.5 1 1.5 2 2.5 3
−6
−4
−2
0
2
4
time (sec)
q
2
p
2
0 0.5 1 1.5 2 2.5 3
−60
−40
−20
0
20
40
time (sec)
c
o
n
t
r
o
l
(
t
o
r
q
u
e
)
−4 −3 −2 −1 0 1 2 3 4
−6
−4
−2
0
2
4
6
q
2
p
2
(a) (b)
Figure 522: (a) Aggressive swingup of an inverted pendulum on a rotating arm from
the downward position (q
2
, p
2
) = (π, 0), (b) Trajectory of the pendulum in (q
2
, p
2
)
plane.
system in nontriangular quadratic normal form. Stabilization of the Pendubot to its
upright equilibrium point using a single (i.e. without switching) static state feedback
is addressed in chapter 7. The inertia matrix for the pendubot is given by (see section
q
q
l
1
L
1
m I
m I
2 2
1 1
1
2
x
y
Figure 523: The Pendubot
A.3)
m
11
(q
2
) = m
1
l
2
1
+ m
2
(L
2
1
+ l
2
2
+ 2L
1
l
2
cos(q
2
)) + I
1
+ I
2
m
12
(q
2
) = m
21
(q
2
) = m
2
(l
2
2
+ L
1
l
2
cos(q
2
)) + I
2
m
22
(q
2
) = m
2
l
2
2
+ I
2
where q
i
, m
i
, L
i
, l
i
, and I
i
denote angles, masses, lengths, lengths of center of masses,
and inertia, respectively. For further use, it is simpler to use the following expressions
for the inertia matrix
m
11
(q
2
) = a + 2b cos(q
2
)
m
12
(q
2
) = m
21
(q
2
) = c + b cos(q
2
)
m
22
(q
2
) = c
145
where a, b, c > 0 are given by
a = m
1
l
2
1
+ m
2
(L
2
1
+ l
2
2
) + I
1
+I
2
b = m
2
l
2
L
1
c = m
2
l
2
2
+ I
2
The potential energy of the pendubot is
V (q
1
, q
2
) = (m
1
l
1
+ m
2
L
1
)g cos(q
1
) + m
2
l
2
g cos(q
1
+ q
2
).
The inertia matrix of the pendubot only depends on q
2
. Thus, q
2
is the shape variable
of the pendubot. Since the shape variable q
2
is unactuated, the pendubot is a Class–II
underactuated system. The EulerLagrange equations of motion for the pendubot are
as the following
m
11
(q
2
)¨ q
1
+ m
12
(q
2
)¨ q
2
−2b sin(q
2
) ˙ q
1
˙ q
2
−b sin(q
2
) ˙ q
2
2
+ g
1
(q
1
, q
2
) = τ
m
21
(q
2
)¨ q
1
+ m
22
(q
2
)¨ q
2
+ b sin(q
2
) ˙ q
2
1
+ g
2
(q
1
, q
2
) = 0
(5.26)
where the gravity terms are
g
1
(q
1
, q
2
) = −(m
1
l
1
+ m
2
L
1
)g sin(q
1
) −m
2
l
2
g sin(q
1
+ q
2
)
g
2
(q
1
, q
2
) = −m
2
l
2
g sin(q
1
+ q
2
)
Applying the feedback law
τ = (m
12
−
m
11
m
22
m
21
)u −b sin(q
2
)(
m
11
m
21
p
2
1
+ 2p
1
p
2
+p
2
2
) + g
1
(q
1
, q
2
) −
m
11
m
21
g
2
(q
1
, q
2
)
reduces the dynamics of q
2
to a double integrator
˙ q
2
= p
2
, ˙ p
2
= u
Applying the change of coordinates
q
r
= q
1
+ γ(q
2
)
p
r
= m
11
(q
2
)p
1
+ m
22
p
2
where
γ(q
2
) =
q
2
0
m
22
m
21
(θ)
dθ =
2c
√
c
2
−b
2
arctan
c −b
c + b
tan(
q
2
2
)
; c > b, q
2
∈ [−π, π]
and denoting q
s
= q
2
, we get
˙ q
r
= p
r
/m
21
(q
s
)
˙ p
r
= m
2
l
2
g sin(q
r
−γ(q
s
) + q
s
) −b sin(q
s
)
(p
r
−m
22
p
s
)(p
r
−(m
11
+m
22
)p
s
)
m
2
11
˙ q
s
= p
s
˙ p
s
= u
146
which apparently the equation of ˙ p
r
consists of a quadratic term in (p
r
, p
s
) and a
reduced gravity term
g
r
(q
r
, q
s
) = m
2
l
2
g sin(q
r
−γ(q
s
) + q
s
)
as in theorem 4.2.2. To further simplify the reduced dynamics of the pendubot, deﬁne
the change of coordinates
z
1
= q
1
+ γ(q
2
)
z
2
= p
r
/m
21
(q
s
)
we get
˙ z
1
= z
2
˙ z
2
=
m
2
l
2
g sin(z
1
−γ(q
2
) +q
2
)
m
21
(q
2
)
−
b sin(q
2
)
m
21
(q
2
)
[(z
2
−
m
22
m
21
(q
2
)
p
2
)
2
+
m
22
m
21
(q
2
)
p
2
2
]
˙ q
2
= p
2
˙ p
2
= u
(5.27)
Clearly, the dynamics of the pendubot in (5.27) is neither in strict feedback form, nor
in strict feedforward form. Therefore, the overall dynamics of the pendubot cannot
be stabilized using famous backstepping and forwarding approaches. Stabilization of
nonlinear systems in nontriangular forms like Pendubot is addressed in chapter 7.
5.7 The BeamandBall System
The beamandball system consists of a beam and a ball with radius r on it (see
Figure 524). The task is to bring the ball from any initial position (and speed) to
the origin. Let d ≥ 0 denote the distance between the center of the mass of the ball
and the beam (d = r in Figure 524). In the current literature [34, 102, 80, 79], what
is known as the beamandball system corresponds to the special case where d = 0.
We refer to this case as the conventional beamandball system and to the case where
d > 0 as the beamandball system. The following treatment applies to the both
cases of d = 0 and d > 0. From section A.4, the elements of the inertia matrix of the
d
x
r
I
1
θ
τ
m , I
2
Figure 524: The BeamandBall System.
147
beamandball system are given by
m
11
(x) = I
1
+ m(x
2
+d
2
)
m
12
= m
21
= −md
m
22
= mλ
where I
1
, m, I
2
denote the inertia of the beam, the mass of the ball , and the inertia
of the ball, respectively. Also, λ = 1 + I
2
/mr
2
> 1. The inertia matrix of the
beamandball system only depends on the position of the ball x. This means that
the position of the ball is the shape variable of the beamandball system. Since the
position is unactuated, the beamandball system is a Class–II underactuated system.
The equations of motion for the beamandball system with d > 0 is as the following
[I
1
+ m(x
2
+d
2
)]
¨
θ − md¨ x + 2mx ˙ x
˙
θ + mg(xcos(θ) −d sin(θ)) = τ
d
−md
¨
θ + mλ¨ x −mx
˙
θ
2
+mg sin(θ) = 0
(5.28)
While the dynamics of the conventional beamandball system with d = 0 is in the
form
(I
1
+ mx
2
)
¨
θ + 2mx ˙ x
˙
θ + mgxcos(θ) = τ
0
mλ¨ x −mx
˙
θ
2
+ mg sin(θ) = 0
(5.29)
Clearly,
¨
θ does not appear in the second equation and after applying
τ
0
= (I
1
+ mx
2
)u + 2mx ˙ x
˙
θ +mgxcos(θ)
after normalizing the units of (x, v) by 1/λ, the dynamics of the conventional beam
andball system can be written as
˙ x = v
˙ v = −g sin(θ) + xω
2
˙
θ = ω
˙ ω = u
(5.30)
The last equation is exactly the same as the model of the beamandball system
originally obtained in [34] and used in [102, 80, 79]. In contrast, consider the beam
andball system with d > 0 in (5.28). After applying the following global change of
coordinates
z
1
= −dθ + λx
z
2
= −dω +λv
(5.31)
and collocated partially linearizing feedback law
τ
d
= (I
1
+ m(x
2
+ d
2
) −
md
2
λ
)u −
mdx
λ
ω
2
+
mdg sin(θ)
λ
+ 2mxvω + mg(xcos(θ) −d sin(θ))
(5.32)
148
the dynamics of the beamandball system with d > 0 transforms into
˙ z
1
= z
2
˙ z
2
= −g sin(θ) + (z
1
+ dθ)ω
2
˙
θ = ω
˙ ω = u
(5.33)
where the units of (z
1
, z
2
) are normalized by 1/λ. Equation (5.33) has many similar
ities to the normal form of the conventional beamandball system in (5.30).
5.8 Control of Multilink Underactuated Planar
Robot Arms
In this section, we discuss reduction of a planar multilink planar robot arm. Let us
start by a planar threelink robot. The inertia matrix of a threelink arm is as the
following (see section A.2)
M(q
2
, q
3
) =
m
11
(q
2
, q
3
) m
12
(q
2
, q
3
) m
13
(q
2
, q
3
)
m
21
(q
2
, q
3
) m
22
(q
3
) m
23
(q
3
)
m
31
(q
2
, q
3
) m
32
(q
3
) m
33
¸
¸
Notice that there is an interesting structure in the way the shape variables (q
2
, q
3
)
appear in the inertia matrix of a threelink robot. The following result shows that
such a nested structure in general exists for any nlink planar robot.
Theorem 5.8.1. Consider an nlink planar robot with revolute joints q
1
, . . . , q
n
. De
ﬁne the following index sets
I
k
:= ¦α[α = ij; (k, k) ≤ (i, j) ≤ (n, n)¦, k = 1, . . . , n
J
k
:= I
k
` I
k+1
, k = 1, . . . , n; I
n+1
= 0
Then ∀α ∈ J
k
: m
α
= m
α
(q
k+1
, . . . , q
n
), ∀k = 1, . . . , n−1 and m
α
= const. for α ∈ J
n
.
Proof. The proof is by direct calculation using the rotation matrices R
i
= R(q
i
) ∈
SO(2) which represent the orientation of each link in R
2
. Notice that by deﬁnition
J
k
’s are disjoint sets.
Remark 5.8.1. The inertia matrix of an nlink planar robot arm is independent of q
1
.
i.e. q
x
= q
1
and q
s
= (q
2
, . . . , q
n
) are the vector of external variables and the vector
of shape variables for a multilink planar robot arm, respectively.
In examples 4.5.1 and 4.5.2, we showed that there are three possible actuation
conﬁgurations for a threelink planar robot with two actuators. Furthermore, based
on the method of momentum decomposition, we proved that stabilization of a three
link robot with two actuators reduces to stabilization of the Acrobot, or the Pendubot.
In the following, we show that this fact can be generalized to a multilink planar robot
underactuated by one.
149
Theorem 5.8.2. There exists a change of coordinates in explicit form that transforms
the dynamics of an nlink planar robot (n ≥ 2) with (n−1) actuators into the perturbed
reduced normal forms of twolink planar robots (i.e. the Acrobot or the Pendubot) in
cascade with (n −2) doubleintegrators.
Proof. First, assume q
1
is not actuated and (n −1) of shape variables are actuated.
Apparently, it is possible to linearize the dynamics of (n − 2) shape variables using
(n − 2) control inputs. Only one actuated shape variable, say q
2
, and q
1
are left.
Following the line of momentum decomposition procedure in chapter 4, set all other
joint angles to a constant locked conﬁguration (e.g. zero). The inertia matrix of
the overall system only depends on q
2
and the reduced form for this system is the
perturbed normal form of the Acrobot. Now, assume q
1
is actuated. Then, (n − 2)
controls are remained for (n−1) shape variables. Therefore, the dynamics of all shape
variables except for one, say q
2
, can be linearized using collocated partial feedback
linearization. Only an unactuated shape variable q
2
and an actuated external variable
q
1
are left. Thus, by applying momentum decomposition, one obtains the perturbed
normal form of the Pendubot in cascade with (n − 2) doubleintegrators (resulted
from partial feedback linearization).
Remark 5.8.2. The preceding proof applies to any underactuated (simple) mechanical
system with n degrees of freedom, one external variable q
1
, (n − 1) shape variables
(q
2
, . . . , q
n
), and (n−1) control inputs. The special structure of the inertia matrix of
the nlink robot here plays no major role except for the fact that M is independent
of q
1
.
5.9 Trajectory Tracking for a Flexible OneLink
Robot
In this paper, we introduce a method that provides a nonlinear noncollocated out
put for trajectory tracking of a ﬂexible onelink robot. The link is modeled as a
ﬁniteorder Lagrangian system obtained from truncated modal analysis. This non
collocated output is derived based on viewing a ﬂexiblelink robot arm as an under
actuated mechanical system and then applying an appropriate change of coordinates
that transforms the system into a cascade nonlinear system with a minimumphase
zero dynamics. The obtained output is the angle of rotation augmented with the
(generalized) saturation of the weighted amplitudes of the deﬂection modes of the
ﬂexible link.
Flexiblelink robots are ﬁnding their way in industrial and space robotics appli
cations due to their lighter weight and faster response time compared to rigid robots.
Control of ﬂexible robots has been studied extensively for more than a decade by
several researchers [12, 18, 25, 81, 110, 107] (see [24] for a recent survey). Despite
their applications, control of ﬂexible link robots has proven to be rather complicated.
The simplest possible ﬂexible arm is a robot with a single ﬂexible link. The dynamic
model for a ﬂexiblelink robot that has been used by almost all of the researchers is
the EulerBernoulli model of a beam. This model is a fourth order equation that leads
150
to an originally inﬁnitedimensional model of a ﬂexible link. The common collocated
output for trajectory tracking of ﬂexible onelink robots is the angle of the rotation
of the link at the base. The performance of this output measurement turns out to be
not satisfactory due to a weak control of the vibrations of the link [20]. This initiated
ﬁnding other noncollocated output measurements like the position of the endpoint of
the link [18]. Based on [85], this choice of the output leads to a nonminimum phase
zerodynamics. In [20], a noncollocated output is proposed as a linear combination
of the angle of rotation and the slope of the beam at its tip. The model used in [20]
is an inﬁnite dimensional linear model of a ﬂexible link robot. Then, an approximate
ﬁnite order compensator is designed based on [107]. Here, we take a fundamentally
diﬀerent approach to ﬁnd a noncollocated output. We use a ﬁniteorder statespace
model of a ﬂexible link derived by truncated modal analysis based on a Lagrangian
formalism due to De Luca and Siciliano [25]. This model is naturally in the form of
an underactuated mechanical system with (m + 1) degrees of freedom and a single
actuator where m is the number of deformation modes of the ﬂexible link. Based on
our previous results on normal forms for underactuated mechanical systems [68, 66]
and theorem 4.2.3, we propose a noncollocated output as the angle of the rotation
augmented with a generalized saturated weighted linear combination of the deforma
tion amplitudes of the link. We prove that the zerodynamics corresponding to this
output is minimumphase.
5.9.1 Model of a Flexible Link
The ﬂexible link depicted in Figure 525 is modeled as an EulerBernoulli beam [25]
satisfying
EI
∂
4
w(ζ, t)
∂ζ
4
+ ρAL
4
∂
2
w(ζ, t)
∂t
2
= 0
where ζ = x/L is the normalized position along the link of length L and A, I, E, ρ
are physical parameters of the link. By the assumption of separability in time and
space, the deformation of the beam can be expressed as W(ζ, t) = φ(ζ)δ(t). After
w
M J
x
y
θ
α
L L
X
Y
Figure 525: A ﬂexible onelink robot arm.
performing a truncated modal analysis with m modes, the Lagrangian equations of
151
the ﬂexible link can be written as the following [25]
¸
m
11
(δ) m
12
m
21
m
22
¸
¨
θ
¨
δ
+
¸
c
1
(
˙
θ, δ,
˙
δ)
c
2
(
˙
θ, δ)
+
¸
0
Kδ + F
˙
δ
=
¸
τ
0
(5.34)
where θ is the angle of rotation, δ ∈ R
m
denotes the deformation amplitudes, M(δ) =
¦m
ij
¦ is the inertia matrix that is a positive deﬁnite symmetric matrix and K, F are
positive deﬁnite matrices as well. The expressions for m
ij
’s are explicitly given in
[25]. Here, we only need an explicit form of m
11
(δ) (the rest of m
ij
’s are constant) as
m
11
(δ) = k
1
+ k
2
(Φ
T
e
δ)
2
(5.35)
where k
1
> 0, k
2
= M
L
i.e. payload mass, Φ
e
∈ R
m
is a constant column vector that
is determined by the m eigenfunctions obtained from the modal analysis. In addition,
Coriolis and centrifugal terms can be explicitly given as
c
1
(
˙
θ, δ,
˙
δ) = 2M
L
˙
θ(Φ
T
e
δ)(Φ
T
e
˙
δ)
c
2
(
˙
θ, δ) = −M
L
˙
θ
2
(Φ
T
e
δ)Φ
e
Apparently, from (5.34), the ﬂexible onelink robot is an underactuated mechanical
system with m + 1 degrees of freedom and one control. The important feature of
(5.34) is that the system has kinetic symmetry w.r.t. θ. In the next section, we show
that this symmetry property has a crucial role in derivation of a minimum phase
noncollocated output for the ﬂexible onelink robot.
To apply theorem 4.2.3 to the case of the ﬂexible onelink robot in (5.34), all the
elements of
ω
1
= m
11
(δ)
−1
m
12
dδ
must be exact oneforms. If this holds, then a nonlinear noncollocated output in the
form of y = θ+γ(δ) exists that transforms the system into a cascade nonlinear system.
However, this is false and the oneform m
11
(δ)
−1
m
12
dδ is not exact due to the fact
that m
12
= λΦ
T
e
for any λ ∈ R. This suggests an alternative way of constructing the
desired output. By inspection of m
11
(δ) in (5.35), one can observe that the following
modiﬁed oneform
ω
2
= m
11
(δ)
−1
Φ
T
e
dδ
is exact and thus the expression
˙
θ + m
11
(δ)
−1
Φ
T
e
˙
δ
is integrable as an output y = θ +γ(δ). Next, we show that this is indeed the output
we were seeking.
5.9.2 The Noncollocated Output
In this section, we derive a noncollocated output and prove that its corresponding
zerodynamics is minimum phase. But ﬁrst we need to make some assumptions and
152
give a lemma.
Notation. For any two vectors x, y ∈ R
m
and the positive deﬁnite mm matrix Q,
deﬁne the following inner product
'x, y`
Q
:= x
T
Qy
Assumption 5.9.1. Suppose the following properties hold
i) Φ
e
satisﬁes 'm
21
, m
21
`
m
−1
22
> 'Φ
e
, m
21
`
m
−1
22
.
ii) The matrix Q
0
= m
21
m
12
−m
21
Φ
T
e
is positive deﬁnite.
Remark 5.9.1. By the speciﬁc structure of the inertia matrix M(δ) for a ﬂexible link
robot in [25], assumptions (i) and (ii) are very reasonable and can be algebraicly
checked.
Lemma 5.9.1. Suppose Assumption 5.9.1 holds. Then the following matrices are
positive deﬁnite and thus invertible
Q
1
= m
11
(δ) −Φ
T
e
m
−1
22
m
21
Q
2
= m
22
−m
21
m
−1
11
Φ
T
e
Proof. First, note that the following matrices are symmetric and positive deﬁnite
Q
3
= m
11
(δ) −m
12
m
−1
22
m
21
Q
4
= m
22
−m
21
m
−1
11
(δ)m
12
The proof for this can be found in [90]. This is true regardless of how many deforma
tion modes are chosen as long as m
ij
’s are elements of a symmetric positive deﬁnite
inertia matrix M. Noting that Q
1
can be rewritten as
Q
1
= Q
3
+ (m
12
−Φ
T
e
)m
−1
22
m
21
by i) in Assumption 5.9.1, the second term is positive and Q
3
is positive as well.
Thus, Q
1
is positive deﬁnite. To prove Q
2
is positive deﬁnite, notice that for any
x ∈ R
m
we have
x
T
Q
2
x = x
T
Q
4
x + x
T
m
21
m
11
(δ)
−1
(m
12
−Φ
T
e
)x
or
x
T
Q
2
x = x
T
Q
4
x + m
11
(δ)
−1
x
T
Q
0
x
and because by Assumption 5.9.1 Q
0
is positive deﬁnite and also m
11
(δ)
−1
> 0, Q
2
is
a positive deﬁnite matrix.
Now, we are ready to present our main result for output tracking of a ﬂexiblelink
robot:
153
Theorem 5.9.1. The nonlinear noncollocated output
y = θ + γ(δ) (5.36)
where γ : R
m
→R is given by
γ(δ) =
1
√
k
1
k
2
arctan(
k
1
k
2
Φ
T
e
δ) (5.37)
has global uniform relative degree 3 w.r.t. u = ˙ τ and determines a global change of
coordinates
z
1
= δ, z
2
=
˙
δ, ξ
1
= y, ξ
2
= ˙ y, ξ
3
= ¨ y
that transforms the dynamics of a ﬂexible onelink robot augmented with an integrator
u = ˙ τ into the following partially linear cascade nonlinear system
˙ z
1
= z
2
˙ z
2
= f(z, ξ
1
, ξ
2
, ξ
3
)
˙
ξ
1
= ξ
2
˙
ξ
2
= ξ
3
˙
ξ
3
= v
(5.38)
with a new control
v = α(δ)u +β(δ,
˙
δ,
˙
θ)
In addition, there exists an > 0 with the property ∝
σ
min
(F) such that over
a neighborhood U
(0) = ¦(z
1
, z
2
) : [z
1
[ < , [z
2
[ < ¦ of (z
1
, z
2
) = 0, for the zero
dynamics corresponding to this output the origin is locally uniformly asymptotically
stable (i.e. the noncollocated output y is minimum phase).
Proof. By direct calculations, for y = θ + γ(δ) in (5.36), we have
˙ y =
˙
θ +
Φ
T
e
˙
δ/k
2
k
1
/k
2
+ (Φ
T
e
δ)
2
=
˙
θ +
Φ
T
e
˙
δ
k
1
+k
2
(Φ
T
e
δ)
2
=
˙
θ + m
11
(δ)
−1
(Φ
T
e
˙
δ)
also
¨ y =
¨
θ +m
11
(δ)
−1
Φ
T
e
¨
δ −2k
2
m
11
(δ)
−2
(Φ
T
e
δ)(Φ
T
e
˙
δ)
2
but from the second line of equation (5.34),
¨
δ can be obtained as
¨
δ = −m
−1
22
m
21
¨
θ −m
−1
22
(c
2
+ Kδ + F
˙
δ)
154
Thus, combining the last two equations, we get
¨ y = m
11
(δ)
−1
(m
11
(δ) −Φ
T
e
m
−1
22
m
21
)
¨
θ
− m
11
(δ)
−1
Φ
T
e
m
−1
22
(c
2
+ Kδ + F
˙
δ)
− 2k
2
m
11
(δ)
−2
(Φ
T
e
δ)(Φ
T
e
˙
δ)
2
On the other hand, after cancelling
¨
δ from the ﬁrst and second lines of (5.34), one
obtains
(m
11
(δ) −m
12
m
−1
22
m
21
)
¨
θ + c
1
−m
12
m
−1
22
(c
2
+ Kδ +F
˙
δ) = τ
Hence, taking
˜ m
11
(δ) = (m
11
(δ) −m
12
m
−1
22
m
21
) > 0
we have
¨ y = m
11
(δ)(m
11
(δ) −Φ
T
e
m
−1
22
m
21
) ˜ m
11
(δ)
−1
(τ −c
1
+ m
12
m
−1
22
(c
2
+Kδ +F
˙
δ))
− m
11
(δ)
−1
Φ
T
e
m
−1
22
(c
2
+ Kδ +F
˙
δ) −2k
2
m
11
(δ)
−2
(Φ
T
e
δ)(Φ
T
e
˙
δ)
2
This means that applying the change of control
w = (m
11
(δ) −m
12
m
−1
22
m
21
) (m
11
(δ) −Φ
T
e
m
−1
22
m
21
)
−1
m
11
(δ)τ + β(δ,
˙
δ,
˙
θ)
(where β can be explicitly calculated from the last equation of ¨ y) partially linearizes
the dynamics of the system as
¨ y = w
and therefore the output y has global relative degree 2 w.r.t. τ and global relative
degree 3 w.r.t. v. To determine the stability of the zerodynamics corresponding to
y, let z
1
= δ, z
2
=
˙
δ and set y ≡ 0 (thus, ˙ y, ¨ y ≡ 0). From the second equation in
(5.34), the zerodynamics system can be determined as
(m
22
−m
21
m
11
(δ)
−1
Φ
T
e
)
¨
δ + 2k
2
m
21
m
11
(δ)
−2
(Φ
T
e
δ)(Φ
T
e
˙
δ)
2
+ c
1
+ Kδ + F
˙
δ = 0
but c
1
= −k
2
˙
θ
2
(Φ
T
e
δ)Φ
e
and
˙
θ = −m
11
(δ)
−1
(Φ
T
e
˙
δ) (because ˙ y = 0), thus
c
1
= −k
2
m
11
(δ)
−2
(Φ
T
e
˙
δ)
2
(Φ
T
e
δ)Φ
e
The equations of the zerodynamics can be expressed as the following
(m
22
−m
21
m
11
(δ)
−1
Φ
T
e
)
¨
δ+k
2
m
11
(δ)
−2
(Φ
T
e
δ)(Φ
T
e
˙
δ)
2
(2m
21
−Φ
e
)+Kδ+F
˙
δ = 0 (5.39)
Denoting
Q(δ) = (m
22
−m
21
m
11
(δ)
−1
Φ
T
e
)
and noting that Q(δ) is a positive deﬁnite matrix, the following Lyapunov function
for the zerodynamics system can be proposed
V (δ,
˙
δ) =
1
2
˙
δ
T
Q(δ)
˙
δ +
1
2
δ
T
Kδ
155
Calculating
˙
V along the solutions of the zerodynamics in (5.39), we obtain
˙
V =
˙
δ
T
Q(δ)
¨
δ +
˙
δ
T
Kδ +
1
2
˙
δ
T
˙
Q
˙
δ
where
˙
Q =
∂Q
∂δ
˙
δ = 2k
2
m
21
m
11
(δ)
−2
(Φ
T
e
δ)(Φ
T
e
˙
δ)Φ
T
e
and
1
2
˙
δ
T
˙
Q
˙
δ = k
2
m
11
(δ)
−2
(m
12
˙
δ)(Φ
T
e
δ)(Φ
T
e
˙
δ)
2
Hence
˙
V = −
˙
δ
T
F
˙
δ −k
2
m
11
(δ)
−2
(Φ
T
e
˙
δ)
2
(Φ
T
e
δ) ((m
12
˙
δ) −(Φ
T
e
˙
δ))
and because m
11
(δ)
−2
≤ k
−2
1
, we get
˙
V ≤ −σ
min
(F)[
˙
δ[
2
+
k
2
k
2
1
[Φ
e
[
3
([m
21
−Φ
e
[)[
˙
δ[
3
[δ[
where σ
min
(F) is the smallest singular value of F. Denoting
k
3
=
k
2
k
2
1
[Φ
e
[
3
[m
21
−Φ
e
[
and deﬁning
=
σ
min
(F)
k
3
for [δ[, [
˙
δ[ < (i.e. over U
(0)),
˙
V ≤ 0. But the largest invariant set in z
2
=
˙
δ = 0
(i.e.
˙
V = 0) for (5.39) is (z
1
, z
2
) = (0, 0). Therefore, from LaSalle’s invariance
principle the origin (δ,
˙
δ) = (z
1
, z
2
) = 0 is uniformly locally asymptotically stable for
the zerodynamics system, i.e. the output y is (locally) minimum phase.
Remark 5.9.2. Viewing σ
min
(F) as the strength of damping of the ﬂexible link, the
result of the preceding theorem can be interpreted as follows. The higher the strength
of damping, the larger the region of attraction of the origin for the zerodynamics.
Remark 5.9.3. For a suﬃciently small deformation amplitudes δ. The output y in
(5.36) can be expressed as a linear combination of the angle of rotation and weighted
deformation amplitudes as
y = θ +
1
M
L
(Φ
T
e
δ)
which depends on the payload mass M
L
. This is in agreement with our intuition that
the payload mass matters in trajectory tracking control design for a ﬂexible arm.
156
5.9.3 Tracking Control
The control for trajectory tracking can be given as the following. Let y
d
(t) denote
the desired trajectory and deﬁne the error functions
e
1
= y −y
d
(t), e
2
= ˙ y − ˙ y
d
, e
3
= ¨ y − ¨ y
d
Then
˙ e
1
= e
2
˙ e
2
= e
3
˙ e
3
= u −y
(3)
d
where y
(k)
d
=
d
k
y
d
(t)
dt
k
. Thus, setting
u = y
(3)
d
+ c
p
(y −y
d
) + c
d
( ˙ y − ˙ y
d
) + c
a
(¨ y − ¨ y
d
)
with c
p
, c
d
, c
a
< 0, or applying the partial state feedback
u = y
(3)
d
+ c
p
(ξ
1
−y
d
) + c
d
(ξ
2
− ˙ y
d
) + c
a
(ξ
3
− ¨ y
d
) (5.40)
guarantees that (e
1
, e
2
, e
3
) →0 as t →∞ and for suﬃciently small initial conditions
(δ(0),
˙
δ(0)) asymptotic output tracking can be achieved [36].
5.10 Conﬁguration Stabilization for the VTOL Air
craft
In this section, we give a static state feedback law for global conﬁguration stabilization
of the VTOL (vertical take oﬀ and landing) aircraft. The simpliﬁed dynamics of the
u
ε u
u
θ
1
2
2
x
y
Figure 526: The VTOL aircraft.
157
VTOL aircraft, as shown in Figure 526, is given in [35, 55] as the following
˙ x
1
= x
2
˙ x
2
= −u
1
sin(θ) + u
2
cos(θ)
˙ y
1
= y
2
˙ y
2
= u
1
cos(θ) + u
2
sin(θ) −g
˙
θ = ω
˙ ω = u
2
(5.41)
where = 0. Here, θ denotes the roll angle and the plane moves in a vertical (x
1
, y
1
)
plane. Clearly, the system has three degrees of freedom and only two actuators–thus
it is underactuated. The eﬀect of the body torque u
2
appears in the translational
dynamics of the VTOL aircraft by a factor of . This captures a general property
of real aircrafts with 6 DOF that exhibit a similar input coupling eﬀect. The case
with [[ < 1 resulting in a weak input coupling has been mostly considered in the
literature [35, 55, 80]. Here, we are interested in the strong input coupling case with
an arbitrary = 0. To decouple the eﬀect of u
2
in the translational dynamics of the
VTOL aircraft, ﬁrst let rewrite the dynamics of the VTOL aircraft as
m
xx
¨ q
x
−g
x
(q
x
) = F
r
(q
s
)τ
r
+ F
x
(q
s
)τ
m
ss
¨ q
s
= F
s
(q
s
)τ
(5.42)
where q
x
= (x
1
, y
1
)
T
, q
s
= θ, m
xx
= I
2×2
, m
ss
= 1, F
s
= 1, and F
r
, F
x
, g
x
are given by
F
r
(q
s
) =
¸
−sin(θ)
cos(θ)
, F
x
(q
s
) =
¸
cos(θ)
sin(θ)
, g
x
=
¸
0
−g
which clearly shows F
r
(q
s
) is a unit vector. Based on theorem 4.3.2, we need to check
whether the vector of oneforms ω deﬁned as ω = m
−1
xx
F
x
(q
s
)F
−1
s
(q
s
)m
ss
dq
s
has exact
elements or not. By direct calculation, we get
ω = F
x
(q
s
)dq
s
=
¸
cos(θ)dθ
sin(θ)dθ
which has exact elements. Moreover, ω has an exact diﬀerential γ(θ) = [ sin(θ), −(cos(θ)−
1)]
T
that satisﬁes dγ(θ) = ω. Therefore, applying the following global change of co
ordinates
z
1
= x
1
− sin(θ)
z
2
= x
2
− cos(θ)ω
w
1
= y
1
+ (cos(θ) −1)
w
2
= y
2
− sin(θ)ω
ξ
1
= θ
ξ
2
= ω
(5.43)
158
transforms the dynamics of the VTOL aircraft into
˙ z
1
= z
2
˙ z
2
= −sin(ξ
1
)¯ u
1
=: v
1
˙ w
1
= w
2
˙ w
2
= cos(ξ
1
)¯ u
1
−g =: v
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
2
(5.44)
and decouples the translational and attitude dynamics. Here, ¯ u
1
= u
1
− ξ
2
2
is the
new control. Due to [35], the system in (5.44) is dynamic feedback linearizable. In
this work, we take a diﬀerent approach that makes use of a static state feedback and
backstepping procedure for global conﬁguration stabilization of the VTOL aircraft.
Clearly, (z
1
, w
1
) are two ﬂat outputs of the VTOL aircraft which are obtained auto
matically as the byproducts of the decoupling change of coordinates (see [29, 106, 105]
for more details on diﬀerential ﬂatness and ﬂat outputs).
Applying the linear state feedback v
1
= c
1
z and bounded feedback v
2
= c
0
σ(c
2
w)
where σ(s) = tanh(s) and 0 < c
0
< g, globally asymptotically and locally exponen
tially stabilizes (z, w) = (0, 0) for the (z, w)subsystem in (5.44) where c
1
, c
2
∈ R
2
are
coeﬃcients of a Hurwitz polynomial (a rather similar choice is made for v
1
in [80] but
with = 0 which is not the case here). An important point is that, the unbounded
choice of v
2
= c
2
w is not possible due to a singularity that appears later in control
design. To stabilize the origin for the composite system, we use the backstepping
procedure as follows. Solving for ¯ u
1
and tan ξ
1
in equation (5.44) gives
¯ u
1
= k
1
(z, w) :=
v
2
1
+ (v
2
+g)
2
tan ξ
1
= k
2
(z, w) =
−v
1
v
2
+ g
Then, deﬁning the change of coordinates and control
µ
1
= tan ξ
1
−k
2
(z, w)
µ
2
= (1 + tan
2
ξ
1
)ξ
2
−
˙
k
2
¯ u
2
= (1 + tan
2
ξ
1
)(u
2
+ 2 tan ξ
1
ξ
2
2
) −
¨
k
2
we get
˙ µ
1
= µ
2
˙ µ
2
= ¯ u
2
Thus, applying ¯ u
2
= −d
1
µ
1
− d
2
µ
2
with d
1
, d
2
> 0 globally exponentially stabilizes
(µ
1
, µ
2
) = (0, 0) for the µsubsystem. The dynamics of the closedloop system is in
the form
˙ η = f(η, µ)
˙ µ = Aµ
(5.45)
where η = col(z, w), A is a Hurwitz matrix. Given µ = 0, for ˙ η = f(η, 0), η = 0 is
globally asymptotically and locally exponentially stable. It can be shown that for any
159
solution of the µsubsystem the solution of the ηsubsystem is uniformly bounded and
the asymptotic stability of (η, µ) = 0 for the cascade system in (5.45) follows from a
theorem due to Sontag in [82]. This guarantees global asymptotic stability and local
exponential stability of the origin for the VTOL aircraft. Figure 527 (a) shows the
trajectories of the VTOL aircraft from initial condition (2, 3, 4, 1, π/3, 1) with = 0.1.
The path of the center of mass of the aircraft is shown in Figure 527 (b). Based on
the simulation result in Figure 527, the controller performs an aggressive maneuver
to stabilize the position and attitude of the aircraft. In the sense that, after a rather
short transient period, the states of the system almost exponentially converge to zero.
0 2 4 6 8 10 12
−1
0
1
2
3
4
5
time (sec)
x
y
roll
0 2 4 6 8 10 12
−4
−3
−2
−1
0
1
2
time (sec)
dx/dt
dy/dt
d(roll)/dt
−0.5 0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x
y
(a) (b)
Figure 527: (a) The state trajectories of the VTOL aircraft, and (b) The path of the
center of mass of the aircraft in (x
1
, y
1
)plane.
160
5.11 Trajectory Tracking and Stabilization of An
Autonomous Helicopter
This section is devoted to trajectory tracking and attitude stabilization nonlinear con
trol design for an autonomous helicopter capable of performing aggressive maneuvers.
A helicopter is an underactuated mechanical system with 6 DOF (i.e. the position
and the attitude in R
3
SO(3)) and 4 control inputs which includes the thrust of
the main rotor and 3 control moments in the body frame. As the dynamic model
of the helicopter, we use a model introduced in [43]. The reason for this choice is
that the helicopter model in [43] is rather accurate. In the sense that, both the eﬀect
of the body moments on the translational dynamics and the eﬀect of the thrust of
the main rotor on the attitude dynamics are taken into account. These two mutual
eﬀects complicate control design for the helicopter. To derive the diﬀerentially ﬂat
outputs [29, 106, 105] for this helicopter model, the eﬀects of the body moment on the
translational dynamics are ignored in [42, 43]. We present a method for decoupling
of the translational and attitude dynamics of the helicopter. As a byproduct, this
decoupling method provides the diﬀerentially ﬂat outputs for the accurate model of
the helicopter. In addition, the problem of global exponential tracking of feasible
trajectories of a helicopter is addressed.
We use a rotation matrix R ∈ SO(3) to represent the attitude of the helicopter.
This way, the singularities of the Euler angle parameterization can be avoided. In the
past, the use of a rotation matrix in SO(3) as the attitude, has proven to be eﬃcient
for attitude control and tracking of satellites [108, 16], spacecraft [17], and helicopters
[43, 31].
T
W
g
X
Y
Z
(x,R)
Figure 528: A helicopter.
5.11.1 Dynamic Model of A Helicopter
The dynamic model of the helicopter used in this work is based on a model of the
helicopter given in [43]. Let q = (x, R) ∈ R
3
SO(3) denote the conﬁguration of
the helicopter depicted in Figure 528. Here, R denotes that transformation from the
161
body frame to the reference frame and is deﬁned as
R = exp(ψˆ e
3
) exp(θˆ e
2
) exp(φˆ e
1
)
where e
i
’s are standard basis in R
3
and η = (φ, θ, ψ) denotes the vector of three Euler
angles associated with rotations around x, y, z axes. The angles φ, θ, ψ are called roll,
pitch, and yaw, respectively. The attitude kinematics is given by
˙
R = Rˆ ω
where ω = (ω
1
, ω
2
, ω
3
)
T
denotes the vector of angular velocities in the body frame
and ˆ ω (i.e. ˆ ω = ω ) is a skewsymmetric matrix
ˆ ω =
0 −ω
3
ω
2
ω
3
0 −ω
1
−ω
2
ω
1
0
¸
¸
The relation between η and ω is as
˙ η = Φ(η)ω
where Φ(η) is called the Euler matrix and is given by
Φ(η) =
1 sin φtan θ cos φtan θ
0 cos φ −sin φ
0 sin φsec θ cos φsec θ
¸
¸
(5.46)
The matrix Φ(η) has a singularity at θ = ±π/2. For this reason, we use R for the
tracking control design.
The equations of motion for the helicopter are as the following
˙ x = v
m˙ v = RF
b
˙
R = Rˆ ω
J ˙ ω = −ω Jω +τ
b
(5.47)
where F
b
, τ
b
denote the forces and moments in the body frame and (x, v) denote the
position and the velocity of the center of mass of the helicopter in the reference frame.
The forces and moments in (5.47) are in the form
F
b
= mgR
T
e
3
−Te
T
+ T
t
e
2
τ
b
= Q(T)e
T
+Dτ
(5.48)
where T is the thrust of the main rotor, T
t
is the thrust of the tail rotor, e
T
is a unit
162
vector along the shaft of the main rotor deﬁned by
e
T
(a, b) =
sin(a)
−cos(a) sin(b)
cos(a) cos(b)
¸
¸
(5.49)
with a, b denoting the pitch and roll tilt angles of the tip of the main rotor, τ =
(bT, aT, T
t
)
T
, D = diag(l
b
, l
a
, l
t
) is a constant matrix with positive diagonal elements,
and
Q(T) = Q
0
+ Q
1
T
1.5
with relatively small constants Q
0
, Q
1
> 0.
Assumption 5.11.1. Notice that for (a, b) = (0, 0), e
T
= e
3
. In addition, for
[a[, [b[ <1, e
T
can be approximated as
e
T
≈
a
−b
1
¸
¸
= e
3
+
a
−b
0
¸
¸
Since, a, b are small in practice, we use this approximation of e
T
instead of e
T
itself.
Moreover, we use the following approximation
Q(T)e
T
≈ Q(T)e
3
This is justiﬁed by the fact that if Q
0
, Q
1
, a, b, ∼ O(), then the ﬁrst two elements of
Q(T)e
T
are of the order O(
2
) and can be ignored.
The following simpliﬁed forces and body moments are directly used in [31] as
the original expressions of forces and moments. For the sake of clarity, we restate
the approximate forces and moments in the body frame under the approximation
assumption 5.11.1.
Corollary 5.11.1. Suppose assumption 5.11.1 holds and let u = Dτ ∈ R
3
. Then
F
b
, τ
b
take the following forms
F
b
= mgR
T
e
3
−Te
3
+ E u
τ
b
= Q(T)e
3
+ u
(5.50)
where
E =
0
2
0
1
0
3
0 0 0
¸
¸
;
1
= 1/l
b
,
2
= −1/l
a
, ,
3
= 1/l
t
Proof. The proof is elementary and by direct substitution.
Based on corollary 5.11.1, the ﬁrstlevel approximate model of the helicopter is in
163
the form
˙ x = v
m˙ v = mge
3
−(Re
3
)T + REu
˙
R = Rˆ ω
J ˙ ω = −ω Jω + Q(T)e
3
+ u
(5.51)
Under a further approximation, by setting E = 0
3×3
in 5.51, one obtains a secondlevel
approximate model of the helicopter as
˙ x = v
m˙ v = mge
3
−(Re
3
)T
˙
R = Rˆ ω
J ˙ ω = −ω Jω + Q(T)e
3
+ u
(5.52)
where the control input of the attitude dynamics u has no eﬀect on the translational
dynamics. This is the “approximate model” used in [42, 43, 42]. Here, our main goal
is to address trajectory tracking for model (5.51) where E = 0
3×3
.
5.11.2 Input Moment Decoupling for A Helicopter
Consider the ﬁrstlevel approximate model of the helicopter in (5.51) with input
moment u. In this section, we seek a diﬀerentially ﬂat output [29, 106, 105] in the
form y = x + γ(η) such that ¨ y is independent of u. If ¨ y is independent of u for all
η, we call y = x + γ(η) an exact ﬂat output. On the other hand, if y = x + γ(η) is
independent of u at η = η
0
, we call y = x +γ(η) a weak ﬂat output. In the following,
we show that there exists no exact ﬂat output y = x + γ(η) for the helicopter model
(5.51). Before doing so, let us present the dynamics of the helicopter using the Euler
angle parameterization and identify the shape variables of the helicopter.
The ﬁrstlevel approximate model of the helicopter in equation (5.51) can be
rewritten as
˙ x = v
m˙ v = mge
3
−n(η)T + R(η)Eu
˙ η = Φ(η)ω
J ˙ ω = −ω Jω + Q(T)e
3
+ u
(5.53)
where n(η) = R(η)e
3
and we refer to it as the unit normal vector. R(η) is given by
R(η) =
cos ψ −sin ψ 0
sin ψ cos ψ 0
0 0 1
¸
¸
cos θ 0 sin θ
0 1 0
−sin θ 0 cos θ
¸
¸
1 0 0
0 cos φ −sin φ
0 sin φ cos φ
¸
¸
or equivalently
R(η) =
cos θ cos ψ sin φsin θ cos ψ −cos φsin ψ cos φsin θ cos ψ + sin φsin ψ
cos θ sin ψ sin φsin θ sin ψ + cos φcos ψ cos φsin θ sin ψ −sin φcos ψ
−sin θ sin φcos θ cos φcos θ
¸
¸
164
Deﬁne Ψ(η) = Φ
−1
(η) and note that Ψ(η) can be calculated as
Ψ(η) =
1 0 −sin θ
0 cos φ cos θ sin φ
0 −sin φ cos θ cos φ
¸
¸
Denote the conﬁguration vector of the helicopter by q = (x, η). The Lagrangian of
the helicopter is in the form
L(q, ˙ q) =
1
2
mv
T
v +
1
2
ω
T
Jω + mgx
3
which can be rewritten as
L(x, η, ˙ x, ˙ η) =
1
2
¸
˙ x
˙ η
T
¸
mI
3
0
0 Ψ(η)
T
JΨ(η)
¸
˙ x
˙ η
+ mgx
3
(5.54)
The inertia matrix of the helicopter is a block diagonal matrix
M(η) = diag(mI
3
, Ψ(η)
T
JΨ(η))
Therefore, the Euler angles are the shape variables of a helicopter, or q
s
= (φ, θ, ψ)
and the position q
x
= x is the vector of external variables of a helicopter. We need
the following result to check whether a ﬂat output exists for a helicopter.
Proposition 5.11.1. The attitude dynamics of the helicopter can be expressed in the
form
M
s
(η)¨ η +C
s
(η, ˙ η) ˙ η = Ψ(η)
T
τ
b
(5.55)
with a welldeﬁned positive deﬁnite inertia matrix M
s
(η) = Ψ(η)
T
JΨ(η) (for θ =
±π/2) given by
M
s
(η) =
J
1
0 −J
1
s
θ
0 J
2
s
2
φ
+ J
3
c
2
φ
(J
2
−J
3
)c
θ
c
φ
s
φ
−J
1
s
θ
(J
2
−J
3
)c
θ
c
φ
s
φ
J
1
s
2
θ
+ c
2
θ
(J
2
s
2
φ
+ J
3
c
2
φ
)
¸
¸
(5.56)
(“c
θ
” and “s
θ
” denote cos(θ) and sin(θ), respectively) and the Coriolis and centrifugal
matrix
C
s
(η, ˙ η) = Ψ(η)
T
J
˙
Ψ(η, ˙ η) −Ψ
T
(η)[(Ψ(η) ˙ η)
∧
]JΨ(η) (5.57)
where J = (diag(J
1
, J
2
, J
3
), τ
b
= Q(T)e
3
+ u denotes the total body moment, and
∧
denotes the hat operation. In addition, for the case J
2
= J
3
(i.e. the antisymmetric
case) the shape variables of the helicopter are φ, θ; while given J
2
= J
3
(i.e. the
yzsymmetric case) the only shape variable of the helicopter is θ.
Proof. Let us ﬁrst rewrite the attitude dynamics as the following
¨ η =
˙
Φ(η, ˙ η)ω + Φ(η) ˙ ω
=
˙
Φ(η, ˙ η)Ψ(η) ˙ η −Φ(η)J
−1
ˆ ωJω + Φ(η)J
−1
τ
b
165
which multiplying both sides of the last equation by m
s
(η) = Ψ(η)
T
JΨ(η) gives the
attitude dynamics in the question with
C
s
(η, ˙ η) = Ψ(η)
T
JΨ(η)
˙
Φ(η, ˙ η)Ψ(η) −Ψ(η)
T
[(Ψ(η) ˙ η)
∧
]JΨ(η)
Noting that
˙
Ψ = Ψ(η)
˙
Φ(η, ˙ η)Ψ(η) ﬁnishes the proof of the ﬁrst part. The second
part of the result for the antisymmetric case follows from the fact that M
s
(η) is
independent of ψ. In addition, for the symmetric case where J
2
= J
3
, the inertia
matrix of the helicopter in (5.56) takes the form
M
s
(θ) =
J
1
0 −J
1
s
θ
0 J
2
0
−J
1
s
θ
0 J
1
s
2
θ
+J
2
c
2
θ
¸
¸
(5.58)
which does not depend on φ, ψ.
Due to the fact that the attitude dynamics in (5.55) is a fullyactuated mechanical
system for θ = ±π/2, it is exact feedback linearizable. For the sake of completeness,
this is stated as the following corollary.
Corollary 5.11.2. The collocated partially linearizing feedback law
τ
b
= JΨ(η)˜ u + Φ(η)
T
C
s
(η, ˙ η) ˙ η
which is invertible for θ = ±π/2, transforms the attitude dynamics in (5.55) into a
3dimensional doubleintegrator
˙ η = ν
˙ ν = ˜ u
with a new control ˜ u.
Following the line of theorem 4.3.1 and based on proposition 5.11.1, the dynamics
of the helicopter in (5.53) can be expressed as
m
xx
¨ q
x
−g
x
= F
r
(q
s
)T + F
x
(q
s
)u
m
ss
(q
s
)¨ q
s
+ h
s
(q
s
, ˙ q
s
) = F
s
(q
s
)Q(T)e
3
+ F
s
(q
s
)u
(5.59)
where (q
x
, q
s
) = (x, η) and
m
xx
= mI
3
m
ss
(q
s
) = Ψ
T
(η)JΨ(η)
h
s
(q
s
, ˙ q
s
) = C
s
(η, ˙ η) ˙ η
F
r
(q
s
) = −n(η)
F
x
(q
s
) = R(η)E
F
s
(q
s
) = Ψ(η)
T
Based on theorem 4.3.1, if the following vector of coupling oneforms
ω
xs
= m
−1
xx
F
x
(q
s
)F
−1
s
(q
s
)m
ss
(q
s
)dq
s
166
which can be rewritten as
ω
xs
= m
−1
R(η)EJΨ(η)dη (5.60)
has exact elements. Then, there exists an exact ﬂat output y = x + γ(η) such that
ω
xs
= dγ(η). Without loss of generality assume m = 1. The following result assures
that such an exact ﬂat output does not exist for a helicopter.
Proposition 5.11.2. There exists no exact ﬂat output y = x+γ(η) for the helicopter,
i.e. there exists no local diﬀeomorphism γ(η) : R
3
→R
3
such that ω
xs
= dγ(η).
Proof. Let f
i
(η) denote the ith column of R(η)EJΨ(η). Then, ω
xs
has exact elements
if and only if f
i
’s satisfy the following property
∂f
i
(η)
∂η
j
=
∂f
j
(η)
∂η
i
, ∀i, j = 1, 2, 3, i = j (5.61)
where (η
1
, η
2
, η
3
) = (φ, θ, ψ). By direct symbolic calculations in MATLAB, it has
been shown that there exists no pair (i, j) with i = j that satisﬁes condition (5.61)
(see Appendix B for the code that symbolicly generates the diﬀerence of the left and
right hand side of the equation (5.61)). Therefore, none of the elements of ω
xs
are
exact one forms.
The following proposition, provides the weak ﬂat outputs for the ﬁrstlevel approx
imate model of the helicopter using momentum decomposition procedure in chapter
4.
Proposition 5.11.3. Consider the dynamics of the helicopter as the following
˙ x = v
m˙ v = mge
3
−(R(η)e
3
)T + R(η)E¯ u
˙ η = ν
M
s
(η) ˙ ν = −C
s
(η, ν)ν + Ψ(η)
T
¯ u
(5.62)
where ¯ u = Q(T)e
3
+ u and it is assumed that EQ(T)e
3
≈ 0. Deﬁne the coupling
inertia µ(η) as the following
µ(η) = R(η)EJΨ(η) (5.63)
and let π = µ(η) ˙ η denote the nonintegrable coupling momentum between the body mo
ments and the translational dynamics of a helicopter. (The nonintegrability property
of π is due to proposition 5.11.2). Then, using momentum decomposition procedure
π can be represented as
π = π
l
+ π
e
where π
l
= G
l
(η) ˙ η denotes the integrable locked momentum associated with locked
167
conﬁguration ¯ η = 0 which is given by
π
l
=
0
2
J
2
cos θ −
3
J
3
sin ψ
1
J
1
cos φ 0
3
J
3
cos ψ
1
J
1
sin φ −
2
J
2
sin θ 0
¸
¸
˙
φ
˙
θ
˙
ψ
¸
¸
(5.64)
and π
e
denotes the nonintegrable error momentum that vanishes identically at η = ¯ η.
In addition, there exists a diﬀeomorphism γ(η) = (γ
φ
(η), γ
θ
(η), γ
ψ
(η))
T
satisfying
γ(0) = 0 where
γ
φ
(θ, ψ) =
2
J
2
sin θ +
3
J
3
(cos ψ −1)
γ
θ
(φ, ψ) =
1
J
1
sin(φ) +
3
J
3
sin ψ
γ
ψ
(φ, θ) = −
1
J
1
(cos φ −1) +
2
J
2
(cos θ −1)
(5.65)
such that ˙ γ = π
l
. In addition, the following output
y
∗
= mx −γ(η) (5.66)
is a weak ﬂat output (i.e. ¨ y
∗
does not depend on the body moment ¯ u at η = 0) that
satisﬁes
¨ y
∗
= mge
3
−(Re
3
)T + C
e
(η, ˙ η) ˙ η +
˜
G
e
(η)u
where
C
e
(η, ˙ η) = C
l
(η, ˙ η) −G
l
(η)M
−1
s
(η)C
s
(η, ˙ η)
C
l
(η, ˙ η) =
˙
G
l
G
e
(η) = G(η) −G
l
(η)
˜
G
e
(η) = G
e
(η)Ψ
−1
(η)J
−1
and both G
e
(η) and
˜
G
e
(η) vanish at η = ¯ η.
Proof. To obtain the locked momentum π
l
= G
l
(η) ˙ η from the original nonintegrable
momentum π = G(η) ˙ η, we use the momentum decomposition procedure as the fol
lowing. Let
R
1
= exp(φˆ e
1
), R
2
= exp(θˆ e
2
), R
3
= exp(ψˆ e
3
)
then, R = R
3
R
2
R
1
and G(η) can be written as
G(η) = R
3
R
2
R
1
EJΨ(η)
Consider the following matrices
G
1
= G(η)[
η=(φ,0,0)
, G
2
= G(η)[
η=(0,θ,0)
, G
3
(η)[
η=(0,0,ψ)
To ﬁnd π
l
, we need to calculate the ith column C
i
of G
i
for i = 1, 2, 3. This way each
C
i
is a function of a single variable. Note that Ψ(η) = Ψ(φ, θ). By deﬁnition, G
i
’s
168
can be rewritten as
G
1
(φ) = R
1
EJΨ(φ, 0), G
2
(θ) = R
2
EJΨ(0, θ), G
3
(ψ) = R
3
EJΨ(0, 0)
Setting G
l
(η) = [C
1
(φ), C
2
(θ), C
3
(ψ)] gives the locked momentum π
l
= G
l
(η) ˙ η in
the question (see Appendix B for the symbolic calculation of G
l
(η)). Clearly, π
l
is
integrable and ˙ γ = G
l
(η) ˙ η. Calculating ¨ y
∗
, we get
¨ y
∗
= mge
3
−(Re
3
)T + RE¯ u − ¨ γ
where
¨ γ =
˙
G
l
˙ η + G
l
(η)¨ η
= C
l
(η, ˙ η) ˙ η + (G(η) −G
e
(η))[M
−1
s
(η)Ψ(η)
T
¯ u −M
−1
s
(η)C
s
(η, ˙ η) ˙ η]
= [C
l
(η, ˙ η) −G
l
(η)M
−1
s
(η)C
s
(η, ˙ η)] ˙ η + (G(η) −G
e
(η))Ψ
−1
(η)J
−1
¯ u
= C
e
(η, ˙ η) ˙ η + RE¯ u −G
e
(η)Ψ
−1
(η)J
−1
¯ u
where
C
e
(η, ˙ η) = C
l
(η, ˙ η) −G
l
(η)M
−1
s
(η)C
s
(η, ˙ η)
Therefore, ¨ y
∗
takes the following form
¨ y
∗
= mge
3
−(Re
3
)T −C
e
(η, ˙ η) ˙ η + G
e
(η)Ψ
−1
(η)J
−1
¯ u
The weak ﬂat property of the output y
∗
follows from the fact that G
e
(η) vanishes at
η = ¯ η = 0.
Corollary 5.11.3. After applying the change of coordinates
z
1
= y
∗
, z
2
= ˙ y
∗
the dynamics of the helicopter in (5.62) transforms into
˙ z
1
= z
2
˙ z
2
= mge
3
−(Re
3
)T + ∆(η, ν, ¯ u)
˙ η = ν
M
s
(η) ˙ ν = −C
s
(η, ν)ν + Ψ(η)
T
¯ u
(5.67)
where the perturbation ∆ is given by
∆(η, ν, ¯ u) = −C
e
(η, ν)ν +
˜
G
e
(η)¯ u
In addition, assume ∃L
1
, L
2
> 0 : σ
max
(C
e
(η, ν)) ≤ L
1
ν, σ
max
(
˜
G
e
(η)) ≤ L
2
η, ∀η, ν.
Let ¯ u = K
a
(z, η, ν) be a state feedback such that (η(t), ν(t), ¯ u(t)) is exponentially
vanishing with a rate σ
1
> 0. Then, ∆(t) is exponentially vanishing with a rate
σ
2
= 2σ
1
> 0.
Proof. The proof is straightforward and is omitted.
169
Remark 5.11.1. Based on the preceding corollary, one can solve the tracking problem
for the unperturbed system with ∆ = 0 and then perform a Lyapunovbased robust
stability analysis for the overall system with a perturbation ∆.
5.11.3 Nontriangular Normal Form of a Helicopter
The following proposition provides a nontriangular normal form for the ﬁrstlevel
approximate model of a helicopter. This normal form will be used later in chapter
7 for stabilization of the helicopter without decoupling or ignoring the eﬀect of the
input moment on the translational dynamics.
Proposition 5.11.4. Deﬁne the state variables and control
z = col(x, v), ξ
1
= η, ξ
2
= ν, w = ˜ u (5.68)
and assume EQ(T)e
3
≈ 0
3
. Then, the ﬁrstlevel approximate model of the helicopter
can be expressed in the following nontriangular normal form quadratic in ξ
2
and aﬃne
in w
˙ z = f(z) + g
1
(ξ
1
)T + g
2
(ξ
1
, ξ
2
, ) + g
3
(ξ
1
, )w
˙
ξ
1
= ξ
2
˙
ξ
2
= w
(5.69)
where = (
1
,
2
,
3
)
T
and f, g
1
, g
2
, g
3
are deﬁned as the following
f(z) =
¸
v
ge
3
(5.70)
g
1
(ξ
1
) =
¸
0
3
−m
−1
(R(η)e
3
)
(5.71)
g
2
(ξ
1
, ξ
2
, ) =
¸
0
3
R(η)E()Φ(η)
T
C
s
(η, ν)ν
(5.72)
g
3
(ξ
1
, ) =
¸
0
3
R(η)E()JΨ(η)
(5.73)
with the property that g
2
, g
3
vanish at = (0, 0, 0)
T
.
Proof. The proof is by direct calculation and follows from proposition 5.11.1 and
corollary 5.11.2.
5.11.4 Feedback Linearization of the Unperturbed Model of
the Helicopter
The dynamics of the unperturbed model of the helicopter is in the following form
˙ x = v
m˙ v = mge
3
−(Re
3
)T
˙
R = Rˆ ω
J ˙ ω = −ω Jω +τ
b
(5.74)
170
Both the translational and the attitude dynamics of the model of the helicopter in
(5.74) can be linearized as
˙ x = v
˙ v = w
˙
R = Rˆ ω
˙ ω = τ
(5.75)
where w ∈ R
3
is the control of the (x, v)subsystem and τ ∈ R
3
is the new control for
the attitude dynamics. The relation between τ
b
and τ is given by
τ
b
= Jτ +ω Jω (5.76)
Based on equation (5.75), the relation between the new control w, the thrust T, and
the normal vector n = Re
3
is as the following
mge
3
−nT = mw
From the last equation, one can solve for T and n as
T = mge
3
−w
n =
ge
3
−w
ge
3
−w
(5.77)
Assume that the controller for the translational dynamics is a state feedback
w = k(x, v)
Then, the feedback law w = k(x, v) deﬁnes both T and n as the following
T(x, v) = mge
3
−k(x, v)
n
c
(x, v) =
ge
3
−k(x, v)
ge
3
−k(x, v)
(5.78)
We refer to n
c
= n
c
(x, v) as the control normal vector.
Remark 5.11.2. The reader should notice that the normal vector n = Re
3
is a function
of the attitude R, while the control normal vector n
c
(x, v) is a function of the position
and the velocity (for a stabilization problem).
For the stabilization problem, the state feedback k(x, v) is in the form
w = k(x, v) = −c
p
(x −x
0
) −c
d
v , c
p
, c
d
> 0
where (x
0
, 0) is a desired equilibrium point of the translational dynamics. In contrast,
for the tracking problem, one is interested in tracking a trajectory x
d
() in R
3
. The
control law for the tracking case takes the following form
w = k(˜ x, ˜ v) = −c
p
(x −x
d
) −c
d
(v − ˙ x
d
) + ¨ x
d
, c
p
, c
d
> 0
where ˜ x = x − x
d
and ˜ v = v − ˙ x
d
. Thus, for tracking purposes, the control normal
171
vector n
c
= n
c
(˜ x, ˜ v) is a function of the position and velocity errors. Notice that
the stabilization problem is a special case of the tracking problem with (x
d
, ˙ x
d
, ¨ x
d
) =
(x
0
, 0, 0). Therefore, from now on, by n
c
we mean
n
c
(˜ x, ˜ v) =
ge
3
−k(˜ x, ˜ v)
ge
3
−k(˜ x, ˜ v)
To achieve asymptotic tracking, the controller for the attitude dynamics must guar
antee that the normal vector n = Re
3
asymptotically tracks the control normal vec
tor n
c
= n
c
(˜ x, ˜ v). However, this does not uniquely determines the desired attitude
R ∈ SO(3) required for such an asymptotic tracking. This problem is resolved in the
next section.
5.11.5 Desired Control Attitude R
d
for Position Tracking
The desired control attitude is a rotation matrix R
d
∈ SO(3) satisfying R
d
e
3
= n
c
.
However, this only determines one column of R
d
and the other two columns need
to be determined from the trajectory x
d
and (possibly) the heading direction of the
vehicle. For doing so, let us denote the three columns of R
d
by
f
c
= R
d
e
1
b
c
= R
d
e
2
n
c
= R
d
e
3
(5.79)
We refer to f
c
, b
c
as the control forward vector and the control binormal vector,
respectively. This terminology is adopted from the names of the three unit vectors in
the Frenet frame for a curve in R
3
[26]. Given a trajectory x
d
() in R
3
, the control law
w = k(˜ x, ˜ v) determines n
c
. To specify R
d
, we need to determine f
d
which is a unit
vector pointing out towards the nose of the vehicle. Let t
d
(t) denote a unit vector
tangent to x
d
(t). Then, the tangent vector t
d
, given by
t
d
=
v
d
v
d

is deﬁned almost everywhere (a.e.) for v
d
= ˙ x
d
= 0. Depending on whether n
c
 t
d
and/or t
d
 e
3
, or not three situations might arise.
i) n
c
t
d
= ±1: In this case, there is welldeﬁned (n
c
, t
d
)plane. We deﬁne f
c
as a
unit vector in the (n
c
, t
d
)plane orthogonal to n
c
such that f
c
t
d
> 0. Therefore
f
c
=
t
d
−(n
c
t
d
)n
c
1 −(n
c
t
d
)
2
, n
c
t
d
= ±1
(where denotes the inner product of two vectors).
ii) n
c
t
d
= ±1 and e
3
t
d
= ±1: These two conditions imply that n
c
∦ e
3
, or
[n
c
e
3
[ < 1. For this case, we deﬁne f
c
as a unit vector in the (n
c
, e
3
)plane
172
such that f
c
⊥ n
c
. Thus
f
c
= −sgn(n
c
t
d
) sgn(t
d
e
3
)
e
3
−(n
c
e
3
)n
c
1 −(n
c
e
3
)
2
, n
c
t
d
= ±1, e
3
t
d
= ±1
iii) n
c
t
d
= ±1 and e
3
t
d
= ±1: This is equivalent to the case where n
c
 t
d
 e
3
and the vehicle moves in the vertical direction. For this case, we set
f
c
= sgn(n
c
t
d
)e
1
(e
1
in the last equation can be replaced by any other predetermined unit vector
which deﬁnes a desired heading direction of the vehicle).
Given f
c
and n
c
, the control binormal vector is deﬁned as
b
c
= f
c
n
c
Now, the three unit vectors f
d
, b
d
, n
d
determine the desired control attitude R
d
as
R
d
= [f
c
, b
c
, n
c
].
5.11.6 Asymptotic Tracking of a Desired Attitude R
d
∈ SO(3)
To do attitude tracking, we need to deﬁne a meaningful distance between two diﬀerent
attitudes in SO(3). Let R = [f, b, n], R
d
= [f
c
, b
c
, n
c
] denote the current and desired
attitudes, respectively. We deﬁne the Frenet distance between R, R
d
∈ SO(3) as
δ
F
(R, R
d
) = f −f
c

2
+b −b
c

2
+n −n
c

2
On the other hand, there is a wellknown function ϕ : SO(3) →R
≥0
deﬁned as
ϕ(R
0
) :=
1
2
tr(I −R
0
)
which has similar properties to a norm [16] (“tr” denotes the trace operator over
square matrices). In fact, ϕ satisﬁes the following triangulartype inequality:
ϕ(R
1
R
T
2
) + ϕ(R
2
R
T
3
) ≥ ϕ(R
1
R
T
3
), ∀R
1
, R
2
, R
3
∈ SO(3)
Deﬁning the distance δ as
δ(R
1
, R
2
) := ϕ(R
1
R
T
2
),
the triangular inequality over SO(3) takes a more familiar form
δ(R
1
, R
2
) + δ(R
2
, R
3
) ≥ δ(R
1
, R
3
)
In addition, based on Rodrigues’s formula, the rotation by θ around a unit vector k
is given by
R(k, θ) = I + sin θ
ˆ
k + (1 −cos θ)
ˆ
k
2
(5.80)
173
Thus
ϕ(R(k, θ)) = 1 −cos(θ) ≤ 2
In other words, ϕ : SO(3) →[0, 2]. We prove that the Frenet distance δ
F
(R, R
d
) and
ϕ(RR
T
d
) are equivalent upto a constant multiplicative factor. As a distance measure
between two SO(3) matrices R and R
d
, ϕ(RR
T
d
) has been eﬀectively used for satellite
attitude control [16] and helicopter motion planning [31].
Lemma 5.11.1. Let
δ(R, R
d
) := ϕ(RR
T
d
) =
1
2
tr(I −RR
T
d
)
Then δ and δ
F
satisfy the following identity
δ
F
(R, R
d
) = 4δ(R, R
d
)
Proof. We have
δ
F
(R, R
d
) =
¸
3
i=1
(R −R
d
)e
i

2
=
¸
3
i=1
e
T
i
(R −R
d
)
T
(R −R
d
)e
i
=
¸
3
i=1
e
T
i
(2I −RR
T
d
−R
T
d
R)e
i
= tr(I −RR
T
d
) + tr(I −(RR
T
d
)
T
)
= 2tr(I −RR
T
d
) = 4δ(R, R
d
)
From now on, we only use δ(R, R
d
) as a distance measure instead of the Frenet
distance δ
F
. Note that δ(R, R
d
) = 0 ⇔ R = R
d
. Since δ ≥ 0, the main idea in
attitude tracking is to design ω, as the control input for the attitude kinematics, such
that
˙
δ < 0, ∀δ > 0. Before presenting our control design for attitude tracking, we
need to deﬁne some notations. Let 'A, B` denote the inner product of two 3 3
matrices A, B deﬁned as
'A, B` :=
1
2
tr(A
T
B)
Then, for two skewsymmetric matrices ˆ α
1
, ˆ α
2
, we have
'ˆ α
1
, ˆ α
2
` = α
1
α
2
, 'ˆ α, ˆ α` = α
2
2
where denotes the inner product of vectors in R
3
. Notice that any matrix A can be
decomposed as
A = skew(A) + sym(A)
where
skew(A) =
A −A
T
2
, sym(A) =
A + A
T
2
Keeping this in mind, the main property of the distance δ(R, R
d
) is given by the
following lemma [16].
Lemma 5.11.2. Let ω
d
= (R
T
d
˙
R
d
)
∨
, then δ(R, R
d
) =
1
2
tr(I − RR
T
d
) satisﬁes the
174
following property
˙
δ = skew(RR
T
d
)
∨
(ω −ω
d
)
(
∨
over skewsymmetric matrices denotes the inverse of the hat operation ˆ).
Proof. By direct calculation, we have
˙
δ = −
1
2
tr(
d
dt
(RR
T
d
))
= −
1
2
tr(Rˆ ωR
T
d
+ Rˆ ω
T
d
R
T
d
)
= −
1
2
tr(R(ˆ ω − ˆ ω
d
)R
T
d
)
= −
1
2
tr(R
T
d
R(ˆ ω − ˆ ω
d
))
= −
1
2
tr((skew(R
T
d
R) + sym(R
T
d
R))(ˆ ω − ˆ ω
d
))
= −
1
2
tr(skew(R
T
d
R)(ˆ ω − ˆ ω
d
))
=
1
2
tr(skew(RR
T
d
)
T
(ˆ ω − ˆ ω
d
))
= 'skew(R
T
d
R), (ˆ ω − ˆ ω
d
)`
= skew(R
T
d
R)
∨
(ω −ω
d
)
and the result follows.
The following proposition is a direct result of lemma 5.11.2.
Proposition 5.11.5. The state feedback law
ω = K
R
(R, R
d
, ω
d
) := ω
d
−c
0
skew(R
T
d
R)
∨
, c
0
> 0
(almost) globally asymptotically stabilizes the equilibrium δ = 0 of
˙
δ = skew(R
T
d
R)
∨
(ω −ω
d
)
for all δ(0) = 2. In other words, ω = K
R
renders R = R
d
for
˙
R = Rˆ ω
(almost) globally asymptotically stable given that R(0) ∈ SO(3) ` ∂SO(3) where
∂SO(3) = ¦R ∈ SO(3) [ δ(R, R
d
) = 2¦
Proof. For the closedloop system
˙
δ = −c
0
skew(R
T
d
R)
∨

2
2
< 0
175
for δ > 0. Now, we need to show that skew(R
T
d
R)
∨
= 0 implies δ = 0. For doing so,
observe that skew(R
T
d
R)
∨
= 0 implies R
T
d
R is a symmetric matrix, or
R
T
d
R = (R
T
d
R)
T
∈ SO(3)
The only symmetric matrices in SO(3) are I and I + 2
ˆ
k
2
where k is an arbitrary
unit vector in R
3
(this follows from Rodrigues’s formula). But R
T
d
R = I + 2
ˆ
k
2
means δ = δ(R, R
d
) = 2 and this contradicts the assumption that δ(0) = 2. Hence,
R
T
d
R = I is the only acceptable solution for skew(R
T
d
R)
∨
= 0 which implies R = R
d
,
or δ = δ(R, R
d
) = 0. In other words, the only invariant equilibrium is δ = 0 that
proves (almost) global asymptotic stability of δ = 0 for δ(0) < 2. This is equivalent
to (almost) global asymptotic tracking of R = R
d
for R(0) ∈ ∂SO(3).
Remark 5.11.3. All the matrices in ∂SO(3) are obtained by rotation of R
d
around a
unit vector k by θ = π (see (5.80). Therefore, to satisfy R(0) ∈ ∂SO(3) in propo
sition 5.11.5, one can perturb the initial condition R(0) via multiplying R(0) by the
rotation matrix R(
ˆ
k, ) deﬁned in (5.80) that slightly rotates R(0). In practice, due
to measurement noise and roundoﬀ errors, it is unnecessary to apply such an initial
perturbation by a small rotation angle around k.
We restate the following lemma from [16] based on the notation and application
of interest in this work.
Lemma 5.11.3. ([16]) For any 0 < ≤ 2, there exist positive constants d
1
, d
2
> 0
such that ϕ(R
0
) ≤ 2 − implies
d
1
skew(R
0
)
∨

2
≤ ϕ(R
0
) ≤ d
2
skew(R
0
)
∨

2
This helps us to prove that the state feedback law in proposition 5.11.5 exponen
tially stabilizes R = R
d
.
Remark 5.11.4. This result in a more general form on Riemannian manifolds has been
proven by Bullo [16], however it is crucial to emphasize the result for the special case
of SO(3) matrices due to its important applications in attitude control of helicopters,
aircrafts, satellites, spacecraft, and underwater vehicles. In addition, in [16] addi
tional dissipation functions are used for attitude tracking/stabilization. Therefore,
the control laws obtained here are slightly diﬀerent and simpler than the ones given
in [16].
Proposition 5.11.6. The state feedback law
ω = K
R
(R, R
d
, ω
d
) := ω
d
−c
0
skew(R
T
d
R)
∨
, c
0
> 0
(almost) globally exponentially stabilizes R = R
d
for
˙
R = Rˆ ω
176
for all the initial conditions R(0) ∈ SO(3) ` ∂SO(3) where
∂SO(3) = ¦R ∈ SO(3) : δ(R, R
d
) = 2¦
Proof. For all R(0) ∈ SO(3)`∂SO(3), δ(0) = ϕ(RR
T
d
) < 2. Setting = (2−δ(0))/2,
based on lemma 5.11.3, we have
˙
δ = −c
0
skew(R
T
d
R)
∨

2
2
≤ −c
0
d
1
δ = −¯ c
0
δ
Therefore, δ = 0 is globally exponentially stable for all initial conditions δ(0) ∈
[0, 2 −
0
], 0 <
0
<1.
Remark 5.11.5. In the following, we provide the interpretation of the (almost) global
exponential attitude tracking control law in proposition 5.11.6 in terms of Rodrigues’s
rotation matrix R(θ, k). For doing so, let
R
0
(θ, k) := R
T
d
R
Then, δ = ϕ(R
T
d
R) = ϕ(R
0
(θ, k)) = 1 −cos(θ) and
˙
δ = skew(R
T
d
R)
∨
(ω −ω
d
)
where
skew(R
T
d
R)
∨
= skew(R
0
(θ, k))
∨
= sin(θ)k
Therefore, the control law in proposition 5.11.6 can be expressed as
ω = ω
d
−c
0
sin(θ)
ˆ
k
which means δ satisﬁes the following equation
˙
δ = −c
0
sin
2
(θ) = −c
0
(1 + cos(θ))δ
If for some 0 < < 1, ϕ(R
0
) ∈ [0, 2 − ], then [θ[ < π and there exist d
1
, d
2
∈ (0, 2]
such that
d
1
≤ 1 +cos(θ) ≤ d
2
= 2
As a result, we get
˙
δ ≤ −c
0
d
1
δ = ¯ c
0
δ
which guarantees (almost) global exponential stability of δ = 0. Apparently, if ini
tially δ(0) = ϕ(R
0
) = 2, θ = π and
˙
δ = 0. The stable solution in this case is
δ(t) = 2, ∀t > 0 and thus δ does not converge to zero. Furthermore, substituting
δ = 1 −cos(θ) in the diﬀerential equation of δ gives
˙
θ = −sin(θ), θ ∈ (−π, π)
which has an exponentially stable equilibrium point θ = 0. Two possible Lyapunov
functions for this system are V = θ or V = 1 −cos(θ) = δ. Here, we prefer the latter
177
one due to its direct relation with the distance function δ.
Having a stabilizing state feedback ω = K
R
, using the backstepping procedure,
the controller for the composite system can be obtained in a straightforward way.
Proposition 5.11.7. Consider the attitude dynamics of a helicopter as the following
˙
R = Rˆ ω,
˙ ω = τ
(5.81)
Then, the static state feedback law
τ = K
a
(ω, R, ω
d
, R
d
) := −c
1
(ω −ω
d
) −c
1
c
0
skew(R
T
d
R)
∨
+
˙
K
R
; c
0
, c
1
> 0 (5.82)
where
˙
K
R
= ˙ ω
d
−c
0
skew(R
T
d
Rˆ ω − ˆ ω
d
R
T
d
R)
∨
˙ ω
d
= (R
T
d
¨
R
d
− ˆ ω
2
d
)
∨
achieves (almost) global asymptotic tracking for the solution (R
d
, ω
d
) of the closedloop
system. In addition, the following positive deﬁnite function
V = 2c
0
c
1
δ(R, R
d
) +
1
2
ω −K
R

2
is a valid Lyapunov function for the system.
Proof. Calculating
˙
V , we get
˙
V = 2c
0
c
1
˙
δ −c
1
(ω −ω
d
+ c
0
skew(R
T
d
R)
∨
) (ω −ω
d
+ c
0
skew(R
T
d
R)
∨
)
= −c
1
c
2
0
skew(RR
T
d
)
∨

2
2
−c
1
ω −ω
d

2
2
< 0
for (R, ω) = (R
d
, ω
d
), because skew(R
T
d
R)
∨
 = 0 ⇔δ(R, R
T
d
) = 0. Thus, δ = 0 (i.e.
R = R
d
) and ω = ω
d
are (almost) globally asymptotically stable for the closedloop
system, given that δ(0) = 2.
Here is our main result for the trajectory tracking/stabilization (in position) for
an autonomous helicopter:
Theorem 5.11.1. Consider a trajectory x
d
() as a curve in R
3
with an associated
attitude R
d
(). Assume L
d
:= ¨ x
d

∞
< ∞ and denote z = col(˜ x, ˜ v). Then, there
exists a closed ball B
r
= ¦z ∈ R
6
[ z ≤ r¦ around z = 0 and a ﬁnite time t
0
> 0
such that given the state feedback law
τ
b
= ω Jω −J[c
1
(ω −ω
d
) + c
1
skew(R
T
d
R)
∨
− ˙ ω
d
+ c
0
skew(R
T
d
Rˆ ω − ˆ ω
d
R
T
d
R)
∨
]
T(˜ x, ˜ v) = mge
3
−k(x −x
d
, v −v
d
)
(5.83)
the tracking error z(t) in position and velocity, for the closedloop system, globally
exponentially converges to B
r
, i.e.
∃t
0
, λ
0
, d
0
> 0 : min
p∈Br
z(t) −p ≤ d
0
exp(−λ
0
(t −t
0
)), ∀t ≥ t
0
178
and remains in B
r
thereafter. In addition, the total tracking error (˜ x, ˜ v, δ, ˜ ω) (almost)
globally asymptotically converges to zero, given that δ(0) = 2.
The following theorem, introduces a highgain nonlinear controller for the approx
imate model of the helicopter with T, ω as the input. This controller achieves an
(almost) global exponential tracking of feasible trajectories for a helicopter.
Theorem 5.11.2. Consider a trajectory x
d
() as a curve in R
3
with an associated
attitude R
d
(). Assume L
d
:= ¨ x
d

∞
< ∞ and denote z = col(˜ x, ˜ v). Then, there exist
a ﬁnite time t
0
> 0 and a positive constant c
∗
> 0 such that given the state feedback
law
ω = ω
d
−c
0
skew(R
T
d
R)
∨
, c
0
> 0
T(˜ x, ˜ v) = mge
3
−k(x −x
d
, v −v
d
)
(5.84)
for all c
0
> c
∗
, after time t
0
, the tracking error (z(t), δ(t)) for the closedloop system,
(almost) globally exponentially converges to zero, given that δ(0) = 2.
We need the following lemma, before presenting the proof of the last two theorems.
Lemma 5.11.4. Assume L
d
= ¨ x
d

∞
< ∞ and denote z = col(˜ x, ˜ v). Then, there
exists a closed ball B
r
= ¦z ∈ R
6
: z ≤ r¦, r > 0 around z = 0 in the form and a
positive constant C
1
> 0 such that the thrust state feedback T(z) = T(˜ x, ˜ v) in (5.83)
satisﬁes the following properties:
i) T(z) < C
1
z, ∀z ∈ (R
6
` B
r
),
ii) T(z) ≤ C
1
r, ∀z ∈ B
r
.
Proof. Let c
m
:= max¦c
p
, c
d
¦ > 0. Since k(˜ x, ˜ v) = −c
p
˜ x − c
d
˜ v + ¨ x
d
and T(˜ x, ˜ v) =
mge
3
−k(˜ x, ˜ v), the thrust feedback T(z) = T(˜ x, ˜ v) satisﬁes the following inequality
T(z) ≤ m(g +L
d
) +
√
2c
m
z)
due to ˜ x +˜ v ≤
√
2z. Setting r = r(c
m
) := (g + L
d
)/(c
m
√
2), we get
T(z) ≤
1
2
C
1
(r +z)
where C
1
= 2
√
2mc
m
. Thus, for all z > r, T(z) < C
1
z and for all z ≤ r,
T(z) ≤ C
1
r.
Proof.(theorem 5.11.1) Denote ˜ x = x−x
d
and ˜ v = v −v
d
. The closedloop dynamics
of the helicopter in (5.74) can be written as the following
˙
˜ x = ˜ v
˙
˜ v = −ge
3
+ R
d
e
3
T(˜ x, ˜ v) + ∆
˙
R = Rˆ ω
˙ ω = u
179
where the perturbation ∆ ∈ R
3
is given by
∆ =
1
m
(R −R
d
)e
3
T(˜ x, ˜ v)
Calculating the upper bound of ∆, we get
∆ ≤
1
m
(R −R
d
)e
3
T(˜ x, ˜ v)
but (R −R
d
)e
3
 ≤ δ
F
(R, R
d
) = 4δ(R, R
d
) and therefore
∆ ≤ C
2
δ(R, R
d
)T(z)
where z = col(˜ x, ˜ v) and C
2
:= 4/m. Deﬁning the new perturbation
¯
∆ = col[0, ∆], the
closedloop translational dynamics in zcoordinates can be written as
˙ z = Az +
¯
∆ (5.85)
where
A =
¸
0
3
I
3
−c
p
I
3
−c
d
I
3
is a Hurwitz matrix and z ∈ R
6
. Consider the Lyapunov function V (z) = z
T
Pz where
P is the symmetric and positive deﬁnite solution of the Lyapunov equation
A
T
P +PA = −I
For all z ∈ B
r
, we have
˙
V = −z
2
+ 2z
T
P
¯
∆
≤ −z
2
+ 2λ
max
(P)C
2
zT(z)δ
< −z
2
(1 −2
√
2λ
max
(P)C
1
C
2
δ) (by lemma 5.11.4 and the property δ ≥ 0)
On the other hand, δ = 0 is exponentially stable for the attitude dynamics, thus
∃t
0
> 0 : δ(t) ∈ [0, r
0
], ∀t ≥ t
0
with
r
0
:=
1
4
√
2λ
max
(P)C
1
C
2
=
1
64c
m
λ
max
(P)
, c
m
= max¦c
p
, c
d
¦ > 0
This implies that for t ≥ t
0
,
˙
V < −0.5V and therefore any solution z(t) outside the
ball B
r
exponentially converges to B
r
and remains in it thereafter. Since with δ = 0,
¯
∆ = 0 and z = 0 is globally asymptotically stable for ˙ z = Az, from Sontag’s theorem
on stability of cascade systems (see [83], or theorem 7.3.3 in chapter 7), (z, δ, ˜ ω) = 0 is
(almost) globally asymptotically stable for the closedloop system given that δ(0) = 2.
Proof.(theorem 5.11.2) Based on the proof of theorem 5.11.1, after some ﬁnite time
t
0
> 0, the position and velocity tracking error z(t) enters a closed ball B
r
around
180
z = 0 exponentially fast and remains in it thereafter. From lemma 5.11.4, over B
r
we
have T(z) ≤ C
3
:= C
1
r. From proposition 5.11.6,
˙
δ ≤ −¯ c
0
δ where ¯ c
0
= c
0
d
1
and
d
1
> 0 is a constant. Deﬁning
W(z, δ) = V (z) +
1
2
δ
2
and keeping in mind that δ ≥ 0, we get
˙
W = −z
2
+ 2z
T
P
¯
∆ + δ
˙
δ
≤ −z
2
+ 2rC
1
λ
max
(P)zδ − ¯ c
0
δ
2
≤ −
1
2
z
2
−(¯ c
2
0
−2r
2
C
2
1
λ
2
max
(P))δ
2
where the last inequality is obtained from
kab ≤ la
2
+
k
2
4l
b
2
with a = z, b = δ, k = 2rC
1
λ
max
(P), l = 1/2. Notice that C
4
:= r(c
m
)C
1
(c
m
) does
not depend on c
m
and is a constant because
C
4
= rC
1
=
g + L
d
√
2c
m
(2
√
2mc
m
) = 2m(g +L
d
)
Taking
¯ c
0
≥ ¯ c
∗
0
:= (2C
2
4
λ
2
max
(P) +
1
4λ
max
(P)
)
1
2
or c
0
> c
∗
:= ¯ c
∗
0
/d
1
, we get
˙
W ≤ −
1
2
z
2
−
1
4λ
max
(P)
δ
2
≤ −
1
2λ
max
(P)
[λ
max
(P)z
2
+
1
2
δ
2
)
≤ −
1
2λ
max
(P)
W
and therefore W = 0, or (z, δ) = 0 is exponentially stable over B
r
. This implies
(z(t), δ(t)) (almost) globally exponentially converges to zero, after some ﬁnite time
t
0
> 0, given that δ(0) = 2.
5.11.7 Thrust for Exponential Attitude Stabilization
For exponential attitude stabilization regardless of how the position in (e
1
, e
2
)plane
changes, one has to use a diﬀerent control law for the thrust T than the one given
before (for tracking/stabilization of the position). This situation arises during doing
181
a maneuver from a (normal) hover mode to an (inverted) upsidedown hover mode.
This is equivalent to having no motions in the e
3
direction, thus
e
T
3
(Re
3
)T = mg
or
T = T(R) =
mg
e
T
3
Re
3
However, the denominator in the above ratio vanishes as Re
3
passes through a hori
zontal plane where the roll angle is ±π/2 and T →∞. This singularity is inherent in
the dynamics of the helicopter and is impossible to avoid regardless of any parame
terization of the attitude [31]. To avoid this problem, we suggest to apply a saturated
thrust given by
˜
T = T
max
sat(T/T
max
)
where T
max
1 is a prespeciﬁed maximum thrust. This results in a slight drop
of the height of the position of the helicopter in the e
3
direction during performing
this particular maneuver (or in general, any maneuver passing through θ = π/2 or
φ = π/2). Still, the main controller for stabilization of the attitude remains the same
as in proposition 5.11.7 (with ω
d
= 0). Notice that both T and τ are independent of
(x, v) in this case.
182
Chapter 6
Reduction and Control of
Underactuated Nonholonomic
Systems
This chapter is devoted to reduction and control of underactuated mechanical sys
tems with nonholonomic ﬁrstorder constraints and symmetry. A variety of reallife
control systems including Cartype Vehicles, Mobile Robots, Surface Vessels, and Au
tonomous Underwater Vehicles (AUV) are examples of underactuated control systems
with nonholonomic velocity constraints. All of these examples of nonholonomic sys
tems possess certain symmetry properties, i.e. their kinetic energy, potential energy,
or both are independent some conﬁguration variables. This motivated us to focus
mainly on reduction and control of nonholonomic systems with symmetry.
Our main theoretical result is that an underactuated system with ﬁrstorder non
holonomic constraints and kinetic symmetry can be reduced to a welldeﬁned reduced
order Lagrangian system in cascade with the constraint equation. Depending on
whether the number of control inputs and constraint equations is less than or equal
to the number of conﬁguration variables, the reduced Lagrangian system is fully
actuated or underactuated, respectively. We present a number of examples with de
tailed reduction process. Namely, a rolling disk, a mobile robot [5], a dynamic model
of a car, and the snakeboard [50]. Moreover, (almost) global exponential stabilization
and tracking for a twowheeled mobile robot is provided in great details.
Based on a result due to Brockett [15], there exists no smooth static state feedback
that asymptotically stabilizes a mechanical system with ﬁrstorder nonholonomic con
straints to the origin. As a consequence, a twowheeled nonholonomic mobile robot
(Figure 62), cannot be asymptotically stabilized to the origin using a smooth feed
back. Due to Astolﬁ [5], this mobile robot can be stabilized using a discontinuous
change of coordinates and a discontinuous static state feedback. Our contribution
is to introduce a smooth dynamic state feedback which achieves global exponential
stabilization/tracking to a point/trajectory that is far from the desired equilib
rium/trajectory ( < 1) for a twowheeled mobile robot. This is based on the use
of a new class of diﬀeomorphisms parameterized by λ that we call call nearidentity
diﬀeomorphism. A nearidentity diﬀeomorphism is equal to an identity function when
183
ever λ vanishes. In addition, for all λ, a nearidentity diﬀeomorphism remains within
a distance λ of an identity function. The dynamic state feedback is designed such
that λ(t) → as t →∞ and λ(0) ≈ O(1).
6.1 Nonholonomic Systems with Symmetry
This section is devoted to reduction of underactuated mechanical systems with non
holonomic ﬁrstorder constraints and symmetry. In the following, we precisely char
acterize a broad class of underactuated nonholonomic systems of interest by giving
three assumptions.
To begin, consider a Lagrangian system with an ndimensional conﬁguration vec
tor q, force matrix F(q), and m nonholonomic ﬁrstorder constraints as the following
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= W
T
(q)λ + F(q)τ
W(q) ˙ q = 0
(6.1)
where λ ∈ R
m
is the vector of Lagrange multipliers, τ ∈ R
l
is the control input,
l = rank(F(q)), and m + l ≤ n. Due to m ≥ 1, l ≤ (n − 1) and the nonholonomic
system in (6.1) is an underactuated system. The term W
T
(q)λ in (6.1) represents
the eﬀect of the constraint forces [64, 11, 8, 48]. This is based on the principle of
virtual work which states “the constraint forces do not work on motions allowed by
the constraints”. The principle of virtual work is an axiom of mechanics.
Remark 6.1.1. The last condition that m + l ≤ n is one of the main conditions that
distinguishes this work from the result of Bloch et al. in [11] on reduction of Caplygin
control systems. In that work, it is assumed that m + l ≥ n. Later, we show this
assumption leads to overactuated or fullyactuated reduced systems.
Assumption 6.1.1. Assume W(q) has full row rank. Then, q can be partitioned
as (q
1
, q
2
) such that W(q) = (W
1
(q), W
2
(q)) where W
1
(q) is an invertible matrix.
Therefore, the constraint equation in (6.1) can be rewritten as
W
1
(q) ˙ q
1
+ W
2
(q) ˙ q
2
= 0 (6.2)
Assumption 6.1.2. Assume M(q), F(q), W(q) are all independent of q
1
.
Assumption 6.1.3. Assume the potential energy is in the form
V (q) = k
T
q
1
+ U(q
2
) (6.3)
where k ∈ R
n−m
is a constant.
Remark 6.1.2. Under Assumption 6.1.3 with k = 0, the notions of kinetic symmetry
and classical symmetry coincide. Therefore, by saying “symmetry”, we refer to both
of these notions.
184
Remark 6.1.3. Broad classes of mechanical systems with nonholonomic velocity con
straints including mobile robots, cartype vehicles, surface vessels, rolling disk, Caply
gin control systems [64, 11], and the snakeboard system [50, 9, 49, 73, 74] satisfy
Assumptions 6.1.1 through 6.1.3. Moreover, all the aforementioned examples of non
holonomic systems are underactuated.
Under assumptions 6.1.1 through 6.1.3, the dynamics of the underactuated non
holonomic system in (6.1) can be expressed as
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= W
T
(q
2
)λ + F(q
2
)τ
W
1
(q
2
) ˙ q
1
+ W
2
(q
2
) ˙ q
2
= 0
(6.4)
Nonholonomic systems in the form (6.4) (with F(q) ≡ 0) are called Caplygin systems
[64]. Caplygin control systems in the form (6.4) were ﬁrst introduced by Bloch et
al. in [11]. To eliminate λ from (6.4), one can multiply both sides of the forced
EulerLagrange equation in (6.4) by a matrix A(q) that annihilates W
T
(q) [11], i.e.
A(q)W
T
(q) = 0. For doing so, let us deﬁne
w
12
(q
2
) = −W
−1
1
(q
2
)W
2
(q
2
)
then ˙ q = D(q
2
) ˙ q
2
where
D(q
2
) =
¸
w
12
(q
2
)
I
n−m
By direct calculation, it can be readily shown that A(q
2
) = D
T
(q
2
) annihilates
W
T
(q
2
). This eventually leads to the following reduction theorem for underactuated
nonholonomic systems with symmetry.
Theorem 6.1.1. Consider the underactuated mechanical control system with non
holonomic constraints and symmetry in (6.4). Then, system (6.4) with (2n + m)
ﬁrstorder equations can be reduced to a system of (2n − m) ﬁrstorder equations in
the following cascade form
˙ q
x
= w
r
(q
r
) ˙ q
r
M
r
(q
r
)¨ q
r
+ C
r
(q
r
, ˙ q
r
) ˙ q
r
+ G
r
(q
r
) = F
r
(q
r
)τ
(6.5)
where (q
x
, q
r
) = (q
1
, q
2
), w
r
(q
r
) = w
12
(q
2
), and
M
r
(q
r
) = D(q
2
)
T
M(q
2
)D(q
2
)
G
r
(q
r
) = w
T
r
(q
r
)k +∇
qr
U(q
r
)
F
r
(q
r
) = D
T
(q
2
)F(q
2
)
In addition, if V (q) = U(q
2
) (or k = 0), the reduced system is a welldeﬁned La
grangian system with conﬁguration vector q
r
= q
2
and Lagrangian function
L
r
(q
r
, ˙ q
r
) =
1
2
˙ q
T
r
M
r
(q
r
) ˙ q
r
−U(q
r
) (6.6)
185
which satisﬁes a forced EulerLagrange equation in cascade with the constraint equa
tion as the following
˙ q
x
= w
r
(q
r
) ˙ q
r
d
dt
∂L
r
∂ ˙ q
r
−
∂L
r
∂q
r
= F
r
(q
r
)τ
(6.7)
Moreover, if l + m < n (or l + m = n), the reduced system with conﬁguration vector
q
r
is an underactuated (or a fullyactuated) mechanical control system.
Proof. The forcedEulerLagrange equation in (6.4) can be rewritten as
M(q
2
)¨ q + C(q
2
, ˙ q) ˙ q + G(q) = W
T
(q
2
)λ + F(q
2
)τ (6.8)
where C(q
2
, ˙ q) satisﬁes
˙
M = C + C
T
and G(q) = ∇
q
V (q). Multiplying both sides of
the last equation by A(q
2
) = D
T
(q
2
) eliminates λ and gives
D
T
(q
2
)M(q
2
)¨ q + D
T
(q
2
)C(q
2
, ˙ q) ˙ q + G
r
(q) = F
r
(q
2
)τ (6.9)
where F
r
(q
2
) = D
T
(q
2
)F(q
2
) and
G
r
(q) = D
T
(q
2
)G(q) =
w
T
12
(q
2
) I
¸
∇
q
1
V (q)
∇
q
2
V (q)
= w
T
12
(q
2
)k +∇
q
2
U(q
2
) =: G
r
(q
2
)
On the other hand, ˙ q = D(q
2
) ˙ q
2
implies that
¨ q = D(q
2
) ˙ q
2
+
˙
D(q
2
, ˙ q
2
) ˙ q
2
Substituting ¨ q from the last equation in (6.9), we get
D
T
(q
2
)M(q
2
)D(q
2
)¨ q
2
+[D
T
(q
2
)C(q
2
, ˙ q)D(q
2
) +D
T
(q
2
)M(q
2
)
˙
D] ˙ q
2
+G
r
(q
2
) = F
r
(q
2
)τ
which after renaming q
r
= q
2
can be equivalently rewritten as
M
r
(q
r
)¨ q
r
+ C
r
(q
r
, ˙ q
r
) ˙ q
r
+ G
r
(q
r
) = F
r
(q
r
)τ (6.10)
where
M
r
(q
r
) := D
T
(q
2
)M(q
2
)D(q
2
) (6.11)
C
r
(q
r
, ˙ q
r
) := D
T
(q
2
)C(q
2
, D(q
2
) ˙ q
2
)D(q
2
) + D
T
(q
2
)M(q
2
)
˙
D(q
2
, ˙ q
2
) (6.12)
To establish that (6.10) is in fact equivalent to the forced EulerLagrange equation
for the reduced system with the Lagrangian function L
r
(q
r
, ˙ q
r
) and the force matrix
F
r
(q
r
), we need to prove C
r
(q
r
, ˙ q
r
) satisﬁes
˙
M
r
= C
r
+C
T
r
. By direct calculation, we
186
have
˙
M
r
= D
T
˙
MD +
˙
D
T
MD + D
T
M
˙
D
= D
T
(C + C
T
)D +
˙
D
T
M
˙
D + D
T
M
˙
D
= (D
T
CD + D
T
M
˙
D) + (D
T
C
T
D +
˙
D
T
MD)
= C
r
+ C
T
r
and the result follows.
Remark 6.1.4. Since the Lagrangian reduced system in (6.7) with the reduced con
ﬁguration vector q
r
and the reduced Lagrangian L
r
is itself an underactuated system
for m + l < n, its reduction process can be addressed based on the reduction theory
developed in Chapter 4 of this thesis. Also, the case of m + l = n is presented in
corollary 6.1.1.
Many wellstudied cases of nonholonomic systems including the rolling disk, a one
legged hopping robot in freeﬂying mode [51, 63], a twowheeled mobile robot [5], and
a planar threelink diver [22, 32] are special examples of underactuated nonholonomic
systems considered in theorem 6.1.1 which all possess a fullyactuated reduced system.
We formalize this special case in the following corollary (which has also been found
independently in [11]).
Corollary 6.1.1. Consider the underactuated nonholonomic system in (6.4). Assume
the number of inputs and constraints add up to n, i.e. m + l = n. Suppose that q
2
is actuated and the system has noninteracting inputs, i.e. F(q) = col(0, F
2
(q
2
))
where F
2
(q
2
) is an invertible matrix. Then, there exists a change of control input that
transforms the dynamics of (6.5) (or (6.4)) into the following nontriangular normal
form with a vector doubleintegrator linear part
˙ z = f(ξ
1
, ξ
2
) := w
r
(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(6.13)
Proof. Notice that F
r
(q
r
) = D
T
(q
2
)F(q
2
) = F
2
(q
2
) and thus F
r
is invertible. After
renaming the variables as z = q
x
, ξ
1
= q
r
, ξ
2
= ˙ q
r
and applying the change of control
τ = F
−1
r
(q
r
)(M
r
(q
r
)u +C
r
(q
r
, ˙ q
r
) ˙ q
r
+ G
r
(q
r
))
we get ¨ q
r
= u and the result follows.
6.2 Applications to Nonholonomic Robots
In this section, we provide applications of the theory developed in the preceding sec
tion for reduction and control of several examples of nonholonomic systems including
Caplygin systems and mobile robots.
187
6.2.1 A Rolling Disk
A vertical rolling disk (see Figure 61) which is not allowed to slip is an example of
mechanical system with nonholonomic velocity constraints. These constraints can be
expressed as
˙ x = r cos(θ) ˙ ϕ
˙ y = r sin(θ) ˙ ϕ
To view the rolling disk as a control system, assume the rotation angle ϕ and the
r
ϕ
θ
x
y
z
(x,y)
Figure 61: A vertical rolling disk
heading angle θ of the disk can be controlled using torques τ
1
and τ
2
, respectively.
The Lagrangian of the rolling disk with conﬁguration vector q = (x, y, ϕ, θ) is
L =
1
2
˙ q
T
M ˙ q
with a constant inertia matrix
M = diag(m, m, J
1
, J
2
)
where m is the mass of the disk and J
1
and J
2
are the inertia of the disk. Thus, the
rolling disk is a ﬂat underactuated mechanical system with actuated variables (ϕ, θ)
and unactuated variables (x, y). The constraint equation is in the form W(q) ˙ q = 0
with
W(q) =
¸
1 0 −r cos(θ) 0
0 1 −r sin(θ) 0
Apparently W(q) has full row rank. Setting q
1
= (x, y) and q
2
= (ϕ, θ), one obtains
W
1
(q
2
) = I
2×2
which is an invertible matrix. Since the number of controls and
constraints add up to n = 4, based on theorem 6.1.1, the reduced system for this
underactuated nonholonomic system is a fully actuated system. The conﬁguration
of the reduced system is q
r
= (ϕ, θ) and the overall dynamics of the system can be
188
expressed as
˙ x = r cos(θ) ˙ ϕ
˙ y = r sin(θ) ˙ ϕ
¨ ϕ = u
1
¨
θ = u
2
(6.14)
where u
i
= τ
i
/J
i
, i = 1, 2. The reduced system of the rolling disk is a vector double
integrator ¨ q
r
= u (u = (u
1
, u
2
)
T
). Controllability of the vertical rolling disk has been
addressed in [11, 49].
Assume the input controls of the system are ˙ ϕ = ω
1
and
˙
θ = ω
2
. Then, after
normalization of (x, y) units by r, the kinematic model of the rolling disk can be
written as
˙ x = cos(θ)ω
1
˙ y = sin(θ)ω
1
˙
θ = ω
2
(6.15)
Applying the following change of coordinates and control for θ ∈ (−π/2, π/2)
x
1
= x
x
2
= tan(θ)
x
3
= y
v
1
= ω
1
/ cos(θ)
v
2
= ω
2
/(1 + tan
2
(θ)
the kinematics of the rolling disk in (6.15) transforms into a ﬁrstorder chainedtype
nonholonomic system as the following
˙ x
1
= v
1
˙ x
2
= v
2
˙ x
3
= x
2
v
1
(6.16)
This is a special case of the general chainedtype system in (2.8). Exponential sta
bilization of nonholonomic systems in the chainedform (6.16) using a discontinuous
state feedback is addressed in [4]. Also, a homogeneous timevarying state feedback
is used in [60, 59] for local exponential stabilization of such chainedform systems.
6.2.2 A Nonholonomic Mobile Robot
Consider the mobile robot depicted in Figure 62 [5, 4]. The robot has two rolling
wheels that can be controlled independently using input torques. It is not diﬃcult
to see that the dynamics of this mobile robot is exactly the same as the vertical
rolling disk in equation (6.14). Clearly, the same arguments hold for the mobile robot
and using a change of coordinates and control the dynamics of this mobile robot
can be transformed into the chainedform system (6.16). Exponential stabilization
of this mobile robot using a discontinuous change of coordinates and a discontinu
ous state feedback (both with singularities at the origin) has been addressed in [5].
Other researchers have tried timevarying and discontinuous state feedback or both
189
θ
(x,y)
τ
1
x
y
τ
2
Figure 62: A mobile robot
to stabilize similar classes of nonholonomic systems [76, 19, 60]. Based on a theorem
due to Brockett [15], systems with ﬁrstorder nonholonomic constraints cannot be
asymptotically stabilized using a C
1
smooth state feedback.
We propose a new approach for tracking and stabilization of this mobile robot us
ing a smooth timevarying change of coordinates and a smooth dynamic state feedback
law that globally exponentially stabilizes the system to an equilibrium point arbitrary
close to the origin (i.e. within a distance of the desired equilibrium for <1). To
make this more precise, we need to deﬁne new notions of asymptotic stability and
tracking. This is done in section 6.3. The details regarding stabilization and tracking
for the twowheeled nonholonomic robot are presented in section 6.4.
6.2.3 A Car
In this section, we address reduction of a dynamic model of a car as shown in Fig
ure 63. The kinematic model of a car is a more common model of a car that has
been studied in the literature [63]. This includes steering using sinusoids [63] and
stabilization using discontinuous state feedback of the type used for a twowheeled
mobile robot [4]. Here, we take a diﬀerent approach that later allows us to apply
our reduction result for a car, to locomotion control design for the snakeboard [50].
The snakeboard is a complex nonholonomic system with certain similarities to a car
and will be discussed in the next section. A dynamic model of a car is an example
of an underactuated nonholonomic system with ﬁve degrees of freedom, two con
trols, and two velocity constraints corresponding to the rear and front wheels. Let
q = (x, y, θ, ψ, φ) ∈ SE(2) S
1
S
1
denote the conﬁguration vector of a dynamic
car. The position of the middle of the rear wheels of the car is denoted by (x, y), the
orientation of the body of the car is denoted by θ, the angle of rotation of each wheel
is denoted by ψ, and the orientation of the front wheels w.r.t. the body is denoted
by φ. The velocity constraints for the front and rear wheels are as the following
sin(θ + φ)
d
dt
(x +l cos θ) −cos(θ +φ)
d
dt
(x +l sin θ) = 0 (6.17)
sin(θ) ˙ x −cos(θ) ˙ y = 0 (6.18)
190
θ
φ
ψ
x
y
(x,y)
Figure 63: The model of a car.
or equivalently
sin(θ + φ) ˙ x −cos(θ +φ) ˙ y −l cos φ
˙
θ = 0 (6.19)
sin(θ) ˙ x −cos(θ) ˙ y = 0 (6.20)
These two constraints can be rewritten as W(q) ˙ q = 0 where W = (W
1
, W
2
) is par
titioned according to q
x
= (x, y) and q
r
= (θ, ψ, φ). The matrices W
1
, W
2
are given
by
W
1
(θ, φ) =
¸
sin(θ + φ) −cos(θ +φ)
sin θ −cos θ
, W
2
(θ, φ) =
¸
−l cos φ 0 0
0 0 0
,
Notice that W
1
(θ, φ) is not invertible at φ = 0. We have
w
12
(θ, φ) = −W
−1
1
W
2
=
¸
ρ
1
0 0
ρ
2
0 0
where ρ
1
, ρ
2
are deﬁned as
ρ
1
(θ, φ) =
l cos θ
tan φ
ρ
2
(θ, φ) =
l sin θ
tan φ
(6.21)
Thus, from w
12
(q), D(q) can be obtained as the following
D(q) = D(θ, φ) =
ρ
1
0 0
ρ
2
0 0
1 0 0
0 1 0
0 0 1
¸
¸
¸
¸
¸
¸
191
Later, we use D(q) to calculate the reduced inertia matrix M
r
(q
r
). The Lagrangian
of a dynamic car is as the following
L =
1
2
m[( ˙ x −
l
2
sin(θ)
˙
θ)
2
+ ( ˙ y +
l
2
cos(θ)
˙
θ)
2
]
+
1
2
(J
b
+ 2J
v
)
˙
θ
2
+
1
2
[(2J
h
)(1 +
1
cos
2
(φ)
)]
˙
ψ
2
+
1
2
(2J
v
)(
˙
θ +
˙
φ)
2
where m is the mass of the car, J
b
is the inertia of the body, J
h
is the inertia of each
wheel along a horizontal axes, and J
v
is the inertia of each wheel along the vertical
axes. After some simpliﬁcation, the Lagrangian can be expressed as
L =
1
2
˙ q
T
m 0 −(ml/4) sin θ 0 0
0 m (ml/4) cos θ 0 0
−(ml/4) sin θ (ml/4) cos θ J
θ
0 2J
v
0 0 0 2J
h
(2 + tan
2
(φ)) 0
0 0 2J
v
0 2J
v
¸
¸
¸
¸
¸
¸
˙ q
(6.22)
where J
θ
= J
b
+ 4J
v
+ ml
2
/4. The diﬀerentialalgebraic equations of motion for the
dynamic car are as the following
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= W
T
(q)
¸
λ
1
λ
2
+
¸
0
3×2
I
2×2
¸
τ
1
τ
2
W(q) ˙ q = 0
(6.23)
where λ
1
, λ
2
∈ R are Lagrange multipliers and τ
1
, τ
2
∈ R are torques applied to
the rear wheels and the steering wheel, respectively. Based on theorem 6.1.1, the
dynamics of the car in (6.23) can be reduced to the cascade of the constraint equation
and a reducedorder Lagrangian system with conﬁguration vector q
r
= (θ, ψ, φ) as the
following
M
r
(q
r
)¨ q
r
+ C
r
(q
r
, ˙ q
r
) ˙ q
r
= F
r
(q
r
)τ
where
M
r
(q
r
) = D
T
(q)M(q)D(q)
C
r
(q
r
, ˙ q
r
) = D
T
C(q, ˙ q)D + D
T
M(q)
˙
D
F
r
(q
r
) = D
T
(q)F(q)
By direct calculation, we have
F
r
(q
r
) = F
r
=
0 0
1 0
0 1
¸
¸
192
and
M
r
(φ) =
˜
J
θ
(φ) 0 2J
v
0 2J
h
(2 + tan
2
(φ)) 0
2J
v
0 2J
v
¸
¸
with
˜
J
θ
(φ) = J
θ
+m(ρ
2
1
+ρ
2
2
) −
ρ
1
ml
2
sin(θ) +
ρ
2
ml
2
cos(θ) = J
θ
+
ml
2
tan
2
(φ)
Clearly, the reduced Lagrangian system is itself an underactuated system with three
degrees of freedom (θ, ψ, φ) and two controls. In addition, (θ, ψ) are the external
variables and φ is the shape variable of the car. The dynamics of the actuated
variables (φ, ψ) of the reduced system can be linearized as
¨
ψ = u
1
¨
φ = u
2
using an explicit collocated change of control in the form
τ = α(φ)u + β(φ, ˙ q
r
)
where
α(φ) =
2J
h
(2 + tan
2
(φ)) 0
0 2J
v
(1 −
2J
v
tan
2
(φ)
ml
2
+ J
θ
tan
2
(φ)
)
¸
¸
is welldeﬁned and positive deﬁnite for all φ. From the ﬁrst constraint equation, one
can solve for
˙
θ to get
˙
θ = tan(φ)(cos(θ) ˙ x + sin(θ) ˙ y)
From this equation, the overall dynamics of the car can be written as
˙ x = r cos(θ)
˙
ψ
˙ y = r sin(θ)
˙
ψ
˙
θ =
r
l
tan(φ)
˙
ψ
¨
ψ = u
1
¨
φ = u
2
(6.24)
where r is the radius of the wheel. After normalization of the units of (x, y) by r, the
kinematic model of the car can be expressed as
˙ x = cos(θ)ω
1
˙ y = sin(θ)ω
1
˙
θ =
1
l
tan(φ)ω
1
˙
φ = ω
2
(6.25)
193
where
˙ ω
1
= u
1
, ˙ ω
2
= u
2
Now, applying the following change of coordinates and control [63]
x
1
= x
x
2
=
tan(φ)
l cos(θ)
3
x
3
= tan(θ)
x
4
= y
v
1
= cos(θ)ω
1
v
2
=
(1 + tan
2
(φ))
l cos(θ)
3
ω
2
+
3 tan(θ) tan
2
(φ)
l
2
cos(θ)
3
ω
1
(6.26)
transforms the nonholonomic system in (6.25) into a chainedform system as the
following
˙ x
1
= v
1
˙ x
2
= v
2
˙ x
3
= x
2
v
1
˙ x
4
= x
3
v
1
(6.27)
The system in (6.27) can be stabilized using a discontinuous state feedback [4, 47],
or sinusoid inputs [63].
6.2.4 The Snakeboard
The snakeboard system depicted in Figure 64 was ﬁrst introduced in [50]. In that
work, an experimental study was performed for locomotion of the snakeboard using
diﬀerent “gait” settings (i.e. a set of sinusoid inputs). The snakeboard example was
one of the main motivations for developing a general framework for reduction of non
holonomic systems with symmetry on Riemannian manifolds in [9]. Controllability
and motion planning for the snakeboard example as a control system was also consid
ered in [49] and [73, 74], respectively. Here, we show that the snakeboard system can
be reduced to an underactuated Lagrangian system in cascade with the constraint
equation. Then, we use this representation to prove that locomotion control of the
snakeboard is equivalent to stabilization of a kinematic car.
Dynamics and Reduction of the Snakeboard
The conﬁguration vector for the snakeboard is q = (x, y, θ, ψ, φ
1
, φ
2
) ∈ R
2
S
4
1
.
The velocity constraints for the front and back wheels of the snakeboard are as the
following
d
dt
(x + l cos θ) sin(θ + φ
1
) −
d
dt
(x + l sin θ) cos(θ +φ
1
) = 0
d
dt
(x −l cos θ) sin(θ −φ
2
) −
d
dt
(x −l sin θ) cos(θ −φ
2
) = 0
194
φ
ψ
φ
θ
2
1
l
x
y
Figure 64: The Snakeboard.
These constraints can be simpliﬁed as
sin(θ + φ
1
) ˙ x −cos(θ + φ
1
) ˙ y −l cos(φ
1
)
˙
θ = 0 (6.28)
sin(θ −φ
2
) ˙ x −cos(θ −φ
2
) ˙ y + l cos(φ
2
)
˙
θ = 0 (6.29)
Therefore, the constraints are in the form W(q) ˙ q = 0 with the constraint matrix W(q)
given by
W(q) =
¸
sin(θ + φ
1
) −cos(θ + φ
1
) −l cos(φ
1
) 0 0 0
sin(θ −φ
2
) −cos(θ −φ
2
) l cos(φ
2
) 0 0 0
Partitioning q as q
1
= (x, y) and q
2
= (θ, ψ, φ
1
, φ
2
) and deﬁning the vector η =
(θ, φ
1
, φ
2
) (i.e. all the variables that appear in W(q)), the constraint matrix W(q)
can be partitioned as (W
1
, W
2
) where W
1
and W
2
are deﬁned as
W
1
(η) =
¸
sin(θ + φ
1
) −cos(θ + φ
1
)
sin(θ −φ
2
) −cos(θ −φ
2
)
, W
2
(η) =
¸
−l cos(φ
1
) 0 0 0
l cos(φ
2
) 0 0 0
and W
1
(η) is invertible except for at isolated points φ
1
+ φ
2
= kπ, k ∈ Z. Assume
φ
1
+ φ
2
= kπ. By direct calculation, we have
w
12
(η) = −W
−1
1
W
2
=
¸
ρ
1
0 0 0
ρ
2
0 0 0
where
ρ
1
=
l cos φ
1
cos(θ −φ
2
) + l cos φ
2
cos(θ + φ
1
)
sin(φ
1
+ φ
2
)
ρ
2
=
l cos φ
1
sin(θ −φ
2
) +l cos φ
2
sin(θ + φ
1
)
sin(φ
1
+ φ
2
)
(6.30)
195
Hence
D(q
2
) = D(η) =
¸
w
12
(η)
I
4×4
=
ρ
1
0 0 0
ρ
2
0 0 0
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
¸
¸
¸
¸
¸
¸
¸
¸
It is assumed that the torque τ = (τ
1
, τ
2
, τ
3
)
T
∈ R
3
is applied to variables ψ, φ
1
, φ
2
and the rest of variables are unactuated. The Lagrangian of the snakeboard is in the
form
L(q, ˙ q) =
1
2
m( ˙ x
2
+ ˙ y
2
) +
1
2
J
b
˙
θ
2
+
1
2
J
r
(
˙
ψ +
˙
θ)
2
+
1
2
J
v
(
˙
φ
1
+
˙
θ)
2
+
1
2
J
v
(
˙
φ
2
−
˙
θ)
2
where J
b
, J
r
, and J
v
are the moment of inertia of the body, the rotor, and the wheel
along the vertical axis, respectively [50]. This Lagrangian can be rewritten as
L(q, ˙ q) =
1
2
˙ x
˙ y
˙
θ
˙
ψ
˙
φ
1
˙
φ
2
¸
¸
¸
¸
¸
¸
¸
¸
T
m 0 0 0 0 0
0 m 0 0 0 0
0 0 J
θ
J
r
J
v
−J
v
0 0 J
r
J
r
0 0
0 0 J
v
0 J
v
0
0 0 −J
v
0 0 J
v
¸
¸
¸
¸
¸
¸
¸
¸
˙ x
˙ y
˙
θ
˙
ψ
˙
φ
1
˙
φ
2
¸
¸
¸
¸
¸
¸
¸
¸
with J
θ
= J
b
+ J
r
+ 2J
v
. Clearly, the inertia matrix M is constant. Thus, the
Coriolis and centrifugal matrix C(q, ˙ q) associated with M vanishes identically. The
algebraicdiﬀerential equations of motion for this system is as the following
d
dt
∂L
∂ ˙ q
−
∂L
∂q
= W
T
(q)
¸
λ
1
λ
2
+
¸
0
3×3
I
3×3
τ
1
τ
2
τ
3
¸
¸
W(q) ˙ q = 0
(6.31)
which can be expressed as
m¨ x = λ
1
sin(θ + φ
1
) + λ
2
sin(θ −φ
2
)
m¨ y = −λ
1
cos(θ + φ
1
) −λ
2
cos(θ −φ
2
)
J
θ
¨
θ +J
r
¨
ψ + J
v
¨
φ
1
−J
v
¨
φ
2
= −λ
1
l cos φ
1
+λ
2
l cos φ
2
J
r
¨
ψ + J
r
¨
θ = τ
1
J
v
¨
φ
1
+ J
v
¨
θ = τ
2
J
v
¨
φ
2
−J
v
¨
θ = τ
3
(6.32)
Based on theorem 6.1.1, the reduced Lagrangian system for the snakeboard system
in (6.32) is an underactuated system with conﬁguration vector q
r
= (θ, ψ, φ
1
, φ
2
) as
196
the following
M
r
(q
r
)¨ q
r
+ C
r
(q
r
, ˙ q
r
) ˙ q
r
= F
r
(q
r
)τ (6.33)
where
M
r
(q
r
) = M
r
(η) = D(η)
T
MD(η)
C
r
(q
r
, ˙ q
r
) = C
r
(η, ˙ η) = D
T
(η)M
˙
D(η, ˙ η)
F
r
(q
r
) = F
r
= D
T
(q)F(q) =
¸
0
1×3
I
3×3
By direct calculation, we obtain the following inertia matrix
M
r
(q
r
) =
˜
J
θ
(φ
1
, φ
2
) J
r
J
v
−J
v
J
r
J
r
0 0
J
v
0 J
v
0
−J
v
0 0 J
v
¸
¸
¸
¸
where
˜
J
θ
= J
θ
+ m(ρ
2
1
+ ρ
2
2
)
Setting b := ρ
2
1
+ ρ
2
2
, we get
b(φ
1
, φ
2
) = l
2
cos
2
φ
1
+ cos
2
φ
2
+ 2 cos φ
1
cos φ
2
cos(φ
1
+ φ
2
)
sin
2
(φ
1
+ φ
2
)
which means b ≥ 0 does not depend on θ, ψ. Thus, M
r
(q
r
) = M
r
(φ
1
, φ
2
). In other
words, (φ
1
, φ
2
) are the shape variables and (θ, ψ) are the external variables of the
reduced snakeboard. Notice that b(φ
1
, φ
2
) has an isolated singularity at (φ
1
, φ
2
) = 0
(under the assumption that [φ
i
[ < π/2; i = 1, 2).
The reduced system in (6.33) has noninteracting control inputs (i.e. F
r
=
col(F
1
r
, F
2
r
) such that F
1
r
= 0). Therefore, using collocated partialfeedback lineariza
tion, the dynamics of all three actuated conﬁguration variables (ψ, φ
1
, φ
2
) can be
linearized using an explicit change of control in the form
τ = α(φ)u + β(φ, ˙ q
r
)
where φ = (φ
1
, φ
2
) and α(φ) is a positive deﬁnite matrix with a welldeﬁned p.d. limit
as φ →0. After collocated partial feedback linearization, we get
¨
ψ = u
1
¨
φ
1
= u
2
¨
φ
3
= u
3
(6.34)
This gives the following asymptotic output tracking result for the snakeboard.
Proposition 6.2.1. Denoting ξ
1
= (ψ, φ
1
, φ
2
) and ξ
2
= (
˙
ψ,
˙
φ
1
,
˙
φ
2
), we obtain
˙
ξ
1
= ξ
2
,
˙
ξ
2
= u
197
with u ∈ R
3
. Let ξ
d
(t) ∈ R
3
be a desired C
2
smooth trajectory for the snakeboard.
Then, the following partial state and output feedback
u = k
p
(ξ
1
−ξ
d
) +k
d
(ξ
2
−
˙
ξ
d
) +
¨
ξ
d
, k
p
, k
d
< 0
achieves global exponential tracking of the trajectory ξ
d
for the snakeboard.
Proof. Set e
1
= ξ
1
−ξ
d
and e
2
= ξ
2
−
˙
ξ
d
. Then
˙ e
1
= e
2
, ˙ e
2
= k
p
e
1
+ k
d
e
2
and the e
i
’s vanish exponentially.
Proposition 6.2.1 allows exponential tracking of certain “gaits” or sinusoid inputs
(ψ, φ
1
, φ
2
) for locomotion of the snakeboard [50]. Here, the contribution is that instead
of algebraic manipulation of the equations of motion (6.32) in [50], we propose a
systematic method for reduction and motion generation of a complex example like the
snakeboard which is applicable to higherorder underactuated nonholonomic systems
as well.
Locomotion Control of the Snakeboard
The key to locomotion control of the snakeboard is the invariance of the general
ized momentum conjugate to θ in the Lagrangian reduced system of the snakeboard
in equation (6.33). Since, the potential energy of the system is zero and θ is an
unactuated variable of the reduced system (6.33), the generalized momentum
π
θ
=
∂L
r
∂
˙
θ
=
˜
J
θ
(φ
1
, φ
2
)
˙
θ + J
r
˙
ψ + J
v
˙
φ
1
−J
v
˙
φ
1
is a conserved quantity. Assuming that the system was initially at rest (i.e. had zero
initial velocity), π
θ
= 0, ∀t > 0 and
˙
θ = −
J
r
˜
J
θ
(φ
1
, φ
2
)
˙
ψ −
J
v
˜
J
θ
(φ
1
, φ
2
)
(
˙
φ
1
−
˙
φ
2
) (6.35)
Based on the following assumption, both the last equation and the expression for
ρ
1
, ρ
2
in (6.30) can be highly simpliﬁed.
Assumption 6.2.1. Assume φ
1
= φ
2
= φ and [φ[ < π/2.
Lemma 6.2.1. Under Assumption 6.2.1, the following equations hold:
ρ
1
(φ) =
l cos(θ)
tan(φ)
ρ
2
(φ) =
l sin(θ)
tan(φ)
˙
θ = −
J
r
tan(φ)
2
ml
2
+ J
θ
tan(φ)
2
˙
ψ
(6.36)
198
In addition, the constraint equation for the snakeboard is equivalent to the following
nonholonomic constraints
sin(θ + φ) ˙ x −cos(θ +φ) ˙ y −l cos(φ)
˙
θ = 0
sin(θ) ˙ x −cos(θ) ˙ y = 0
(6.37)
Proof. Setting φ
1
= φ
2
= φ, we get
ρ
1
= l
cos(φ
1
) cos(θ −φ
2
) + cos(φ
2
) cos(θ + φ
1
)
sin(φ
1
+ φ
2
)
= l
cos(φ)[cos(θ −φ) + cos(θ + φ)]
sin(2φ)
= l
cos(φ)(2 cos(θ) cos(φ))
2 sin(φ) cos(φ)
=
l cos(θ)
tan(φ)
Similarly, one obtains ρ
2
= l sin(θ)/ tan(φ). Thus,
˜
J
θ
can be rewritten as
˜
J
θ
(φ) = J
θ
+ m(ρ
2
1
+ρ
2
2
) = J
θ
+
ml
2
tan
2
(φ)
From this equation, we have
˙
θ = −
J
r
˜
J
θ
(φ)
˙
ψ = −
J
r
tan
2
(φ)
ml
2
+ J
θ
tan
2
(φ)
˙
ψ
In addition, under Assumption 6.2.1, the velocity constraints for the snakeboard can
be expressed as
sin(θ +φ) ˙ x −cos(θ + φ) ˙ y −l cos(φ)
˙
θ = 0
sin(θ −φ) ˙ x −cos(θ −φ) ˙ y + l cos(φ)
˙
θ = 0
By adding the last two equations, one gets
[sin(θ + φ) + sin(θ −φ)] ˙ x −[cos(θ + φ) + cos(θ −φ)] ˙ y = 0
which can be simpliﬁed as
cos(φ)[sin(θ) ˙ x −cos(θ) ˙ y] = 0
Due to [φ[ < π/2, the last constraint reduces to
sin(θ) ˙ x −cos(θ) ˙ y = 0
and the result follows.
Notice that the velocity constraints in (6.37) are exactly the same as the velocity
199
constraints of a car. Roughly speaking, a velocity constraint in the form
sin(θ) ˙ x −cos(θ) ˙ y = 0
for the snakeboard, implies that the back half of the snakeboard can be removed and
two new virtual wheels can be added to the middle of the snakeboard. These virtual
wheels must be along the body of the snakeboard with orientation angles θ w.r.t. the
horizontal axis. This connection between the snakeboard and the car suggests the
following formal equivalence between the two systems.
Proposition 6.2.2. Let q = (x, y, θ, ψ, φ) ∈ SE(2)S
1
S
1
denote the conﬁguration
vector of both the snakeboard and the car with one diﬀerence. For the snakeboard, ψ
denotes the angle of the rotor, while ψ is the angle of the rotation of the rear wheels
in a car. Then, the kinematic model of the snakeboard and the car are the same
for φ = 0 (i.e. the diﬀeomorphism that maps the conﬁguration of the snakeboard
to the conﬁguration of the car is an identity map). In addition, both the kinematic
snakeboard and the kinematic car can be transformed into a chainedform system as
the following
˙ x
1
= v
1
˙ x
2
= v
2
˙ x
3
= x
2
v
1
˙ x
4
= x
3
v
1
(6.38)
where the diﬀeomorphism
(x
1
, x
2
, x
3
, x
4
, v
1
, v
2
) = T(x, y, θ, φ, ω
ψ
, ω
φ
)
for the snakeboard is given by
x
1
= x
x
2
=
tan(φ)
l cos(θ)
3
x
3
= tan(θ)
x
4
= y
v
1
= −
lJ
r
cos(θ) tan(φ)
ml
2
+ J
θ
tan
2
(φ)
ω
ψ
v
2
=
(1 + tan
2
(φ))
l cos
3
(θ)
ω
φ
−
3 sin(θ) tan(φ)
3
J
r
l cos(θ)
3
(ml
2
+J
θ
tan
2
(φ)
ω
ψ
(6.39)
and the dynamics of (ψ, φ) is linear as the following
˙
ψ = ω
ψ
˙
φ = ω
φ
˙ ω
ψ
= u
1
˙ ω
φ
= u
2
(6.40)
200
Proof. The dynamics of the snakeboard with φ
1
= φ
2
= φ can be expressed as
˙ x = ρ
1
(φ)
˙
θ
˙ y = ρ
2
(φ)
˙
θ
˙
θ = −
J
r
tan
2
(φ)
ml
2
+ J
θ
tan
2
(φ)
˙
ψ
¨
ψ = u
1
¨
φ = u
2
(6.41)
where
ρ
1
(φ) =
l cos(θ)
tan(φ)
, ρ
2
(φ) =
l sin(θ)
tan(φ)
Deﬁning
w
1
= −
lJ
r
tan(φ)
ml
2
+ J
θ
tan
2
(φ)
˙
ψ
the kinematic model of the snakeboard takes the following form
˙ x = cos(θ)w
1
˙ y = sin(θ)w
1
˙
θ =
1
l
tan(φ)w
1
˙
φ = w
2
(6.42)
with control inputs (w
1
, w
2
) ∈ R
2
. Equation (6.42) is exactly the same as the kine
matic model of a car with conﬁguration (x, y, θ, φ) where w
1
is the velocity of the
rear wheels of the car and w
2
is the velocity of the steering wheel. Now, applying
the following change of coordinates and control (which is equivalent to the one in the
question)
x
1
= x
x
2
=
tan(φ)
l cos(θ)
3
x
3
= tan(θ)
x
4
= y
v
1
= cos(θ)w
1
v
2
= ˙ x
2
(6.43)
transforms the kinematic model of the snakeboard into the chainedform nonholo
nomic system in (6.38).
Remark 6.2.1. Proposition 6.2.2 reduces the locomotion control for the snakeboard
to the stabilization problem for a car which is a rather wellstudied problem [63, 4].
This is the ﬁrst direct evidence between control of a locomotion system and a mobile
system that are both transformable into chainedform nonholonomic systems. In
other words, this is an example of a case where the locomotion control for a system
reduces to the stabilization of a simple system.
201
Geometric Interpretation of (ρ
1
, ρ
2
) and the Singularity in the Kinematics
of the Snakeboard
The functions (ρ
1
, ρ
2
) in the kinematics and constraint equations of both the snake
board and the car have a meaningful geometric interpretation that is stated in the
following. Consider a car with a virtual rear wheel in the middle of the real rear
wheels and a virtual front wheel. The path and geometry of the rear and front vir
tual wheels for a car are shown in Figure 65. For the snakeboard, the front wheels
are the same but the rear wheel is a virtual wheel along the body of the snakeboard
at the center of mass (x, y). If one ﬁxes the orientation of the front wheel at φ. The
φ
φ
θ
ρ
ρ
θ
l
O
1
2
ρ
y
x
(x,y)
Figure 65: The geometry of the paths of the rear and front wheels of the car with a
radius of rotation ρ.
car rotates around the center of rotation O. The path of both wheels are circles.
Let ρ denote the radius of the rotation of the rear wheel located at point p = (x, y).
Then, the velocity v of point p is tangent to the circle (O, ρ) at p and is obtained by
v = ρ
˙
θ (6.44)
Therefore, the vertical and horizontal components of v are
˙ x = ρ cos(θ)
˙
θ
˙ y = ρ sin(θ)
˙
θ
(6.45)
In other words, setting ρ
1
= ρ cos θ and ρ
2
= ρ sin θ, the components of v satisfy the
following equations
˙ x = ρ
1
˙
θ
˙ y = ρ
2
˙
θ
(6.46)
From Figure 65, we have
ρ =
l
tan(φ)
202
Thus, ρ
1
and ρ
2
are given by
ρ
1
=
l cos(θ)
tan(φ)
, ρ
2
=
l sin(θ)
tan(φ)
This is exactly the same as the deﬁnition of ρ
1
, ρ
2
for the car and the snakeboard
with φ
1
= φ
2
= φ. Notice that ρ uniquely determined by the orientation angle of
the front wheel φ and does not depend on θ. Apparently, when φ → 0, ρ → ∞ and
ρ
1
, ρ
2
→∞. This is equivalent to the fact that θ is constant (or
˙
θ = 0) and the radius
ρ of a straight line is inﬁnity.
Now, let ψ denote the angle of rotation of the rear wheel with unit radius in the
car. Then, v = ρ
˙
θ =
˙
ψ, which means
˙
θ =
1
ρ
˙
ψ =
tan(φ)
l
˙
ψ
This gives the ﬁnal equations of motion for the kinematic car as
˙ x = cos(θ)v
˙ y = sin(θ)v
˙
θ =
tan(φ)
l
v
˙
φ = w
The last system is obtained without any explicit use of nonholonomic velocity con
straints. However, the fact that the rear and front wheels always move on a circle with
radius ρ is equivalent to the assumption that the wheels do not slip. The snakeboard
does not satisfy the last equation since it does not have a real rotating rear wheel.
Instead, it has a rotor with angle of rotation ψ. The invariance of the momentum
conjugate to θ in the reduced Lagrangian system of the snakeboard plays the role
of the kinematic equation
˙
θ =
˙
ψ/ρ in the car. One can observe that deﬁning the
curvature as κ = 1/ρ, we have
˙
θ = κ(φ)
˙
ψ
where κ(φ) is the curvature of the circle of rotation of the rear wheel of the car
corresponding to the front wheel angle φ. If the rear wheels of the car rotate with a
constant nonzero velocity
˙
ψ, the curvature κ(φ) plays the role of the control input
for the motion of the group element (x, y, θ) ∈ SE(2).
A rather similar property holds for the snakeboard with arbitrary front wheel and
rear wheel angles (φ
1
, φ
2
). The geometric interpretation of ρ is shown in Figure 66.
In this ﬁgure, (r
1
, r
2
) denote the radii of rotation corresponding to the front and back
wheels of the snakeboard. In addition, ρ is the radius of rotation of the center of
mass of the snakeboard. This is due to the midsection theorem in planar geometry.
Observe that
ρ
2
= a
2
+ h
2
203
θ
γ
ρ
ρ
ρ
φ
φ
φ
2
2
1
1
1
φ
r
2
r
1
2
x
y
l
l
a
h
O
γ − θ
Figure 66: The geometry of the paths of the rear and front wheels and the center of
mass of the snakeboard with a radius of rotation ρ.
and the radii of rotation r
1
, r
2
satisfy
r
2
1
= h
2
+ (l +a)
2
r
2
2
= h
2
+ (l −a)
2
Adding the last two equation gives
ρ
2
=
r
2
1
+ r
2
2
2
−l
2
In other words, if r
1
, r
2
(or φ
1
, φ
2
) are constant, then ρ is constant and it is uniquely
determined by φ
1
, φ
2
, l. This means that the geometric path of (x, y) is a circle
centered at O with radius ρ.
An elementary calculation shows that
h =
2l
tan(φ
1
) + tan(φ
2
)
From h, the expressions of ρ, r
1
, r
2
, γ can be readily determined as functions of the
shape variables (φ
1
, φ
2
). Particularly, one obtains the following formula for ρ(φ
1
, φ
2
)
ρ
2
= a
2
+ h
2
= (l −htan(φ
2
))
2
+ h
2
= l
2
¸
(tan(φ
1
) −tan(φ
2
))
2
+ 4
(tan(φ
1
) + tan(φ
2
))
2
It is rather elementary to prove that the radius of rotation ρ from the last equation
is exactly the same as ρ =
ρ
2
1
+ ρ
2
2
for ρ
1
, ρ
2
deﬁned in equation (6.30).
204
6.3 Notions of Stabilization and Tracking
In this section, we deﬁne new stability/tracking notions for nonlinear systems called
stability/tracking. Consider the following nonlinear system
˙ x = f(x) (6.47)
where f : R
n
→R
n
is a C
1
smooth function.
Deﬁnition 6.3.1. (virtual equilibrium) We say x
0
∈ R
n
is a virtual equilibrium of
(6.47) within a distance > 0, if there exists an equilibrium x
∗
0
of (6.47) that is within
a distance from x
0
, i.e. ∃x
∗
0
∈ R
n
: f(x
∗
0
) = 0, x
∗
0
−x
0
 ≤ . We call x
∗
0
the adjacent
equilibrium to x
0
.
Deﬁnition 6.3.2. (stable virtual equilibrium) We call x
0
an (exponentially or
asymptotically) stable virtual equilibrium of (6.47), if there exists an adjacent equi
librium x
∗
0
within a distance > 0 from x
0
that is (exponentially or asymptotically)
stable for (6.47) in the sense of Lyapunov.
We are interested in considering the solutions of (6.47) under speciﬁc classes of
parameterized transformations where the notion of stability of solutions is useful.
Assume there exists a global diﬀeomorphism ψ(x, λ) : R
n
R
p
→ R
n
in x that is
continuous in λ and satisﬁes ψ(x, λ) = x =⇒ λ = 0. By continuity at λ = 0, we
have
∀ > 0, ∃λ = λ() : ψ(x, λ) −x ≤ (6.48)
We call ψ(x, λ) a near identity diﬀeomorphism for λ < 1. Denote the inverse of
y = ψ(x, λ) by x = φ(y, λ) so that
φ(ψ(x, λ), λ) = x, ψ(φ(y, λ), λ) = y
By deﬁnition φ(x, 0) = x and φ(x, λ) = x =⇒ λ = 0. Also, φ(x, λ) is continuous
w.r.t. λ. Applying the following change of coordinates
x
λ
= ψ(x, λ)
one obtains a parameterized nonlinear system that we call λsystem
˙ x
λ
=
˜
f(x
λ
, λ) (6.49)
where
˜
f(x
λ
, λ) =
¸
∂ψ(x, λ)
∂x
f(x)
x=φ(x
λ
,λ)
Apparently, any equilibrium point of (6.47) is an equilibrium point of (6.49) and vice
versa.
Deﬁnition 6.3.3. (stability under transformation) For any > 0, let λ = λ()
be such that (6.48) holds. We say x = x
0
is (exponentially/asymptotically) stable
205
virtual equilibrium of (6.47) under the transformation ψ(x, λ), iﬀ x
λ
= x
0
is a (ex
ponentially/asymptotically) stable equilibrium of the λsystem (6.49) in the sense
of Lyapunov with λ = λ(), or equivalently iﬀ x
0
= φ(x
0
, λ()) is (an) a (exponen
tially/asymptotically) stable equilibrium of (6.47) in the sense of Lyapunov, i.e. x
0
is an adjacent equilibrium to x
0
for the λsystem.
The following result is a common sense corollary of the deﬁnition of stability
which means the original system has an asymptotically stable equilibrium point near
x
0
in the usual sense of stability.
Corollary 6.3.1. If x
0
is an (exponentially/asymptotically) stable equilibrium of
(6.47) under the transformation ψ(x, λ), then there exists an equilibrium point x
0
of
(6.47) that is (exponentially/asymptotically) stable equilibrium of (6.47) in the sense
of Lyapunov within a distance > 0 from x
0
, i.e. x
0
.
Proof. Let x
0
= φ(x
0
, λ()), then
x
0
−x
0
 = x
0
−ψ(x
0
, λ()) <
Remark 6.3.1. Two simple examples of ψ(x, λ) are the following
ψ(x, λ) = x +λu, λ ∈ R, u = 1
and
ψ(x, λ) = x + Aλ, x ∈ R
n
, λ ∈ R
p
, A has full rank
Now, consider a nonlinear control system with a dynamic state feedback
˙ x = f(x, u)
˙
λ = g(x, λ)
(6.50)
Let ψ(x, λ) be a global diﬀeomorphism as before. Applying the change of coordinates
x
λ
= ψ(x, λ)
we obtain the following λsystem
˙ x
λ
=
˜
f(x
λ
, λ, u)
˙
λ = ˜ g(x
λ
, λ)
(6.51)
where
˜
f(x, λ, u) =
¸
∂ψ(x, λ)
∂x
f(x, u) +
∂ψ(x, λ)
∂λ
g(x, λ)
x=φ(x
λ
,λ)
˜ g(x
λ
, λ) = g(φ(x
λ
, λ), λ)
Deﬁnition 6.3.4. (stabilization) For a ﬁx > 0, let λ
f
= λ() satisfy (6.48). Let
x
0
be a desired equilibrium point of the system with zero control, or f(x
0
, 0) = 0. We
206
say the dynamic state feedback
u = k(x, λ)
˙
λ = g(x, λ)
(6.52)
achieves globally asymptotic stabilization of x
0
for (6.50) under the transformation
ψ(x, λ), iﬀ for the closedloop λsystem
˙ x
λ
=
˜
f(x
λ
, λ,
˜
k(x
λ
, λ))
˙
λ = ˜ g(x
λ
, λ)
where
˜
k(x
λ
, λ) = k(φ(x
λ
, λ), λ), (x
0
, λ
f
) is a globally asymptotically stable equilib
rium in the sense of Lyapunov, or equivalently x
0
= φ(x
0
, λ
f
) and λ
f
are globally
asymptotically stable equilibriums of (6.50).
The notion of tracking for a nonlinear control system ˙ x = f(x, u) can be deﬁned
for an equivalent transformed control system with dynamic state feedback based on
the notion of stabilization.
Deﬁnition 6.3.5. (tracking) Consider the following nonlinear system with input u
and output y
˙ x = f(x, u)
y = h(x)
(6.53)
We say a control law u achieves asymptotic tracking of a trajectory y
d
() for (6.53)
under transformation ψ(x, λ), if there exists a control law u that achieves asymptotic
tracking of the desired trajectory y
λ
= y
d
for
˙ x
λ
=
˜
f(x
λ
, λ, u)
˙
λ = ˜ g(x
λ
, λ)
y
λ
= h(φ(x, λ))
(6.54)
6.4 Global Exponential Stabilization/Tracking for
a Mobile Robot
In this section, we design a nonlinear controller for global exponential stabilization
and tracking of a nonholonomic mobile robot (Figure 62).
6.4.1 Dynamics of a Mobile Robot in SE(2)
In the majority of previous control design strategies for this nonholonomic mobile
robot, the kinematic model of the robot is used and the input torque are then obtained
using adding an integrator and highgain control. Here, we directly use the dynamic
207
model of the mobile robot as the following
˙ x
1
= cos(θ)v
1
˙ x
2
= sin(θ)v
1
˙
θ = v
2
˙ v
1
= τ
1
˙ v
2
= τ
2
(6.55)
Since this mobile robot moves on a surface, a more appropriate form of representation
of the dynamics of the mobile robot is in SE(2). SE(2) is the Special Euclidean
group in R
2
with elements g = (R, x) where R is the rotation matrix satisfying
R
T
R = I, det(R) = 1 (i.e. R ∈ SO(2)) and x ∈ R
2
. Equivalently g can be deﬁned as
g =
¸
R x
0 I
where the group product g
1
g
2
is matrix multiplication. From (6.55), the dynamics of
the mobile robot in SE(2) can be expressed as the following
˙ x = (Re
1
)v
1
˙
R = Rˆ v
2
˙ v = τ
(6.56)
where
R =
¸
cos(θ) −sin(θ)
sin(θ) cos(θ)
and the skewsymmetric matrix ˆ v
2
is deﬁned as follows
ˆ v
2
=
¸
0 −v
2
v
2
0
or equivalently the evolution of the group element g in (6.56) can be expressed as
˙ g = g Ω, ˙ v = τ
where Ω ∈ se(2) (i.e. the Lie algebra associated with SE(2))
Ω =
¸
ˆ v
2
v
1
0 0
We use the representation in (6.56) as the model of this nonholonomic mobile robot.
6.4.2 NearIdentity Diﬀeomorphism
Consider the following near identity smooth change of coordinates
z = ψ(x, λ) := x + λRe
1
(6.57)
208
where λ ∈ R, e
i
is the ith standard basis in R
2
, and R = R(θ) is the rotation matrix
by θ in R
2
. This transformations is schematically depicted in Figure 67. The inverse
x
x
z
z
d
d= λ R e
d= λ
1
2
1
2
1
θ
Figure 67: A near identity transformation with a parameter [λ[ <1.
of this transformation is a near identity diﬀeomorphism given by
x = φ(z, λ) = z −λRe
1
Apparently, for all x ∈ R
2
the following property holds
ψ(x, λ) −x = λRe
1
 = [λ[
and for all > 0, λ = achieves the property
ψ(x, λ) −x ≤
Notice that these properties of the near identity diﬀeomorphism ψ(x, λ) are true for
(R, x) ∈ SE(n) with n ≥ 2 and are not restricted to the special case of n = 2. The
dynamics of the system in zcoordinates (i.e. λsystem) can be expressed as
˙ z = R
λ
v + (R
λ
e
1
)
˙
λ
˙
R
λ
= R
λ
ˆ v
2
(λ,
˙
λ)
˙ v = τ
(6.58)
where v = (v
1
, v
2
)
T
and R
λ
, ˆ v
2
are given by
R
λ
=
¸
cos(θ) −λsin(θ)
sin(θ) λcos(θ)
, ˆ v
2
(λ,
˙
λ) =
0 −λv
2
1
λ
v
2
˙
λ
λ
¸
¸
Note that for λ = 1, R
λ
is not a rotation matrix and R
T
λ
R
λ
= I. In fact, we have
R
T
λ
R
λ
=
¸
1 0
0 λ
2
, det R
λ
= λ
Thus, for λ = 0, the matrix R
λ
is invertible.
209
6.4.3 Control Design for Stabilization/Tracking
In this section, we present our main stabilization/tracking results for a nonholonomic
mobile robot.
Proposition 6.4.1. Any desired position x
d
∈ R
2
for the nonholonomic mobile robot
in (6.56) can be rendered globally exponentially stable using a smooth dynamic state
feedback in an explicit form. Equivalently, this feedback law globally exponentially
stabilizes (i.s.o. Lyapunov) an equilibrium x
d
of (6.56) with a distance from x
d
, i.e.
x
d
−x
d
 = .
Proof. Fix an > 0, and let the dynamics of λ be as
˙
λ = −c
λ
(λ −λ
f
), λ(0) = λ
0
> , λ
f
= , c
λ
> 0
Clearly, λ = λ
f
is exponentially asymptotically stable for this system and λ(t) ≥
> 0, ∀t ≥ 0. The task in the question is to stabilize the position of the mobile
robot to x = x
d
(the orientation of the mobile robot can be trivially exponentially
stabilized by setting v
1
= 0 at x = x
d
). Thus, we set the desired equilibrium of the
λsystem in (6.58) to z
d
= x
d
. The goal is to globally exponentially stabilize z = z
d
for the λsystem. This renders the equilibrium x
d
= φ(x
d
, ) of the original system
is (6.56) globally exponentially stable (i.s.o. Lyapunov). But x
d
− x
d
 = , thus
this achieves global exponential stabilization of x
d
for the original system (6.56).
To render z
d
globally exponentially stable for the λsystem in (6.56) can be rendered
globally exponentially stable. For doing so, let us calculate ¨ z. We have
¨ z = R
λ
˙ v +
˙
R
λ
v + (
˙
R
λ
e
1
)
˙
λ + (R
λ
e
1
)
˙
λ
= R
λ
τ + R
λ
ˆ v
2
(λ,
˙
λ)v + R
λ
(ˆ v
2
(λ,
˙
λ) −c
λ
I)e
1
˙
λ
Thus, deﬁning the change of control
τ = R
−1
λ
u − ˆ v
2
(λ,
˙
λ)v −(ˆ v
2
(λ,
˙
λ) −c
λ
I)
˙
λe
1
where
˙
λ = −c
λ
(λ −), we get
¨ z = u
which is a twodimensional doubleintegrator. Therefore, applying
u = −c
p
(z −z
d
) −c
d
˙ z, c
p
, c
d
> 0
or
u = −c
p
(z −z
d
) −c
d
(R
λ
v + R
λ
e
1
˙
λ)
renders the equilibrium point z = z
d
of the λsystem globally exponentially stable.
The overall smooth dynamic state feedback is in the form
τ = −c
p
R
−1
λ
(z −z
d
) −c
d
v −c
d
˙
λe
1
− ˆ v
2
(λ,
˙
λ)v −(ˆ v
2
(λ,
˙
λ) −c
λ
I)
˙
λe
1
˙
λ = −c
λ
(λ −)
(6.59)
210
Figures 68 and 69 show the path of the mobile robot starting at position x =
(2, 3)
T
for the initial orientation angles θ = kπ/4, k = 0, . . . , 7. The trajectories for
the position and input torques (controls) are shown in Figure 610
Similar results for trajectories starting at x = (0, 2)
T
with the initial orientation
angles θ = 0, π/4, π/2, 3π/2 are shown in Figure 611 (for the complement angles
the trajectories are symmetric w.r.t. vertical axis). Also, trajectories of the posi
tion and input torques (controls) for initial position x = (0, 2)
T
) and orientation
θ = 0 are shown in Figure 612. These results demonstrate that the controller ef
ﬁciently stabilizes the origin for this nonholonomic mobile robot. Each trajectory
exponentially converges to a point within a distance = 0.01 from the desired equi
librium x
d
= 0. This is suﬃciently close to the origin for all practical purposes.
The values of the parameters in all simulations for the mobile robot were chosen as
c
λ
= 1, c
p
= 1, c
d
= 2, λ(0) = 0.5, λ
f
= = 0.01. The trace trajectories of this non
holonomic robot are shown in Figure 613. Based on this ﬁgure, graphically, the robot
makes a turn to face the desired destination (i.e. the origin) and goes exponentially
fast towards the destination.
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=0
x1
x
2
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=pi/4
x1
x
2
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=pi/2
x1
x
2
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=3pi/4
x1
x
2
Figure 68: Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane for initial
position x = (2, 3)
T
(v = 0) and orientation angles θ = 0, π/4, π/2, 3π/4.
Asymptotic tracking of a desired output z
d
can be obtained based on the follow
ing result.
Proposition 6.4.2. The following dynamic state feedback law achieves asymptotic
211
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=pi
x1
x
2
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=5pi/4
x1
x
2
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=3pi/2
x1
x
2
0 0.5 1 1.5 2 2.5
0
0.5
1
1.5
2
2.5
3
3.5
theta=7pi/4
x1
x
2
Figure 69: Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane for initial
position x = (2, 3)
T
(v = 0) and orientation angles θ = π, 5π/4, 3π/2, 7π/4.
tracking for the desired output y
d
() of the nonholonomic mobile robot (6.55)
τ = −c
p
R
−1
λ
z −c
d
v −c
d
˙
λe
1
− ˆ v
2
(λ,
˙
λ)v −(ˆ v
2
(λ,
˙
λ) −c
λ
I)
˙
λe
1
+ R
−1
λ
(c
p
y
d
+ c
d
˙ y
d
+ ¨ y
d
)
˙
λ = −c
λ
(λ −)
(6.60)
Proof. Set the desired trajectory in zcoordinates to z
d
= y
d
. Then, the following
feedback
u = −c
p
(z −z
d
) −c
d
( ˙ z − ˙ z
d
) + ¨ z
d
guarantees that the error e = z − z
d
globally asymptotically converges to zero and
the result follows from the relation between τ and u.
Figure 614 demonstrates simulation results of the tracking for a nonholonomic
robot starting at position (4, 4) with orientation angle π/2. The desired trajectory
is an ellipse (x, y) = (3 sin t, 4 cos t). Clearly, the robot very quickly converges to an
neighborhood of the desired trajectory (e.g. = 0.01).
212
0 2 4 6 8 10 12
0
0.5
1
1.5
2
2.5
3
time (sec)
P
o
s
i
t
i
o
n
x1
x2
0 2 4 6 8 10 12
−30
−20
−10
0
10
20
30
time (sec)
C
o
n
t
r
o
l
tau1
tau2
Figure 610: Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane starting
at x = (2, 3)
T
(v = 0) with angle θ = π/4
−1 −0.5 0 0.5 1
0
0.5
1
1.5
2
2.5
theta=0
x1
x
2
−1 −0.5 0 0.5 1
0
0.5
1
1.5
2
2.5
theta=pi/4
x1
x
2
−1 −0.5 0 0.5 1
0
0.5
1
1.5
2
2.5
theta=pi/2
x1
x
2
−1 −0.5 0 0.5 1
0
0.5
1
1.5
2
2.5
theta=3pi/4
x1
x
2
Figure 611: Trajectories of the nonholonomic mobile robot in (x
1
, x
2
)plane for initial
position x = (0, 2)
T
(v = 0) and initial orientation angles θ = 0, π/4, π/2, 3π/4.
213
0 1 2 3 4 5 6 7 8 9 10
−0.5
0
0.5
1
1.5
2
time (sec)
P
o
s
i
t
i
o
n
x1
x2
0 1 2 3 4 5 6 7 8 9 10
−5
0
5
time (sec)
C
o
n
t
r
o
l
tau1
tau2
Figure 612: Evolution of position and controls for the nonholonomic mobile robot in
(x
1
, x
2
)plane starting at x = (0, 2)
T
(v = 0) with angle θ = 0
−6 −4 −2 0 2 4 6
−5
−4
−3
−2
−1
0
1
2
3
4
5
x
y
−6 −4 −2 0 2 4 6
−5
−4
−3
−2
−1
0
1
2
3
4
5
x
y
(a) (b)
−6 −4 −2 0 2 4 6
−5
−4
−3
−2
−1
0
1
2
3
4
5
x
y
−6 −4 −2 0 2 4 6
−5
−4
−3
−2
−1
0
1
2
3
4
5
x
y
(c) (d)
Figure 613: Trace trajectories of a twowheel nonholonomic robot with initial posi
tion (5, 4) and heading angles (a) θ = 0, (b) θ = π/2, (c) θ = π, and (d) θ = 3π/2.
214
0 2 4 6 8 10 12 14 16 18
−4
−2
0
2
4
6
time (sec)
x
x
x
d
0 2 4 6 8 10 12 14 16 18
−5
0
5
time (sec)
y
y
y
d
Figure 614: Position trajectory tracking for a mobile robot from the initial point
(4, 4, π/2).
215
Chapter 7
Control of Nonlinear Systems in
Nontriangular Normal Forms
This chapter is devoted to control of nonlinear systems in normal forms that pos
sess nontriangular structures. Control of nonlinear systems in triangular forms, i.e.
(strict) feedback and feedforward forms, has been extensively studied in the past
decade. However, few results are available that deal with nonlinear systems with
nontriangular structures. later, we will be more precise on what we mean by nontri
angular forms and present the challenging problems and stabilization limitations of
nonlinear systems in nontriangular forms. The following notations are convenient in
this chapter.
Notation. GAS stands for globally asymptotically stable.
Notation.   = 'x, x`
1
2
denotes the Euclidean norm and  
p
denotes the L
p
norm
in R
n
.
Notation. (Lie derivative) Let f(x) : R
n
→R and g : R
n
→R
n
be smooth functions.
The directional derivative of f(x) with respect to g(x) is deﬁned as any of the following
equivalent forms
L
g
f(x) :=
∂f(x)
∂x
g(x) = '∇f(x), g(x)` = Df(x) g(x) (7.1)
where 'x, y` = x
T
y denotes the inner product in R
n
. Moreover, the kth order Lie
derivative of f w.r.t. g is denoted by L
(k)
g
f and is deﬁned as
L
(k)
g
f(x) := L
g
(L
((k−1)
g
f(x)), k ∈ Z
+
(7.2)
with L
(0)
g
f(x) = f(x).
The outline of this chapter is as follows: First, we provide some background
on stabilization of nonlinear systems in triangular forms, namely, (strict) feedback
and feedforward forms. Secondly, we focus on control of nonlinear systems with
nontriangular structures.
216
7.1 Introduction
In the preceding chapters, we obtained three types of normal forms for underactuated
mechanical systems. Namely, cascade nonlinear systems in strict feedback form, strict
feedforward form, and diﬀerent variations of nontriangular linearquadratic forms. In
the past, eﬀective methods have been developed for stabilization of nonlinear systems
in triangular forms (i.e. feedback/feedforward) [57, 36, 102, 103, 101, 56, 80]. Namely,
the backstepping procedure [57, 36, 103] and forwarding methods [102, 101, 56, 80] are
analytic tools for stabilization of nonlinear systems in feedback forms and feedforward
forms, respectively.
The class of nonlinear systems aﬃne in control
˙ x = f(x) + g(x)u, x ∈ R
n
, u ∈ R
p
(7.3)
is one of the most wellstudied types of nonlinear systems. Assume there exists an
output y = h(x) that has uniform relative degree r ≤ n with respect to the input u
over an open neighborhood U of x = x
0
∈ R
n
, i.e. there exist α(x), β(x) such that
the following properties hold over U for a constant r
L
(k)
g
h(x) = 0, ∀k ∈ ¦0, 1, . . . r −1¦
y
(r)
= α(x)u + β(x)
where α(x) is an invertible matrix. According to [36], there exists a change of coor
dinates and control in the form
(z, ξ) = Φ(x), v = α(x)u + β(x)
that transforms the nonlinear system in (7.3) into the ByrnesIsidori normal form of
degree r ≤ n
˙ z = F(z, ξ
1
, ξ
2
, . . . , ξ
r
)
˙
ξ
1
= ξ
2
˙
ξ
r
= v
(7.4)
Here, Φ(x) is a diﬀeomorphism over U. If U = R
n
, then y = h(x) has global uniform
relative degree r and the aforementioned change of coordinates is global. The z
subsystem
˙ z = F(z, ξ)
in (7.4) with ξ = (ξ
1
, . . . , ξ
r
) is called the zerodynamics of (7.3) [36]. Given ξ
1
=
ξ
2
= . . . = ξ
r
= 0, if z = 0 is asymptotically stable for the zerodynamics, then both
the output y = h(x) and the zerodynamics are called minimum phase.
Under the condition that the vector ﬁeld F in (7.4) does not depend on (ξ
2
, . . . , ξ
r
),
217
the nonlinear system in (7.4) reduces to
˙ z = F(z, ξ
1
)
˙
ξ
1
= ξ
2
˙
ξ
r
= v
(7.5)
which is in strict feedback form [36, 57]. In general, if at most one of ξ
1
, . . . , ξ
r
appears
in f, it is known how to stabilize the composite system in (7.4) under appropriate
technical conditions (see [36], chapter 9). The importance of the strict feedback
normal form in (7.5) is due to the existence of the following theorem called standard
backstepping [36].
Theorem 7.1.1. (standard backstepping) Consider the nonlinear system in (7.5) and
assume there exists a smooth state feedback ξ
1
= α(z) with α(0) = 0 such that for the
closedloop zerodynamics
˙ z = F(z, α(z)) (7.6)
the origin z = 0 is globally asymptotically stable (GAS). Deﬁne K
1
(z) := α(z). Then,
there exists a state feedback
ξ
j+1
= K
j+1
(z, ξ
1
, . . . ξ
j
), j = 1, . . . , r (7.7)
in an explicit form where ξ
r+1
:= v that renders the origin (z, ξ
1
, . . . , ξ
j
) = 0 GAS for
the closedloop (z, ξ
1
, . . . , ξ
j
)subsystem (i.e. jth subsystem) of (7.5). In addition, let
V
0
(z) be a smooth positive deﬁnite and proper Lyapunov function associated with the
closedloop zerodynamics subsystem satisfying
˙
V
0
(z) := ∇V
0
(z) f(z, α(z)) < 0, ∀z =
0. Then, the the Lyapunov function V
j
(z, ξ
1
, . . . , ξ
j
), j = 1, . . . , r associated with the
jth subsystem is given by
V
j
(z, ξ
1
, . . . , ξ
j
) = V
0
(z) +
1
2
(ξ
1
−K
1
(z))
2
+ 1
{j>1}
(j)
1
2
Σ
j
i=2
(ξ
i
−K
i
(z, ξ
1
, . . . , ξ
i−1
))
2
and satisﬁes
˙
V
j
< 0, ∀(z, ξ
1
, . . . , ξ
j
) = 0 (1
A
() denotes the indicator function of the
set A).
Proof. See remark 7.3.2.
In the next section, we provide a constructive method for obtaining both the state
feedback K
j
and the Lyapunov function V
j
in a recursive manner (this can be found
in [36], also see theorem 7.3.1 here).
In contrast to nonlinear systems in strict feedback forms, the following normal
form
˙ z = F(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(7.8)
has a nontriangular structure. Throughout this chapter, we assume F(z, ξ
1
, ξ
2
) de
pends on both ξ
1
, ξ
2
unless otherwise is stated. Our main focus in this chapter is
218
on stabilization of nonlinear systems in the form (7.8). Our main motivation is that
(almost) all nontriangular normal forms for underactuated systems are special classes
of ByrnesIsidori normal form with a vector doubleintegrator linear part as in (7.8).
In addition, the following nontriangular normal form with a tripleintegrator linear
part
˙ z = F(z, ξ
1
, ξ
2
, ξ
3
)
˙
ξ
1
= ξ
2
˙
ξ
2
= ξ
3
˙
ξ
3
= v
(7.9)
appeared in the study of ﬂexiblelink robots and a helicopter model with the eﬀect of
the input of the attitude dynamics in the translational dynamics. The actual input
in both cases is ξ
3
which is augmented with an integrator at the input.
Remark 7.1.1. Notice that by redeﬁning the state variables and zerodynamics vector
ﬁeld as
ζ := col(z, ξ
1
), η
1
:= ξ
2
, η
2
:= ξ
3
, F
0
:= col(F(z, ξ
1
, ξ
2
, ξ
3
), ξ
2
)
The dynamics of system (7.9) can be rewritten as the following
˙
ζ = F
0
(ζ, η
1
, η
2
)
˙ η
1
= η
2
˙ η
2
= v
(7.10)
which is a ByrnesIsidori normal form of degree 2 (the same as (7.8)). Roughly
speaking, this implies that certain stabilization problems for the class of nonlinear
systems in (7.9) with a tripleintegrator linear part reduce to equivalent problems for
the class of nonlinear systems in (7.8) with a (vector) doubleintegrator linear part.
Stabilization of ByrnesIsidori normal forms with a doubleintegrator linear part
as in (7.8) is currently considered a major open problem [36]. Particular cases of
interest include normal forms for mechanical systems where F(z, ξ
1
, ξ
2
) has quadratic
dependence in the velocity ξ
2
(e.g. the beamandball system). Our main contribu
tion in this chapter is to introduce an extension of the backstepping procedure for
stabilization of important special classes of normal form (7.8). This new backstepping
procedure is considerably diﬀerent from standard backstepping procedure that relies
on construction of appropriate Lyapunov functions.
Based on standard backstepping procedure, if a controller ξ
2
= K
1
(z, ξ
1
) exists
that renders the origin GAS for
˙ z = F(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
(7.11)
Then, the existence of a globally stabilizing state feedback v = K
2
(z, ξ
1
, ξ
2
) for the
composite system in (7.8) is trivial. In other words, the main stabilization problem
of interest is stabilization of nonlinear system (7.11) which is nonaﬃne in control ξ
2
.
Observe that it is nontrivial to formulate a stabilization problem for the nonlinear
system (7.8) analogous to the one for a cascade system in strict feedback form, unless
219
α(z) ≡ 0. We formulate this problem for the case of α(z) ≡ 0 as the following:
Problem 7.1.1. Consider the nonlinear system in (7.8). Assume that the zero
dynamics of (7.8) is globally minimum phase, i.e. given ξ
1
= ξ
2
= 0, the origin z = 0
is GAS for the zerodynamics
˙ z = F(z, 0, 0)
Find suﬃcient conditions such that there exists a state feedback v = k(z, ξ
1
, ξ
2
) that
globally asymptotically stabilizes (z, ξ
1
, ξ
2
) = 0 for (7.8).
Due to a counterexample by Sussmann [98], there exists a thirdorder nonlinear
system in the form (7.8) which is not semiglobally stabilizable. This major stabiliza
tion limitation of nonlinear systems in nontriangular forms as (7.8) is because of the
peaking phenomenon that was introduced by Sussmann and Kokotovic [99]. The prob
lem is that driving the state (ξ
1
, ξ
2
) exponentially fast to zero using a linear partial
state feedback does not necessarily asymptotically stabilize the composite system in
(7.8). Such a partial state feedback might cause the zsubsystem to have ﬁnite escape
time. In [99], growth conditions are provided on F(z, ξ
1
, ξ
2
) for global stabilization of
(7.8) using partial state feedback. The problem is that these growth conditions are
rather restrictive. Nevertheless, the work in [99] is one possible solution to problem
7.1.1.
In many applications (e.g. tracking for ﬂexiblelink robots), the zerodynamics is
not globally minimum phase. In this thesis, we are mainly interested in stabilization
problems for normal form (7.8) when the zerodynamics is not minimum phase. A
possible way to formulate a stabilization problem for normal form (7.8) analogous to
the nature of the one addressed using backstepping procedure for systems in strict
feedback form is the following problem.
Problem 7.1.2. Consider the nonlinear system in (7.8). Assume there exists a
smooth state feedback law ξ
1
= α(z) with α(0) = 0 such that for the zerodynamics
locked at ξ
2
= 0
˙ z = F(z, α(z), 0)
z = 0 is globally asymptotically stable. Find suﬃcient conditions such that there exist
state feedback laws ξ
2
= k
1
(z, ξ
1
) and u = k
2
(z, ξ
1
, ξ
2
) which asymptotically stabilize
the origin for the (z, ξ
1
)subsystem and the composite system (7.8), respectively.
A third possible stabilization problem is as follows:
Problem 7.1.3. Consider the nonlinear system in (7.8). Assume there exist smooth
state feedback laws ξ
1
= α(z) and ξ
2
= β(z) satisfying the following conditions:
i) α(0) = 0 and β(0) = 0.
ii) For
˙ z = f(z, α(z), 0)
z = 0 is GAS.
220
iii) (ξ
1
, ξ
2
) = (α(z), β(z)) is an invariant manifold for (7.8), i.e. α(z), β(z) satisfy
the equation
β(z) =
∂α(z)
∂z
f(z, α(z), β(z))
iv) For
˙ z = f(z, α(z), β(z))
the origin z = 0 is GAS.
Find suﬃcient conditions such that there exist state feedback laws ξ
2
= k
1
(z, ξ
1
) and
u = k
2
(z, ξ
1
, ξ
2
) which semiglobally/globally asymptotically stabilize the origin for the
(z, ξ
1
)subsystem and the composite system (7.8), respectively.
Remark 7.1.2. Based on condition iii) in problem 7.1.3, ξ
2
= β(z) is a ﬁxed point of
equation
ξ
2
=
∂α(z)
∂z
f(z, α(z), ξ
2
)
Thus, α(z), β(z) are not arbitrary controllers. The state feedback α(z) must be chosen
such that it both stabilizes the zerodynamics locked at ξ
2
= 0 and the preceding ﬁxed
point equation must be wellposed (i.e. have a solution). Both of these issues are
addressed later in this chapter.
A common example of nonlinear systems with feedforward structures is a chain of
integrators with higherorder perturbations as
˙ x
1
= x
2
+ p
1
(x
2
, x
3
, . . . , x
n
, u)
˙ x
2
= x
3
+ p
2
(x
3
, . . . , x
n
, u)
˙ x
n
= u + p
n
(x
n
, u)
(7.12)
System (7.12) can be rewritten as
˙ x = Ax +Bu + p(x, u) (7.13)
(with obvious deﬁnitions of A, B, p(x, u)) where the perturbation terms p
i
are at least
quadratic with respect to their arguments [102]. Nonlinear systems in feedforward
forms can have more general forms as the following
˙ x
1
= f
1
(x
2
, x
3
, . . . , x
n
, u)
˙ x
2
= f
2
(x
3
, . . . , x
n
, u)
˙ x
n
= f
n
(x
n
, u)
(7.14)
with an apparent upper triangular structure w.r.t. variables (x
1
, x
2
, . . . , x
n
, u) [56].
Feedforward systems of types (7.12) and (7.14) are studied by Teel and MazencPraly.
Similar to systems in feedback forms recursive procedures called feedforwarding are
developed for stabilization of feedforward systems [102, 101, 56]. The main draw back
221
of all existing results on stabilization of feedforward systems is the lowgain nature
of controllers that leads to relatively poor performance of control laws obtained from
feedforwarding. Here, the performance measure is the settling time of the solutions
of the closedloop system. The controllers obtained using backstepping procedure do
not suﬀer from this drawback.
7.2 Structure of Nontriangular Normal Forms of
Underactuated Systems
The classiﬁcation of underactuated mechanical systems presented in section 4.6 allows
us to identify the structure of the vector ﬁeld f in (7.8) (or (7.9)) for underactuated
systems in nontriangular forms and feedforward forms.
Proposition 7.2.1. All the normal forms for classes of underactuated systems in
nontriangular/feedforward forms with a doubleintegrator linear part (minus the per
turbation terms ϕ
i
) are special classes of the following normal form
˙ z
1
= z
2
+ g
0
(ξ
1
)ξ
2
˙ z
2
= f
0
(z
1
, ξ
1
) + z
T
2
g
11
(ξ
1
)z
2
+ z
T
2
g
12
(ξ
1
)ξ
2
+ ξ
T
2
g
22
(ξ
1
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.15)
where z
1
, z
2
∈ R
n
, ξ
1
, ξ
2
, u ∈ R
m
. In addition, the functions g
0
: R
m
→ R
n×n
,
f
0
(z
1
, ξ
1
) : R
n
R
m
→ R
n
and cubic matrices g
11
: R
m
→ R
n×n×n
, g
12
: R
m
→
R
n×m×n
, and g
22
: R
m
→R
m×m×n
are all smooth and g
0
(0) = 0.
Proof. The proof is by the deﬁnition of each class in section 4.6.
Remark 7.2.1. Equation (7.15) can be rewritten in the form
˙ z = F(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.16)
where
F(z, ξ
1
, ξ
2
) = f(z
1
, ξ
1
) + L(ξ
1
)ξ
2
+ (z
2
, ξ
2
)Q(ξ
1
)(z
T
2
, ξ
T
2
)
T
(7.17)
has a linearquadratic structure with respect to ξ
2
with a cubic matrix Q(ξ
1
).
Remark 7.2.2. Global stabilization of the core reduced system
˙ z
1
= z
2
˙ z
2
= f
0
(z
1
, ξ
1
)
(7.18)
is addressed in section 4.7.
The following special classes of normal form (7.15) are important due to their
application in control of underactuated systems:
222
i) g
11
, g
22
= 0: nonlinear system (7.15) is in nontriangular form with a vector ﬁeld
aﬃne in ξ
2
.
ii) f
0
(z
1
, ξ
1
) = f
0
(ξ
1
), g
11
= 0: nonlinear system (7.15) is in feedforward form with
a vector ﬁeld quadratic in ξ
2
.
iii) g
0
= 0: nonlinear system (7.15) is in nontriangular form with a vector ﬁeld
quadratic in ξ
2
. Moreover, ˙ z
1
does not depend on (ξ
1
, ξ
2
).
Lemma 7.2.1. Normal form (7.15) can be represented as
˙ z = f(z, ξ
1
) + g(z, ξ
1
, ξ
2
)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.19)
where f, g are explicitly given by
f(z, ξ
1
) =
¸
z
2
f
0
(z
1
, ξ
1
) + z
T
2
g
11
(ξ
1
)z
2
, g(z, ξ
1
, ξ
2
) =
¸
g
0
(ξ
1
)
z
T
2
g
12
(ξ
1
) +ξ
T
2
g
22
(ξ
1
)
Proof. The proof is by deﬁnition of f and g.
Remark 7.2.3. The main condition in problems 7.1.2 and 7.1.3 is that there should
exist a state feedback ξ
1
= α(z) with α(0) = 0 such that for the closedloop zero
dynamics locked at ξ
2
= 0
˙ z = F(z, α(z), 0)
the equilibrium z = 0 is globally asymptotically stable. Based on lemma 7.2.1, this
is equivalent to the existence of a globally stabilizing state feedback for ˙ z = f(z, ξ
1
)
that can be expressed as
˙ z
1
= z
2
˙ z
2
= f
0
(z
1
, ξ
1
) +z
T
2
g
11
(ξ
1
)z
2
(7.20)
Apparently, if g
11
≡ 0, then global stabilization of (7.20) is equivalent to global
stabilization of the core reduced system which is addressed in section 4.7. In addition,
for the special case where g
11
= g
11
(ξ
1
, ) with the property g
11
(ξ
1
, 0) = 0, semiglobal
stabilization of (7.20) can be achieved (under appropriate technical conditions).
Deﬁnition 7.2.1. (slightly nontriangular nonlinear systems) The following subclass
of nontriangular nonlinear systems in (7.19)
˙ z = f(z, ξ
1
) + g(z, ξ
1
, ξ
2
, )ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.21)
with the property g(z, ξ
1
, ξ
2
, 0) = 0 are called slightly nontriangular nonlinear systems.
223
Control design for slightly nontriangular systems is important due to their appli
cations in aerospace vehicles. As an example, according to proposition 5.11.4, the
normal form of the ﬁrstlevel approximate model of a helicopter is in the form
˙ z = f(z) + g
1
(ξ
1
)T + g
2
(ξ
1
, ξ
2
, ) +g
3
(ξ
1
, )u
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.22)
where z = col(x, v) ∈ R
6
, ξ
1
= col(φ, θ, ψ), (i.e. the three Euler angles), ξ
2
∈ R
3
, and
= (
1
,
2
,
3
)
T
∈ R
3
. Furthermore, both g
2
and g
3
vanish at = (0, 0, 0). Later, we
provide more examples of slightly nontriangular systems in robotics applications.
7.3 Stabilization of Nonlinear Systems in Feedback
Form
Backstepping procedure is an eﬀective method for global stabilization of nonlinear
systems in strict feedback form. There are two types of backstepping procedure. One
that makes an explicit use of a Lyapunov function and is referred to as standard
backstepping procedure. The other one is a type of backstepping procedure which
does not require any explicit knowledge of a Lyapunov function and we call it cascade
backstepping procedure. Here, we present both of them and clarify the diﬀerences
between the two methods.
7.3.1 Standard Backstepping Procedure
We start by presenting the standard backstepping procedure for a nonlinear system
augmented with a single integrator at the input. First, we need to state the following
lemma [36, 40].
Lemma 7.3.1. Assume f(x, y) : R
n
R
m
→ R is a continuously diﬀerentiable
function. Then, the following identity holds:
f(x, y) = f(x, 0) +g(x, y)y (7.23)
where
g(x, y) =
1
0
D
y
f(x, sy)ds
Proof. Fix (x, y) ∈ R
n+m
and let h(s) := f(x, sy) where s ∈ R. We have h
(s) =
D
y
f(x, sy) y. Hence
h(1) −h(0) =
1
0
h
(s)ds = (
1
0
D
y
f(x, sy)ds) y
and the result follows.
224
Remark 7.3.1. An application of lemma 7.3.1 gives
f(x, y +δ) = f(x, y) + g(x, y, δ)δ
The following version of backstepping theorem can be found in [36].
Theorem 7.3.1. (standard backstepping) Consider the following nonlinear system
˙ z = f(z, ξ)
˙
ξ = u
(7.24)
where z ∈ R
n
, ξ, u ∈ R, and f is a smooth function satisfying f(0, 0) = 0. Assume
there exists a smooth state feedback ξ = α(z) : R
n
→R satisfying α(0) = 0 such that
for
˙ z = f(z, α(z))
z = 0 is GAS. Let V (z) be a smooth positive deﬁnite and proper Lyapunov function
associated with the closedloop zsubsystem satisfying
DV (z) f(z, α(z)) < 0, ∀z = 0
Then, the following smooth state feedback
u = −c(ξ −α(z)) +
∂α(z)
∂z
f(z, ξ
1
, ξ
2
) −
∂V (z)
∂z
h(z, µ) (7.25)
renders (z, ξ) = 0 GAS for (7.24) where c > 0 is a constant, µ := ξ −α(z), and
h(z, µ) :=
1
0
D
ξ
f(z, α(z) +sµ)ds
satisﬁes
f(z, α(z) +µ) = f(z, α(z)) + h(z, µ)µ
Proof. Consider the following Lyapunov function candidate
W(z, ξ) = V (z) +
1
2
µ
2
Calculating
˙
W along the solutions of the closedloop system in (7.24), we get
˙
W =
˙
V + ˙ µµ,
=
∂V (z)
∂z
f(z, α(z) + µ) + (u −
∂α(z)
∂z
f(z, ξ
1
, ξ
2
))µ,
=
∂V (z)
∂z
f(z, α(z)) + [u −
∂α(z)
∂z
f(z, ξ
1
, ξ
2
) +
∂V (z)
∂z
h(z, µ)]µ, (lemma 7.3.1)
=
∂V (z)
∂z
f(z, α(z)) −cµ
2
< 0
225
for all (z, µ) = 0. Therefore, (z, ξ) = 0 is GAS for (7.24) and W(z, ξ) is a valid
Lyapunov function for the composite system.
Remark 7.3.2. Successive application of theorem 7.3.1 proves the statements in the
orem 7.1.1.
In [57], the following theorem which is a special case of theorem 7.3.1 (with a
zsubsystem aﬃne in ξ) is considered as standard backstepping procedure.
Theorem 7.3.2. ([57]) Consider the following cascade nonlinear system aﬃne in ξ
˙ z = f(z) + g(z)ξ
˙
ξ = u
(7.26)
where z ∈ R
n
, ξ, u ∈ R, and f, g are smooth functions with f(0) = 0. Let the state
feedback ξ = α(z) with α(0) = 0 render z = 0 GAS for the closedloop zsubsystem
˙ z = f(z) +g(z)α(z)
Then, the following state feedback
u = −c(ξ −α(z)) +
∂α(z)
∂z
f(z, ξ
1
, ξ
2
) −
∂V (z)
∂z
g(z) (7.27)
renders (z, ξ) = 0 GAS for the composite system (7.26).
Proof. Deﬁne
˜
f(z, ξ) := f(z) + g(z)ξ. Then
˜
f(z, ξ +µ) =
˜
f(z, ξ) + h(z, µ)µ
where h(z, µ) := g(z). Thus, state feedback (7.27) reduces to (7.25) and the result
follows from theorem 7.3.1.
The proof of standard backstepping procedure automatically provides a Lyapunov
function W(z, ξ
1
) for the closedloop composite system. This is very valuable for
(possible) robustness analysis of perturbed nonlinear systems in feedback forms (e.g.
normal forms of ClassVI underactuated mechanical systems in chapter 4).
7.3.2 Cascade Backstepping Procedure
The knowledge about the Lyapunov function V (z) for the closedloop zsubsystem
is necessary in the computation of the control law (7.25) for the composite system.
Such a knowledge about V (z) is not always available. This motivates developing a
backstepping procedure which is not Lyapunovbased. We refer to this procedure that
does not require explicit knowledge of any Lyapunov functions as cascade backstepping
procedure.
The most important tool in developing a nonLyapunovbased or cascade back
stepping procedure in the following theorem due to Sontag.
226
Theorem 7.3.3. (global stability of cascade systems [83]) Consider the following
cascade nonlinear system
˙ z = f(z, η)
˙ η = g(η)
(7.28)
where f : R
n
R
m
→R
n
and g : R
m
→R
m
are smooth functions satisfying f(0, 0) = 0
and g(0) = 0. Assume the following conditions hold:
i) For ˙ z = f(z, 0), z = 0 is globally asymptotically stable.
ii) For ˙ η = g(η), η = 0 is globally asymptotically and locally exponentially stable.
iii) For any solution η(t) with initial condition η
0
∈ R
m
, the solution of the z
subsystem of (7.28) exists for all t ≥ 0 and is bounded.
Then, (z, η) = 0 is globally asymptotically stable for (7.28).
The main use of theorem 7.3.3 is in stability analysis of the class of cascade systems
˙ x = f(x, η)
˙ η = g(η, u)
(7.29)
where a control law u = k(η) is available that render η = 0 GAS for ˙ η = g(η, k(η)).
In this chapter, we frequently use theorem 7.3.3 in the aforementioned context.
The only nonconstructive aspect of theorem 7.3.3 is in its third condition where
the boundedness of the solutions of the zsubsystem of (7.28) is assumed. One pos
sible assumption to guarantee this boundedness assumption is to assume that the
zsubsystem
˙ z = f(z, η)
is inputtostate stable, or simply ISSstable [82]. Due to SontagWang [84], a Lyapunov
type characterization of the ISS property reveals that the necessary and suﬃcient
condition for a system to be ISSstable is that there exist classK functions α
1
, α
2
and a positive deﬁnite smooth ISS Lyapunov function V (z) such that the following
property holds:
∂V (z)
∂z
f(z, η) < −α
1
(z), for z > α
2
(η) (7.30)
Condition (7.30) is an extremely restrictive condition which might not hold for large
classes of nonlinear systems of interest. Since, it essentially ignores the existence of
the input η. In the sense that over the region z > α
2
(η),
˙
V < 0 and the solution
of the zsubsystem converges towards the inner level surfaces around the origin z = 0,
until the condition z > α
2
(η) is violated. However, to guarantee the boundedness
of the solution of the zsubsystem it is suﬃcient that V (z(t)) stays bounded. The time
decay of V (z(t)) is unnecessary. Keeping this in mind, here we provide a constructive
condition that guarantees boundedness of the solutions of the zsubsystem. This
condition is less restrictive than the ISSstability property of the zsubsystem. This
is inﬂuenced by the work of Sepulchre et al. [80] (theorem 4.7, p.129) and generalizes
that result. First, we need to deﬁne a class of input disturbances.
227
Deﬁnition 7.3.1. (class(
0
functions) Deﬁne the following class of functions δ :
R
≥0
→R
n
(
0
(l
0
, l
1
) := ¦δ() [ δ
∞
≤ l
0
, δ
1
≤ l
1
, lim
t→∞
δ(t) = 0¦ , l
0
, l
1
> 0
where δ
∞
:= sup
t≥0
[δ(t)[ and δ
1
:=
∞
0
[δ(t)[dt denote L
∞
and L
1
norms of δ(),
respectively. We say a function δ : t → δ(t) is class(
0
with bounds l
0
, l
1
> 0, if
δ() ∈ (
0
(l
0
, l
1
).
Theorem 7.3.4. Consider the nonlinear system
˙ z = f(z, η) (7.31)
with a class(
0
input disturbance η() ∈ (
0
(l
0
, l
1
). Assume for any l
0
> 0, there exist
a smooth positive deﬁnite and proper function V (z), positive constants ρ ≥ 0, k > 0,
and λ ∈ (0, 1] such that the following conditions are satisﬁed:
i) ∇V (z) f(z, 0) ≤ 0
ii) [∇V (z) h(z, η)[ ≤ kV (z)
λ
for all z ≥ ρ and η ≤ l
0
where
h(z, η) =
1
0
D
η
f(z, sη)ds
Then, any solution of the system (7.31) is bounded.
Proof. Based on lemma 7.3.1, we have
f(z, η) = f(z, 0) + h(z, η)η
Calculating
˙
V gives
˙
V = ∇V (z) f(z, η)
= ∇V (z) f(z, 0) +∇V (z) h(z, η)η
≤ ∇V (z) h(z, η)η
≤ [∇V (z) h(z, η)[ η
≤ kV (z)
λ
η
We consider two cases: i) λ ∈ (0, 1) and ii) λ = 1. For λ ∈ (0, 1), over the region
z ≥ ρ, the last inequality can be rewritten as
1
k
V (z(t))
−λ
˙
V (z(t)) ≤ η(t)
Integrating both sides of the last inequality w.r.t. t gives
V (z(t))
(1−λ)
−V (z(0))
(1−λ)
≤ k(1 −λ)
t
0
η(t)dt ≤ k(1 −λ)η
1
≤ k(1 −λ)l
1
228
which means z(t) is bounded uniformly in t, because
V (z(t)) ≤ [V (z(0))
(1−λ)
+ k(1 −λ)l
1
]
1
1−λ
=: L(k, λ, l
1
, z(0)) (7.32)
and V (z) is a proper function. Thus, / := V
−1
([0, L(k, λ, l
1
, z(0)]) is a compact set
and z(t) ∈ /, ∀t ≥ 0. This implies z(t) < L
z
:= max¦ρ, r(/)¦, for all t > 0
where r(/) = max
x∈K
x denotes the radius of the compact set /. The proof of the
boundedness of z(t) for all t > 0 for the case of λ = 1 is very similar and is omitted.
The only diﬀerence is that the inequality (7.32) is in the form
V (z(t)) ≤ V (z(0)) exp(kl
1
) (7.33)
Now, we are ready to introduce a nonLyapunovbased backstepping procedure.
Theorem 7.3.5. (cascade backstepping procedure) Consider the following nonlinear
system
˙ z = f(z, ξ)
˙
ξ = u
(7.34)
Assume there exists a smooth state feedback ξ = α(z) with α(0) = 0 such that for
˙ z = f(z, α(z))
z = 0 is GAS. Let V (z) be a smooth positive deﬁnite and proper Lyapunov function
associates with the closedloop zsubsystem satisfying
∇V (z) f(z, α(z)) ≤ 0
In addition, suppose for any l
0
> 0, there exist ρ ≥ 0, k > 0, and λ ∈ (0, 1] such that
[∇V (z) h(z, µ)[ ≤ kV (z)
λ
, ∀(z, µ) : z ≥ ρ, µ ≤ l
0
(7.35)
where µ = ξ −α(z) and h(z, µ) =
1
0
D
ξ
f(z, α(z) +sµ)ds. Then, the following smooth
state feedback
u = −c(ξ −α(z)) +
∂α(z)
∂z
f(z, ξ), c > 0 (7.36)
renders (z, ξ) = 0 GAS for (7.34).
Proof. The closedloop cascade system in (7.34) with controller (7.36) can be rewrit
ten as
˙ z = f(z, α(z) + µ) =:
˜
f(z, µ)
˙ µ = −cµ
(7.37)
Notice that both µ = 0 is GAS for the µsubsystem and z = 0 is GAS for ˙ z =
˜
f(z, 0).
On the other hand, the solution of the zsubsystem is bounded according to theorem
7.3.4. To see this, note that µ(t) = µ(0) exp(−ct) and thus µ(t) is a class(
0
function
satisfying µ(t) ≤ l
0
:= µ(0), ∀t > 0 and µ(t)
1
= 1/c =: l
1
. But
f(z, α(z) +µ) = f(z, α(z)) + h(z, µ)µ
229
and by the assumption in the question we get
˙
V ≤ kV (z)
λ
µ
which guarantees uniform boundedness of the solution z(t) for all t > 0. Based on
theorem 7.3.3, the origin (z, µ) = 0 is GAS for the cascade closedloop system. This
implies (z, ξ) = 0 is GAS for (7.34).
Example 7.3.1. Consider the following nonlinear system in strict feedback form
˙ z
1
= z
2
˙ z
2
= σ(ξ)
˙
ξ = u
(7.38)
where σ(ξ) is a sigmoidal function that is odd and globally Lipschitz (e.g. tanh(ξ)).
The linear state feedback
ξ = α(z) := −k
1
z
1
−k
2
z
2
, k
1
, k
2
> 0
renders z = 0 GAS for the zsubsystem. To prove this, let us express the dynamics
of the closedloop zsubsystem as
˙ z
1
= z
2
˙ z
2
= −σ(k
1
z
1
+ k
2
z
2
)
(7.39)
Deﬁne φ(z
1
, z
2
) := σ(k
1
z
1
+ k
2
z
2
) and notice that D
z
2
φ(z
1
, z
2
) > 0 for all z ∈ R
2
.
This means
z
2
(φ(z
1
, 0) −φ(z
1
, z
2
)) < 0, ∀(z
1
, z
2
) ∈ R
2
` ¦(0, 0)¦
Let us deﬁne
ψ(z
1
) =
z
1
0
φ(s, 0)ds =
z
1
0
σ(k
1
s)ds
and consider the following Lyapunov function candidate
V (z
1
, z
2
) = ψ(z
1
) +
1
2
z
2
2
Calculating
˙
V along the solutions of the closedloop zsubsystem, we get
˙
V = z
2
(φ(z
1
, 0) −φ(z
1
, z
2
)) < 0, ∀(z
1
, z
2
) = (0, 0)
and therefore (z
1
, z
2
) = 0 is GAS for the zsubsystem. Taking f(z, ξ) := [z
2
, σ(ξ)]
T
,
we have
f(z, α(z) +µ) = f(z, α(z)) + h(z, µ)µ
where h(z, µ) = [0, g(z, µ)]
T
and g(z, µ) =
1
0
σ
(α(z) + sµ)ds. Since σ is globally
Lipschitz, there exists an L > 0 such that [g(z, µ)[ ≤ L for all (z, µ). Now, we prove
230
that based on theorem 7.3.5, the following nonlinear state feedback
u = −c(ξ −α(z)) +
∂α(z)
∂z
f(z, ξ), c > 0
which can be written in an explicit form
u = −c(ξ + k
1
z
1
+ k
2
z
2
) −k
1
z
2
−k
2
σ(ξ) (7.40)
renders (z, ξ) = 0 GAS for (7.38). For doing so, we need to show that condition (7.35)
holds. We have
[∇V (z) h(z, µ)[ = [z
2
g(z, µ)[ ≤ L[z
2
[ ≤ (2L)V (z)
1/2
Therefore, condition (7.35) holds with k = 2L, λ = 1/2 uniformly in µ. This guar
antees boundedness of the solution of the zsubsystem and the GAS property of
(z, ξ) = 0 for the composite system. Using standard backstepping procedure, one
obtains a diﬀerent globally stabilizing control law
u = −c(ξ −α(z)) +
∂α(z)
∂z
f(z, ξ) −E(z, ξ), c > 0
with an extra term E(z, ξ) compared to state feedback (7.40) which is given by
E(z, ξ) := ∇V (z) h(z, µ) =
1
0
z
2
σ
((1 −s)α(z) + sξ)ds
Remark 7.3.3. If the sigmoidal function σ in (7.38) is neither bounded nor globally
Lipschitz. Then, we choose the following control law
ξ = α(z) := −σ
0
(k
1
z
1
+ k
2
z
2
), k
1
, k
2
> 0
where σ
0
is a bounded and globally Lipschitz sigmoidal function. Now, ˜ σ(ξ) =
σ(σ
0
(ξ)) is both bounded and globally Lipschitz. The analysis in example 7.3.1 now
continues in a similar way with replacing σ by ˜ σ.
7.4 Stabilization of Nonlinear Systems in Nontri
angular Forms
The main class of nontriangular nonlinear systems considered in this chapter are in
the form
˙ z = f(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.41)
231
where z ∈ R
n
and f is a smooth function satisfying f(0, 0, 0) = 0. Here, we consider
the single input case with ξ
1
, ξ
2
, u ∈ R for the sake of simplicity of notations. The
generalization of the obtained results to the case of multiple inputs is trivial. De
pending on whether f(z, ξ
1
, ξ
2
) is an aﬃne function of ξ
1
and/or ξ
2
or none, we break
the problem of stabilization of (7.41) into a number of cases and address each case
separately. Throughout this chapter, we put especial emphasis on the structure of f
as it appears in the normal forms of underactuated mechanical systems.
7.4.1 Nontriangular Nonlinear Systems Aﬃne in (ξ
1
, ξ
2
)
In this section, our goal is to stabilize normal form (7.41) in its simplest possible form
given as
˙ z = f(z) + g
1
(z)ξ
1
+ g
2
(z)ξ
2
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.42)
where (z, ξ
1
) ∈ R
n
Rand ξ
2
∈ R is the control input. The vector ﬁelds f(z), g
1
(z), g
2
(z)
are all smooth and f(0) = 0. Due to the linearity property of the vector ﬁled of the z
subsystem w.r.t. ξ
2
, normal form (7.42) does not capture the eﬀects of Coriolis and
centrifugal terms in mechanical systems. However, it facilitates explaining the main
necessary steps that are required to address the stabilization problem of systems in
nontriangular forms nonaﬃne in (ξ
1
, ξ
2
), without excessive technical preliminaries.
More importantly, it allows us to show that the method of standard backstepping
and its particular way of construction of Lyapunov functions is not generalizable to
nonlinear systems in normal form (7.42). In contrast, we provide a cascade backstep
ping procedure for stabilization of nonlinear systems in the form (7.42). Later, we
present generalization of this nonLyapunovbased backstepping procedure to classes
of normal form (7.41) that are nonaﬃne in (ξ
1
, ξ
2
).
First, we demonstrate why the standard backstepping procedure is practically
not applicable to (7.42). In the sense that, it requires conditions that are extremely
restrictive.
Proposition 7.4.1. Consider the nonlinear system in (7.42). Assume there exists a
smooth state feedback α(z) : R
n
→R with α(0) = 0 such that for
˙ z = f(z) + g
1
(z)α(z) =: f
1
(z) (7.43)
z = 0 is GAS. Let V (z) be the smooth positive deﬁnite and proper Lyapunov function
associated with (7.43) satisfying L
f
1
V (z) < 0, ∀z = 0. Assume α(z) globally satisﬁes
the condition L
g
2
α(z) < 1. Deﬁne β(z) as
β(z) :=
L
f
1
α(z)
1 −L
g
2
α(z)
so that ξ
2
= β is the ﬁxedpoint solution of the equation
ξ
2
= ∇α(z) [f(z) + g
1
(z)α(z) + g
2
(z)ξ
2
]
232
Suppose that L
g
2
V (z) = 0, ∀z ∈ R
n
. Then, the following statements hold:
i) For the closedloop zsubsystem, ˙ z = f(z) +g
1
(z)α(z) +g
2
(z)β(z) =: f
2
(z), the
origin z = 0 is GAS.
ii) there exist a smooth state feedback ξ
2
= K
1
(z, ξ
1
) that renders (z, ξ
1
) = 0 GAS
for the (z, ξ
1
)subsystem of (7.42).
iii) there exists a smooth state feedback u = K
2
(z, ξ
1
, ξ
2
) that renders (z, ξ
1
, ξ
2
) = 0
GAS for the composite system (7.42).
Proof. The proof of i) is trivial due to the assumption L
g
2
V (z) = 0. This implies
˙
V = L
f
2
V (z) = L
f
1
V (z) + β(z)L
g
2
V (z) = L
f
1
V (z) < 0
for all z = 0 and therefore z = 0 is GAS. To prove ii), ﬁrst we need to apply a change
of coordinates by setting µ = ξ
1
− α(z). The dynamics of the (z, ξ
1
)subsystem of
(7.42) in new coordinates can be expressed as
˙ z = f
2
(z) + [g
1
(z) + A(z)g
2
(z)]µ
1
+ B(z)g
2
(z)µ
2
˙ µ
1
= µ
2
(7.44)
where µ
2
is the new control and the functions A(z), B(z) : R
n
→ R are obtained as
follows. By deﬁnition of µ
2
, we have
µ
2
= ξ
2
−∇α(z) [f(z) + g
1
(z)ξ
1
+ g
2
(z)ξ
2
]
= (1 −L
g
2
α(z))ξ
2
−L
f
1
α(z) −µ
1
L
g
1
α(z)
This change of variable is globally invertible due to 1 −L
g
2
α(z) > 0. Solving the last
equation for ξ
2
gives
ξ
2
=
L
f
1
α(z) + L
g
1
α(z)µ
1
+ µ
2
1 −L
g
2
α(z)
= β(z) + A(z)µ
1
+B(z)µ
2
(7.45)
where
A(z) =
L
g
1
α(z)
1 −L
g
2
α(z)
, B(z) =
1
1 −L
g
2
α(z)
Substituting ξ
2
from (7.45) in equation (7.42) gives (7.44). Following the proof the
standard backstepping theorem, let us consider the candidate Lyapunov function
W(z, ξ
1
) := V (z) +
1
2
µ
2
1
for the nonlinear system in (7.44). Calculating
˙
W, we get
˙
W =
˙
V + µ
1
µ
2
= ∇V (z) [f(z) + g
1
(z)(α(z) +µ
1
) + g
2
(z)(β(z) + A(z)µ
1
+B(z)µ
2
)
= L
f
2
V (z) + [L
g
1
V (z) + A(z)L
g
2
V (z) + µ
2
]µ
1
+ B(z)(L
g
2
V (z))µ
2
233
Unlike the case of systems in strict feedback form, for the case of nontriangular
nonlinear systems, an extra term δ = B(z)(L
g
2
V (z))µ
2
appears in the expression of
˙
W which is sign indeﬁnite. Unless δ = 0, or L
g
2
V (z) = 0 for all z,
˙
W cannot be made
negative deﬁnite. But by assumption, L
g
2
V (z) = 0 and applying the state feedback
µ
2
= −c
1
µ
1
−L
g
1
V (z), c
1
> 0
renders
˙
W negative deﬁnite for all (z, µ
1
) = 0 because
˙
W = L
f
2
V (z) −c
1
µ
2
1
This means that
ξ
2
= K
1
(z, ξ
1
) := β(z) + [A(z) −B(z)c
1
](ξ
1
−α(z)) −B(z)L
g
1
V (z) (7.46)
renders (z, ξ
1
) = 0 GAS for the (z, ξ
1
)subsystem of (7.42) and W(z, ξ
1
) is a valid
Lyapunov function for the (z, ξ
1
)subsystem. Finally, part iii) follows from part ii)
according to the standard backstepping theorem.
Remark 7.4.1. The most restrictive condition in theorem is the assumption that
L
g
2
V (z) = 0 for all z ∈ R
n
. However, without this assumption, in the opinion of
the author, there does not seem to exist any ways to render
˙
W negative deﬁnite
for all (z, ξ
1
) = 0. This demonstrates the weakness of the particular choice of the
Lyapunov function used in the standard backstepping procedure, in dealing with
nonlinear systems in nontriangular forms.
The following result is the nonLyapunovbased version of proposition 7.4.1.
Proposition 7.4.2. Assume all the conditions in proposition 7.4.1 hold. In addition,
suppose there exist ρ ≥ 0, k > 0, and λ ∈ (0, 1] such that
[L
g
1
V (z)[ ≤ kV (z)
λ
, z ≥ ρ
Then, the smooth state feedback
ξ
2
= K(z, ξ
1
) = β(z) +G(z)(ξ
1
−α(z)) (7.47)
(which does not depend on the Lyapunov function V (z)) renders (z, ξ
1
) = 0 GAS for
the (z, ξ
1
)subsystem of (7.42) where the nonlinear gain G(z) is deﬁned as
G(z) = A(z) −c
1
B(z), c
1
> 0
Proof. Setting µ
2
= −c
1
µ
1
, from equation (7.44), we have ˙ µ
1
= −c
1
µ
1
(thus µ
1
= 0
is globally exponentially stable) and
˙
V = L
f
2
V + (L
g
1
V )µ
1
+ G(z)L
g
2
V
= L
f
2
V + (L
g
1
V )µ
1
≤ [L
g
1
V (z)[[µ
1
[
≤ kV (z)
λ
[µ
1
[
234
Based on theorem 7.3.4, the solutions of the zsubsystem are bounded. Now, accord
ing to theorem 7.3.3, the origin (z, µ
1
) = 0 is GAS for the cascade nonlinear system
(7.44) which proves the statement in the question.
Now, we would like to present a nonLyapunovbased backstepping procedure
for stabilization of nontriangular nonlinear system (7.42) which does not require the
restrictive condition L
g
2
V (z) = 0, ∀z ∈ R
n
. First, we need to make some assumptions.
Assumption 7.4.1. There exists a state feedback α(z) : R
n
→ R with α(0) = 0
satisfying the following conditions:
i) for ˙ z = f(z) + g
1
(z)α(z) =: f
1
(z), the origin z = 0 is globally asymptotically
stable,
ii) there exists L
1
> 0 such that [L
g
1
α(z)[ ≤ L
1
for all z ∈ R
n
,
iii) there exists L
2
> 0 such that L
g
2
α(z) ≤ 1 −1/L
2
for all z ∈ R
n
.
The following proposition is the main analytical tool that justiﬁes the use of the
cascade backstepping procedure for stabilization of a nontriangular system as (7.42).
Proposition 7.4.3. Consider the nonlinear system in (7.42) and suppose assumption
7.4.1 holds. Then, after applying the global change of coordinates µ = Φ(z, ξ) and
change of control deﬁned by
µ
1
= ξ
1
−α
1
(z)
µ
2
= M(z)(ξ
2
−α
2
(z, ξ
1
))
w = M(z)(u −α
3
(z, ξ
1
, ξ
2
))
(7.48)
with M(z) = 1−L
g
2
α(z) > 1/L
2
> 0, the dynamics of the nonlinear system in (7.42)
transforms into
˙ z = f
0
(z) + [g
1
(z) + A(z)g
2
(z)]µ
1
+ B(z)g
2
(z)µ
2
˙ µ
1
= µ
2
˙ µ
2
= w
(7.49)
where f
0
(z) := f(z) + g
1
(z)α
1
(z) + g
2
(z)α
2
(z, α
1
(z)) and α
i
’s are deﬁned recursively
as the following
α
1
(z) := α(z)
α
2
(z, ξ
1
) := [L
f
α
1
(z) + ξ
1
L
g
1
α
1
(z)]/M(z)
α
3
(z, ξ
1
, ξ
2
) := [
∂α
2
∂z
(f(z) + g
1
(z)ξ
1
+ g
2
(z)ξ
2
) +
∂α
2
∂ξ
1
ξ
2
] −
˙
M
M
[ξ
2
−α
2
(z, ξ
1
)]
with
˙
M(z, ξ
1
, ξ
2
) := L
f
M(z) + ξ
1
L
g
1
M(z) + ξ
2
L
g
2
M(z)
235
In addition, ξ = Ψ(z, µ), the inverse of Φ, is given in closedform by
ξ
1
= α(z) + µ
1
ξ
2
= β(z) + A(z)µ
1
+ B(z)µ
2
(7.50)
where
β(z) = α
2
(z, α
1
(z)), A(z) =
L
g
1
α(z)
M(z)
, B(z) =
1
M(z)
(7.51)
Moreover, A(z), B(z) satisfy the following bounds
[A(z)[ ≤ L
1
L
2
, 0 < B(z) ≤ L
2
for all z ∈ R
n
(the constants L
1
, L
2
> 0 are deﬁned in assumption 7.4.1).
Proof. Let us calculate the ﬁrst two time derivatives of µ
1
= ξ
1
−α(z). We get
µ
2
:= ˙ µ
1
= ξ
2
−[L
f
α(z) + ξ
1
L
g
1
α(z) + ξ
2
L
g
2
α(z)]
= (1 −L
g
2
α(z))ξ
2
−[L
f
α(z) + ξ
1
L
g
1
α(z)]
= M(z)(ξ
2
−
L
f
α(z) + ξ
1
L
g
1
α(z)
M(z)
)
Now, according to the deﬁnition of α
2
(z, ξ
1
)
µ
2
= M(z)(ξ
2
−α
2
(z, ξ
1
))
Denoting w := ˙ µ
2
, w can be calculated as
w = M(z)(u − ˙ α
2
) +
˙
M(ξ
2
−α
2
(z, ξ
1
)
= M(z)(u −α
3
(z, ξ
1
, ξ
2
))
where
α
3
(z, ξ
1
, ξ
2
) := ˙ α
2
(z, ξ
1
, ξ
2
) −
˙
M
M
(ξ
2
−α
2
(z, ξ
1
))
So far, we proved that the new variables µ
1
, µ
2
, w satisfy the chain of integrators
˙ µ
1
= µ
2
, ˙ µ
2
= w
To calculate the inverse of µ = Φ(z, ξ), we need to solve µ = Φ(z, ξ) in (ξ
1
, ξ
2
). One
can solve for ξ
1
as ξ
1
= α(z) +µ
1
. Now, solving for ξ
2
from the deﬁnition of µ
2
gives
ξ
2
= α
2
(z, ξ
1
) +
1
M(z)
µ
2
= α
2
(z, α(z) + µ
1
) + B(z)µ
2
= α
2
(z, α
1
(z)) + A(z)µ
1
+ B(z)µ
2
which after denoting β(z) = α
2
(z, α
1
(z)), it agrees with the deﬁnition of β(z) in the
236
question. Substituting ξ
1
, ξ
2
in the ﬁrst equation of (7.42) shows that
˙ z = f(z) + g
1
(z)(α(z) + µ
1
) + g
2
(z)(β(z) +A(z)µ
1
+ B(z)µ
2
)
= f
0
(z) + [g
1
(z) + A(z)g
2
(z)]µ
1
+ B(z)g
2
(z)µ
2
and this completes the proof of equation (7.49). The bounds for A(z) and B(z) follow
from parts ii) and iii) of assumption 7.4.1.
The following assumption guarantees that the zerodynamics of the cascade non
linear system in (7.49) with the output y = µ
1
is globally minimum phase.
Assumption 7.4.2. Assume for
˙ z = f(z) + g
1
(z)α(z) + g
2
(z)β(z) =: f
0
(z) (7.52)
the origin z = 0 is globally asymptotically stable.
Assumption 7.4.3. Let V (z) be a smooth positive deﬁnite and proper Lyapunov
function satisfying L
f
0
V (z) ≤ 0. Suppose there exists ρ ≥ 0, k > 0, and λ ∈ (0, 1]
such that the following conditions are satisﬁed
[L
g
i
V (z)[ ≤ kV (z)
λ
, ∀z ≥ ρ, i = 1, 2 (7.53)
Here is out main result on global asymptotic stabilization of nontriangular non
linear systems aﬃne in (ξ
1
, ξ
2
):
Theorem 7.4.1. Consider the nonlinear system in (7.42). Suppose assumptions
7.4.1, 7.4.2, and 7.4.3 hold. Then, the following smooth state feedback in the form of
a PD controller with nonlinear gains
u = α
3
(z, ξ
1
, ξ
2
) −
c
1
M(z)
(ξ
1
−α
1
(z)) −
c
2
M(z)
(ξ
2
−α
2
(z, ξ
1
)), c
1
, c
2
> 0 (7.54)
that globally asymptotically stabilizes (z, ξ
1
, ξ
2
) = 0 for the nontriangular nonlinear
system (7.42) with M(z) = 1 −L
g
2
α(z).
Proof. Based on proposition 7.4.3, nonlinear system (7.42) can be transformed into
the cascade system
˙ z = f
0
(z) + [g
1
(z) +A(z)g
2
(z)]µ
1
+ B(z)g
2
(z)µ
2
=: F(z, µ
1
, µ
2
)
˙ µ
1
= µ
2
˙ µ
2
= w
(7.55)
such that for ˙ z = F(z, 0, 0) = f
0
(z), z = 0 is GAS. Let us stabilize the µsubsystem
with a PD controller
w = −c
1
µ
1
−c
2
µ
2
, c
1
, c
2
> 0
Then, µ = 0 is globally exponentially stable. Based on theorem 7.3.3, to show that
(z, µ) = 0 is globally asymptotically stable for the closedloop cascade system in
237
(7.55), we need to prove that any solution of the zsubsystem is bounded. For doing
so, let us calculate
˙
V along the solutions of (7.55). One obtains
˙
V = L
f
0
V (z) + [L
g
1
V (z) + A(z)L
g
2
V (z)]µ
1
+B(z)(L
g
2
V (z))µ
2
≤ ([L
g
1
V (z)[ +[A(z)[[L
g
2
V (z)[)[µ
1
[ +B(z)[L
g
2
V (z)[[µ
2
[
≤ (1 +L
1
L
2
+ L
2
)kV (z)
λ
µ (because [µ
i
[ ≤ µ, i = 1, 2)
=
˜
kV (z)
λ
µ
where
˜
k := 1 + L
1
L
2
+ L
2
. On the other hand, µ(t) is exponentially vanishing and
µ()
1
< ∞. Thus, from the proof of theorem 7.3.4, it follows that any solution
of the zsubsystem is bounded. The PD feedback w = −c
1
µ
1
− c
2
µ
2
in the original
coordinate can be written as
u = α
3
(z, ξ
1
, ξ
2
) −
c
1
M(z)
µ
1
−
c
2
M(z)
µ
2
which is a PD controller with nonlinear gains −c
i
/M(z), i = 1, 2.
Remark 7.4.2. The equation of α
3
in the state feedback (7.54) can be expressed in a
slightly more explicit way as follows. Deﬁne p(z), q(z) as
p(z) :=
L
f
α(z)
M(z)
, q(z) :=
L
g
1
α(z)
M(z)
and observe that they satisfy α
2
(z, ξ
1
) = p(z) + ξ
1
q(z). Then, the right hand side of
α
3
(z, ξ
1
, ξ
2
) = ˙ α
2
−
˙
M
M
2
µ
2
can be explicitly determined as the following
˙ α
2
= L
f
p(z) + ξ
1
L
g
1
q(z) + ξ
2
(q(z) + L
g
2
q(z)) (7.56)
˙
M = L
f
M(z) +ξ
1
L
g
1
M(z) + ξ
2
L
g
2
M(z) (7.57)
A crucial assumption in theorem 7.4.1 is assumption 7.4.2. In the following, we
provide suﬃcient conditions on g
2
(z) such that assumption 7.4.2 holds.
Proposition 7.4.4. Consider the following nonlinear system
˙ z = f(z) + g
1
(z)ξ
1
+g
2
(z)ξ
2
and assume the following conditions hold:
i) there exists a smooth state feedback ξ
1
= α(z) such that for
˙ z = f(z) + g
1
(z)α(z) =: f
1
(z)
z = 0 is GAS,
238
ii) L
g
2
α(z) < 1 for all z ∈ R
n
,
iii) there exists a function h(z) : R
n
→R such that g
2
(z) = h(z)f
1
(z).
Let
β(z) :=
L
f
1
α(z)
1 −L
g
2
α(z)
Then, for the closedloop system
˙ z = f(z) + g
1
(z)α(z) +g
2
(z)β(z) (7.58)
z = 0 is GAS.
Proof. The closedloop system with control (ξ
1
, ξ
2
) = (α(z), β(z)) can be written as
˙ z =
f
1
(z) + ∆(z)
1 −L
g
2
α(z)
(7.59)
where
∆(z) := (L
f
1
α(z))g
2
(z) −(L
g
2
α(z))f
1
(z)
But g
2
(z)  f
1
(z), or g
2
(z) = h(z)f
1
(z) implies
(L
f
1
α(z))g
2
(z) = (L
g
2
α(z))f
1
(z), ∀z ∈ R
n
Thus, ∆(z) ≡ 0. This means the closedloop nonlinear system in (7.59) reduces to
˙ z = S(z)f
1
(z) (7.60)
with a scaling factor S(z) =: (1 − L
g
2
α(z))
−1
> 0. Since S(z) is a positive deﬁnite
scalar, z = 0 is GAS for ˙ z = S(z)f
1
(z) and the result follows.
In the following result, we slightly remedy the restrictive condition g
2
(z)  f
1
(z)
in proposition 7.4.4.
Proposition 7.4.5. Consider the following nonlinear system
˙ z = f(z) + g
1
(z)ξ
1
+g
2
(z)ξ
2
and assume the following conditions hold:
i) there exists a smooth state feedback ξ
1
= α(z) such that for
˙ z = f(z) + g
1
(z)α(z) =: f
1
(z)
z = 0 is GAS.
ii) Let V (z) be a smooth positive deﬁnite and proper Lyapunov function associated
with ˙ z = f
1
(z) such that ∇V (z)f
1
(z) ≤ 0. Assume the largest invariant set in
Ω = ¦z [ ∇V (z)f
1
(z) = 0¦ is ¦0¦.
239
iii) L
g
2
α(z) < 1 for all z ∈ R
n
,
iv) Assume V (z) satisﬁes the inequality
(L
f
1
α(z))(L
g
2
V (z)) ≤ (L
g
2
α(z))(L
f
1
V (z)), ∀z ∈ R
n
where the equality is achieved for g
2
(z) = h(z)f
1
(z) with h(z) : R
n
→R.
Deﬁne
β(z) :=
L
f
1
α(z)
1 −L
g
2
α(z)
Then, for the closedloop system
˙ z = f(z) + g
1
(z)α(z) +g
2
(z)β(z) (7.61)
z = 0 is GAS.
Proof. The closedloop system with control (ξ
1
, ξ
2
) = (α(z), β(z)) is in the form
˙ z = S(z)(f
1
(z) + ∆(z)) (7.62)
where S(z) = (1 −L
g
2
α(z))
−1
and
∆(z) = (L
f
1
α(z))g
2
(z) −(L
g
2
α(z))f
1
(z)
Thus
L
∆
V (z) = (L
f
1
α(z))(L
g
2
V (z)) −(L
g
2
α(z))(L
f
1
V (z)) ≤ 0
which implies
˙
V = S(z)(L
f
1
V (z) + L
∆
V (z)) ≤ L
f
1
V (z) ≤ 0
but the largest invariant set in Ω is ¦0¦, therefore z = 0 is GAS for the closedloop
nonlinear system.
Example 7.4.1. Consider the following cascade nonlinear system which is in a non
triangular form aﬃne in (ξ
1
, ξ
2
)
˙ z
1
= z
2
−
z
2
ξ
2
2(1 +
z
2
1
+ z
2
2
)
˙ z
1
= ξ
1
+
tanh(z
1
+ z
2
)ξ
2
2(1 +
z
2
1
+z
2
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
We show that there exists a state feedback ξ
1
= α(z) which satisﬁes all the conditions
in theorem 7.4.1. and thus the last nonlinear system can be globally asymptotically
240
stabilized to the origin. Take α(z) = −tanh(z
1
+ z
2
). Then, for
˙ z = f
1
(z) :=
¸
z
2
−tanh(z
1
+ z
2
)
z = 0 is GAS. This is due to the fact that
V (z) =
z
1
0
tanh(s)ds +
1
2
z
2
2
(7.63)
is a valid positive deﬁnite Lyapunov function for ˙ z = f
1
(z). Since
L
g
2
α(z) =
∂α(z)
∂z
g
2
(z) = (1 −tanh
2
(z
1
+ z
2
))
(z
2
−tanh(z
1
+ z
2
))
2(1 +z)
,
we get
[L
g
2
α(z)[ ≤ [1 −tanh
2
(z
1
+ z
2
)[
1 +[z
2
[
2(1 +z)
≤
1
2
and thus 1/2 ≤ 1 −L
g
2
α(z) ≤ 3/2 (i.e. L
2
= 2 and 1/2 ≤ M(z) ≤ 3/2). In addition,
due to
[L
g
1
α(z)[ = [
∂α(z)
∂z
2
[ = 1 −tanh
2
(z
1
+ z
2
) ≤ 1
we obtain L
1
= 1. So far, we have proved that assumption 7.4.1 holds. Using propo
sition 7.4.4, we can show that assumption 7.4.2 holds as well. This is a consequence
of the fact that g
2
(z) satisﬁes the relation g
2
(z) = h(z)f
1
(z) with
h(z) :=
−1
2(1 +z)
Finally, assumption 7.4.3 can be veriﬁed by direct calculation as follows. The Lya
punov function V (z) in equation (7.63) satisﬁes
[L
g
1
V (z)[ = [
∂V (z)
∂z
2
[ = [z
2
[ ≤
√
2V (z)
1
2
, ∀z ∈ R
2
and
[L
g
2
V (z)[ =
[z
2
[tanh(z
1
+ z
2
) −tanh(z
1
)][
2(1 +z)
≤ [z
2
[ ≤
√
2V (z)
1
2
, ∀z ∈ R
2
The last inequality is due to the global Lipschitz property of tanh(x) which implies
[ tanh(z
1
+ z
2
) −tanh(z
1
)[ ≤ [z
2
[
Hence for k =
√
2, λ = 1/2, the following inequalities hold
[L
g
i
V (z)[ ≤ kV (z)
λ
, ∀z ∈ R
2
, i = 1, 2
241
Therefore, based on theorem 7.4.1, the following PD controller with nonlinear gains
u = α
3
(z, ξ
1
, ξ
2
) −
c
1
M(z)
(ξ
1
−α
1
(z)) −
c
2
M(z)
(ξ
2
−α
2
(z, ξ
1
)), c
1
, c
2
> 0
globally asymptotically stabilizes the origin (z, ξ
1
, ξ
2
) = 0 for the nontriangular non
linear system in the question.
7.4.2 Nontriangular Nonlinear Systems Nonaﬃne in (ξ
1
, ξ
2
)
In this section, we generalize our results on cascade backstepping procedure for sta
bilization of nontriangular nonlinear systems aﬃne in (ξ
1
, ξ
2
) to cascade nonlinear
systems that their vector ﬁelds are nonaﬃne in (ξ
1
, ξ
2
). We consider the following
class of partiallylinear cascade nonlinear systems
˙ z = F(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.64)
where F(z, ξ
1
, ξ
2
) : R
nz
RR →R
nz
is assumed to be a smooth function satisfying
F(0) = 0 and nonaﬃne in (ξ
1
, ξ
2
). The function F(z, ξ
1
, ξ
2
) can be rewritten as
F(z, ξ
1
, ξ
2
) = f(z) + g
1
(z, ξ
1
)ξ
1
+ g
2
(z, ξ
1
, ξ
2
)ξ
2
(7.65)
where
f(z) := F(z, 0, 0) (7.66)
g
1
(z, ξ
1
) :=
1
0
(D
ξ
1
F)(z, sξ
1
, 0)ds (7.67)
g
2
(z, ξ
1
, ξ
2
) :=
1
0
(D
ξ
2
F)(z, ξ
1
, sξ
2
)ds (7.68)
In this section, we provide suﬃcient conditions for global stabilization of nonlinear
systems in normal form (7.64). Depending on the particular structure of f, g
1
, g
2
a
number of important cases arise that will be discussed later in this chapter. The
importance of each case is due to its application as the normal form of special classes
of mechanical systems of interest. The main emphasis will be on control design for
nontriangular normal forms of underactuated systems. Before presenting our main
result, we need to make some assumptions.
Assumption 7.4.4. Assume there exists a smooth state feedback ξ
1
= α(z) with
α(0) = 0 that globally asymptotically stabilizes z = 0 for
˙ z = F(z, ξ
1
, 0) =: f
1
(z, ξ
1
) (7.69)
Let V (z) be the smooth positive deﬁnite and proper Lyapunov function associated
with the closedloop system such that ∇V (z) f
1
(z, α(z)) < 0, ∀z = 0.
242
Assumption 7.4.5. The exists a smooth function β(z) : R
n
→R with β(0) = 0 such
that the following conditions hold:
i) ξ
∗
2
= β(z) is the unique solution of the following ﬁxedpoint equation in ξ
2
ξ
2
=
∂α(z)
∂z
F(z, α(z), ξ
2
)
ii) ∇V (z) F(z, α(z), β(z)) < 0, ∀z = 0.
Assumption 7.4.6. Deﬁne
ψ(z, ξ
1
, ξ
2
) := −c
1
(ξ
1
−α(z)) +
∂α(z)
∂z
F(z, ξ
1
, ξ
2
) (7.70)
and assume there exists an a > 0 such that
∂ψ
∂ξ
2
=
∂α(z)
∂z
∂F(z, ξ
1
, ξ
2
)
∂ξ
2
≤ 1 −a, ∀(z, ξ
1
, ξ
2
)
This implies
M(z, ξ
1
, ξ
2
) := 1 −
∂ψ
∂ξ
2
≥ a > 0, ∀(z, ξ
1
, ξ
2
) (7.71)
Assumption 7.4.7. Assume the change of coordinates µ = T(z, ξ) given by
µ
1
= ξ
1
−α(z)
µ
2
= ξ
2
−ψ(z, ξ
1
, ξ
2
)
is a global diﬀeomorphism for each z ∈ R
n
with an inverse ξ = T
−1
(z, µ) as the
following
ξ
1
= µ
1
+ α(z)
ξ
2
= φ(z, µ
1
, µ
2
)
where φ(z, 0, 0) = β(z).
Now, the following result clariﬁes our purpose of making the last four assumptions.
Proposition 7.4.6. Consider the nonlinear system in (7.64) and suppose assump
tions 7.4.4 through 7.4.7 hold. Then, the change of coordinates µ = T(z, ξ) and the
change of control given by
µ
1
= ξ
1
−α(z)
µ
2
= ξ
2
−ψ(z)
w = M(z, ξ
1
, ξ
2
)(u −γ(z, ξ
1
, ξ
2
))
(7.72)
243
is globally invertible where
γ(z, ξ
1
, ξ
2
) =
D
z
ψ F(z, ξ
1
, ξ
2
) + D
ξ
1
ψ ξ
2
1 −D
ξ
2
ψ
(7.73)
In addition, it transforms the dynamics of (7.64) into the following cascade form
˙ z = F
0
(z, µ
1
, µ
2
)
˙ µ
1
= −c
1
µ
1
+ µ
2
˙ µ
2
= w
(7.74)
where
F
0
(z, µ
1
, µ
2
) = F(z, α(z) + µ
1
, φ(z, µ
1
, µ
2
))
Moreover, for
˙ z = F
0
(z, 0, 0) = F(z, α(z), β(z))
z = 0 is GAS.
Proof. Based on assumptions 7.4.6 and 7.4.7, µ = T(z, ξ) is a global diﬀeomorphism
and M(z, ξ
1
, ξ
2
) ≥ a > 0. Thus, the change of coordinates and control in (7.74) is
globally invertible. By deﬁnitions of µ
1
and ψ, we have
˙ µ
1
= ξ
2
−
∂α(z)
∂z
F(z, ξ
1
, ξ
2
) = −c
1
µ
1
+ µ
2
and because µ
2
= ξ
2
−ψ(z, ξ
1
, ξ
2
), one gets
˙ µ
2
= (1 −D
ξ
2
ψ)u −(D
z
ψ F(z, ξ
1
, ξ
2
) +D
ξ
1
ψ)
= M(z, ξ
1
, ξ
2
)(u −γ(z, ξ
1
, ξ
2
)) =: w
which means the µsubsystem has the following linear dynamics
˙ µ
1
= −c
1
µ
1
+µ
2
˙ µ
2
= w
global asymptotic stability of z = 0 for
˙ z = F
0
(z, 0, 0)
follows from part ii) of assumption 7.4.5.
Assumption 7.4.8. Rewrite F
0
(z, µ
1
, µ
2
) as
F
0
(z, µ
1
, µ
2
) = F
0
(z, 0, 0) + h
1
(z, µ
1
)µ
1
+ h
2
(z, µ
1
, µ
2
)µ
2
and assume for any l
1
, l
2
> 0, there exist ρ
i
≥ 0, k
i
> 0, λ
i
∈ (0, 1] for i = 1, 2 such
244
that V (z) satisﬁes
[∇V (z) h
1
(z, µ
1
)[ ≤ k
1
V (z)
λ
1
, z ≥ ρ
1
, [µ
1
[ ≤ l
1
[∇V (z) h
2
(z, µ
1
, µ
2
)[ ≤ k
2
V (z)
λ
2
, z ≥ ρ
2
, [µ
1
[ ≤ l
1
, [µ
2
[ ≤ l
2
(7.75)
Lemma 7.4.1. For any l > 0, take l
1
, l
2
≥ l and suppose assumption 7.4.8 holds.
Then, the following condition is satisﬁed
[∇V (z)h
i
[ ≤ kV (z)
λ
, z ≥ ρ, µ < l, i = 1, 2 (7.76)
with ρ = max¦ρ
1
, ρ
2
¦ ≥ 0, k = max¦k
1
, k
2
¦ > 0, and λ = max¦λ
1
, λ
2
¦ ∈ (0, 1].
Proof. Since [µ
i
[ ≤ µ for i = 1, 2 and the function ϕ(k, l) = kx
λ
for x > 0 is an
increasing function of λ (or k) uniformly in k (or λ) the result holds.
We are now ready to present out main result on global stabilization of nontrian
gular nonlinear systems nonaﬃne in (ξ
1
, ξ
2
):
Theorem 7.4.2. Consider the nonlinear system in (7.64). Suppose that assumptions
7.4.4 through 7.4.8 hold. Then, the following state feedback in explicit form
u = K(z, ξ
1
, ξ
2
) := γ(z, ξ
1
, ξ
2
) −
c
2
M(z, ξ
1
, ξ
2
)
(ξ
2
−ψ(z, ξ
1
, ξ
2
)), c
2
> 0 (7.77)
globally asymptotically stabilizes the origin (z, ξ
1
, ξ
2
) = 0 for (7.64).
Proof. According to proposition 7.4.6, after applying the change of coordinates and
control in equation (7.72), one obtains cascade system (7.74). The linear µsubsystem
of (7.74) can be globally exponentially stabilized using w = −c
2
µ
2
, c
2
> 0. The closed
loop µsubsystem is in the form ˙ µ = Aµ where
A =
¸
−c
1
1
0 −c
2
is a Hurwitz matrix with eigenvalues −c
1
, −c
2
< 0. In new coordinates, the control
u can be expressed as the following
u = γ(x) −
c
2
M(x)
(ξ
2
−ψ(x))
with x = col(z, ξ
1
, ξ
2
) as given in (7.77). Based on assumption 7.4.8 and lemma 7.4.1,
any solution of the closedloop zsubsystem in (7.74) with state feedback w = −c
2
µ
2
(or (7.77)) is bounded. Thus, based on theorem 7.3.3, (z, µ) = 0 is GAS for the
cascade nonlinear system in (7.74). This implies the origin (z, ξ
1
, ξ
2
) = 0 is GAS for
the closedloop nonlinear system (7.64).
245
7.4.3 Global Existence of FixedPoint Control Laws
A crucial assumption in the global version of the cascade backstepping procedure in
theorem 7.77 is that the ﬁxed point equation in ξ
2
of the form
ξ
2
=
∂α(z)
∂z
F(z, α(z), ξ
2
) =: Ψ(z, ξ
2
) (7.78)
(see assumption 7.4.5) globally (w.r.t z) has solutions in ξ
2
. More precisely, there
exists β : R
n
0
→R such that
β(z) = Ψ(z, β(z)), ∀z ∈ R
n
0
In this section, we provide suﬃcient conditions for global existence of a solution
ξ
∗
2
= β(z) for the ﬁxed point equation
ξ
2
= Ψ(z, ξ
2
) (7.79)
We refer to the solution β(z) of this ﬁxed point equation as a ﬁxed point control law.
Due to the structure of F(z, ξ
1
, ξ
2
) in equation (7.15), we focus on global existence of
ﬁxedpoint control laws for nontriangular systems with the following dynamics
˙ z
1
= z
2
˙ z
2
= f(z, ξ
1
, ξ
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= u
(7.80)
where
f(z, ξ
1
, ξ
2
) = f
0
(z, ξ
1
) + g
0
(z, ξ
1
, ξ
2
)ξ
2
2
(7.81)
Before presenting our main result, we need to make some deﬁnitions and give some
lemmas. Let us focus our attention to the following class of state feedback laws α(z).
Deﬁnition 7.4.1. (bounded functions with compact domain of attention) We say a
function α(z) : R
2n
→ R with α(0) = 0 and z = col(z
1
, z
2
) is bounded with compact
domain of attention around the origin w.r.t. z
2
if there exist a compact neighborhood
/
0
⊂ R
n
of z
2
= 0 and constants L
i
> 0, i = 0, 1, 2 such that
i) [α(z)[ ≤ L
0
, ∀z ∈ R
2n
ii) D
z
2
α(z) = 0, ∀z ∈ R
n
(R
n
` /
0
)
iii) D
z
2
α(z) ≤ L
1
, ∀z ∈ R
2n
iv) D
z
1
α(z) ≤ L
2
, ∀z ∈ R
2n
Deﬁnition 7.4.2. (bounded sigmoidal functions with ﬂat tails) We say σ(x) : R →R
is a bounded sigmoidal function with a ﬂat tail if the following conditions hold:
246
i) ∃L > 0 : [σ(x)[ ≤ L, ∀x.
ii) xσ(x) > 0, ∀x = 0 and σ(0) = 0.
iii) ∃I ⊂ R, 0 ∈ I : σ
(x) = 0, ∀x ∈ I.
Clearly, over the region R`I, σ(x) is constant (which explains the name “ﬂat tail”).
Example 7.4.2. (sigmoidal functions) A nonsmooth example of a sigmoidal function
with a ﬂat tail is the saturation function sat(x) = sgn(x) min¦[x[, 1¦. A C
1
smooth
bounded sigmoidal function with ﬂat tails is
σ(x) =
sin(
π
2
x) x ∈ [−1, 1]
1 x > 1
−1 x < −1
(7.82)
or in a more compact form
σ(x) = sin(
π
2
x) 1
I
(x) + (1 −1
I
(x)) sgn(x), I = [−1, 1] (7.83)
where 1
I
(x), sgn(x) denote the indicator function and the sign function, respectively.
In general, it is possible to construct a C
r
smooth bounded sigmoidal function with
ﬂat tails and (relatively) large linear region as
σ
f
(x; a) =
x [x[ ≤ a
sgn(x) P([x[) a < [x[ ≤ 1
sgn(x) [x[ > 1
(7.84)
where a ∈ (0, 1) is a constant threshold and P(x) is a polynomial of an appropriate
order. For example, with a = 0.75 and I = [−1, 1] the following 5th order polynomial
gives a C
2
smooth bounded sigmoidal function with ﬂat tails
P(x) = 768x
5
−3328x
4
+ 5728x
3
−4896x
2
+ 2080x −351 (7.85)
The function σ
f
(x; 0.75) with its ﬁrst two derivatives is depicted in Figure 71. In
this case, σ
f
(x; 0.75) is strictly increasing over the interval I = [−1, 1].
Example 7.4.3. Consider the following doubleintegrator with a saturation control
˙ z
1
= z
2
˙ z
2
= sat(ξ
1
)
(7.86)
where sat(x) = sgn(x) min¦[x[, 1¦. System (7.86) is a special case of the zsubsystem
of (7.80) under the assumption that f(z, ξ
1
, 0) = sat(ξ
1
). The following state feedback
α(z) = −σ
1
(z
1
+ σ
2
(z
2
)) (7.87)
globally asymptotically stabilize (z
1
, z
2
) = 0 for (7.86) where σ
1
is an strictly increas
ing bounded sigmoidal function and σ
2
is a bounded sigmoidal function with ﬂat tails
247
−1.5 −1 −0.5 0 0.5 1 1.5
−1
−0.5
0
0.5
1
x
σ
(
x
)
0.75 0.8 0.85 0.9 0.95 1
0.75
0.8
0.85
0.9
0.95
1
x
σ
(
x
)
−1.5 −1 −0.5 0 0.5 1 1.5
0
0.5
1
1.5
2
x
σ
’
(
x
)
−1.5 −1 −0.5 0 0.5 1 1.5
−20
−10
0
10
20
x
σ
"
(
x
)
Figure 71: A C
2
smooth sigmoidal function with ﬂat tails with its derivatives.
outside I = [−1, 1]. By deﬁnition, α(z) in (7.87) has compact domain of attention
w.r.t. z
2
with /
0
= I.
Lemma 7.4.2. Consider the nonlinear system in (7.86). Let α(z) be the state feed
back in (7.87) with σ
1
and σ
2
which are bounded sigmoidal functions with ﬂat tails
outside I = [−1, 1]. Then, the following statements hold:
i) z = 0 is globally asymptotically stable for (7.86) with state feedback α(z).
ii) there exists a compact neighborhood / ⊂ R
2
of z = 0 as the following
/ = ¦(z
1
, z
2
) ∈ R
2
[ [z
1
+ σ
2
(z
2
)[ ≤ 1, [z
2
[ ≤ 1¦ ⊂ [−2, 2] [−1, 1] (7.88)
such that D
z
2
α(z) = 0 for z ∈ /.
Proof. Let us deﬁne the following global change of coordinates
y
1
= z
1
+σ
2
(z
2
)
y
2
= z
2
(7.89)
In new coordinates, the dynamics of the closedloop system (7.86) takes the form
˙ y
1
= y
2
−σ
2
(y
2
)σ
1
(y
1
)
˙ y
2
= −σ
1
(y
1
)
(7.90)
Deﬁne
φ(y
1
) =
y
1
0
σ
1
(s)ds
248
and consider the following positive deﬁnite and smooth Lyapunov function candidate
V (y
1
, y
2
) = φ(y
1
) +
1
2
y
2
2
which satisﬁes
˙
V = −σ
2
(y
2
)σ
1
(y
1
)
2
≤ 0
We need to ﬁnd the largest invariant set in Ω = ¦y ∈ R
2
[
˙
V = 0¦, or
Ω = ¦y ∈ R
2
[ [y
2
[ ≥ 1 or y
1
= 0¦
A simple analysis shows that there is no invariant set in the region [y
2
[ > 1 and the
only invariant set for y
1
= 0 is ¦(0, 0)¦. Thus, from LaSalle’s invariance principle,
y = 0 (or z = 0) is GAS. To prove part ii), notice that D
z
2
α(z) = σ
2
(y
2
)σ
1
(y
1
) = 0
for all y ∈ S = ¦y [ [y
1
[ > 1, [y
2
[ > 1¦. Clearly, / is the complement of S in R
2
. Also,
assuming [z
1
[ > 2, we get
[y
1
[ ≥ [z
1
[ −[σ
2
(z
2
)[ > 1
and therefore D
z
2
α(z) = 0. This means outside
˜
/ = [−2, 2] [−1, 1] ⊃ /, D
z
2
α(z)
vanishes.
The main beneﬁt of lemma 7.4.2 is that outside a compact set /, one can explicitly
determine the ﬁxed point control law ξ
∗
2
= β(z) of equation (7.78). Since D
z
2
α(z) = 0
over R
2
` /, we obtain
β(z) =
∂α(z)
∂z
1
z
2
and the stability analysis for the closedloop zsubsystem
˙ z = F(z, α(z), β(z))
reduces to a stability analysis over the sets (R
2
` o) and (o ` /) where the set o is
called a sigmoidal ribbon deﬁned as the following
o := ¦z ∈ R
2
[ [z
1
+ σ
2
(z
2
)[ ≤ 1¦
By deﬁnition, / ⊂ o. We have
β(z) =
D
z
1
α(z) z
2
z ∈ (o ` /)
0 z ∈ (R
2
` o)
(7.91)
Figure 72 shows diﬀerent regions including the sigmoidal ribbon o (light shaded)
and the compact set / (dark shaded). The dashed regions in Figure 72 show where
the control law α(z) is constant (i.e. Dα(z) = 0) It remains to ﬁnd β(z) over the
compact set /. This is done in the following for the general case where z
1
, z
2
∈ R
n
.
Assumption 7.4.9. Suppose that for any compact set 0 ∈ / := /
z
2
/
ξ
1
/
ξ
2
⊂
R
n+2
, there exist constants a
i
> 0, b
i
≥ 0, i = 1, 2, 3 such that f
0
, g
0
, D
ξ
2
f have
incremental linear growth in z
1
for all (z
2
, ξ
1
, ξ
2
) ∈ /, i.e.
249
z
z
2
1
1
S
S
1
1
1
K
α=+1
α=−1
Figure 72: Sigmoidal ribbon o and compact set /.
i) f
0
(z, ξ
1
) ≤ a
1
+ b
1
z
1

ii) g
0
(z, ξ
1
, ξ
2
) ≤ a
2
+ b
2
z
1

iii) D
ξ
2
f(z, ξ
1
, ξ
2
) ≤ a
3
+ b
3
z
1

The following scaling of the function α(z) is inﬂuenced by the work of Sepulchre [79].
Lemma 7.4.3. Let the smooth function α(z) : R
2n
→R with α(0) = 0 be a bounded
function that has a compact domain of attention /
0
around z
2
= 0 with constants
L
i
> 0, i = 0, 1, 2 satisfying all the conditions in deﬁnition 7.4.1. Deﬁne
α
κ
(z) = κ(z
1
)α(z) (7.92)
where
κ(z
1
) :=
1 , if b
i
= 0, i = 1, 2, 3
1
1 +z
1

2
, otherwise
Then, α
κ
(z) is also a bounded function with compact domain of attention /
0
around
z
2
= 0 and associated constants L
κ
0
= L
0
, L
κ
1
= L
1
, and L
κ
2
= L
0
+L
2
.
Proof. Noting that 0 < κ(z
1
) ≤ 1, part i) of deﬁnition 7.4.1 immediately follows. To
prove properties ii) and iii) of α
κ
(z) (in the same deﬁnition), notice that
D
z
2
α
κ
(z) = κ(z
1
)D
z
2
α(z)
Thus, D
z
2
α
κ
(z) vanishes for all z
2
∈ /
0
. In addition, D
z
2
α
κ
(z) ≤ D
z
2
α(z) ≤ L
1
,
∀z ∈ R
2
. To prove the part iv), let us calculate D
z
1
α
κ
(z) as
D
z
1
α
κ
(z) = (D
z
1
κ(z
1
))α(z) + κ(z
1
)D
z
1
α(z)
250
but D
z
1
κ(z
1
) = z
1
κ(z
1
)
3
and D
z
1
κ(z
1
) ≤ z
1
κ(z
1
)κ(z
1
)
2
≤ 1. Hence, we get
D
z
1
α
κ
(z) ≤ L
0
+ L
2
Theorem 7.4.3. Suppose that Assumption (7.4.9) holds. Let the smooth function
α(z) : R
2n
→ R with α(0) = 0 be a bounded function that has a compact domain of
attention /
0
around z
2
= 0 with constants L
i
> 0, i = 0, 1, 2. In addition, let α
κ
(z)
denote the scaled α(z) deﬁned in equation (7.92). Then, there exist L
∗
i
(/
0
) > 0, i =
0, 1, 2 such that for all α(z) with associated constants 0 < L
i
< L
∗
i
, i = 0, 1, 2, the
following ﬁxedpoint equation
ξ
2
= D
z
α
κ
(z) F(z, α
κ
(z), ξ
2
) (7.93)
globally has a unique smooth solution
ξ
∗
2
= β(z)
with β(0) = 0.
Proof. See section C.1 in Appendix C.
Remark 7.4.3. Based on the contraction mapping theorem, the ﬁxed point ξ
∗
2
= β(z)
in theorem 7.4.3 can be obtained using the following iteration in k
ξ
(k+1)
2
= Ψ(z, ξ
(k)
2
), ξ
(0)
2
= 0 (7.94)
and satisﬁes β(z) = lim
k→∞
ξ
(k)
2
= ξ
∗
2
.
Proposition 7.4.7. Assume F has the following structure
F(z, ξ
1
, ξ
2
) =
¸
z
2
f
0
(z, ξ
1
) + g
0
(z, ξ
1
)ξ
2
2
Then, the ﬁxed point equation in theorem 7.4.3 reduces to the following quadratic
equation in ξ
2
ξ
2
= D
z
1
α
κ
(z) z
2
+ D
z
2
α
κ
(z) [f
0
(z, α
κ
(z)) + g
0
(z, α
κ
(z)]ξ
2
2
(7.95)
so that its solutions can be given in explicit form.
Proof. The ﬁxed point equation in (7.95) can be rewritten as a standard quadratic
equation in ξ
2
in the form
a(z)ξ
2
2
+ bξ
2
+ c(z) = 0 (7.96)
where
a(z) = D
z
2
α
κ
(z)g
0
(z, α
κ
(z))
b = −1
c(z) = D
z
1
α
κ
(z) z
2
+D
z
2
α
κ
(z) f
0
(z, α
κ
(z))
251
The condition for global existence of a ﬁxed point solution ξ
∗
2
= β(z) can be readily
expressed as ∆(z) = 1 −4a(z)c(z) > 0. Then the solution of the quadratic equation
(7.96) in ξ
2
with the smallest absolute value [72] is given by
β(z) =
2c(z)
1 +
1 −4a(z)c(z)
(7.97)
The solution in (7.97) is the same as the following standard solution of (7.96)
β(z) =
1 −
1 −4a(z)c(z)
2a(z)
a(z) = 0
c(z) a(z) = 0
Remark 7.4.4. Following the line of proposition 7.4.7, let / be a compact set such
that D
z
2
α(z) = 0 for all z ∈ / (see part (ii) of lemma 7.4.2). Set κ(z) = 1, or let
α
κ
(z) = α(z). Then, the existence condition of the ﬁxed point ξ
∗
2
a(z)c(z) <
1
4
has to hold over the compact set / ⊂ R
n
0
instead of the whole R
n
0
. This tremendously
reduces the complexity of checking the global existence of a ﬁxed point control law.
We demonstrate this in the following. Since / is a compact set, there exists constants
L
f
0
, L
g
0
, L
z
2
> 0 such that
f
0
(z, α(z)) ≤ L
f
0
, g
0
(z, α(z)) ≤ L
g
0
, z
2
 ≤ L
z
2
for all z ∈ /. Hence, the following condition
[a(z)c(z)[ ≤ L
1
L
g
0
(L
2
L
z
2
+ L
1
L
f
0
) <
1
4
is satisﬁed by choosing the constants L
1
, L
2
associated with α(z) according to
0 < L
1
, L
2
< L
∗
:=
1
2
L
g
0
(L
f
0
+ L
z
2
)
(7.98)
where L
∗
can be readily calculated.
Example 7.4.4. (beamandball system) Consider the dynamics of the beamand
ball system given in equation (5.33). The thirdorder subsystem of (5.33) is in the
form
˙ z
1
= z
2
˙ z
2
= −g sin(ξ
1
) + (z
1
+ dξ
1
)ξ
2
2
˙
ξ
1
= ξ
2
(7.99)
252
where (ξ
1
, ξ
2
) = (θ, ω). Deﬁne
f
0
(z, ξ
1
) = −g sin(ξ
1
)
g
0
(z, ξ
1
) = z
1
+dξ
1
(7.100)
Clearly, f
0
, g
0
satisfy assumption 7.4.9. Let α(z) = −c
0
σ
2
(z
1
+σ
1
(z
1
)) where σ
i
’s are
sigmoidal functions with ﬂat tails outside [−1, 1]. In addition, [σ
2
(x)[ ≤ 1, ∀x and
0 < c
0
< π/2 is a constant (to be determined). One can take c
0
to be suﬃciently
small such that condition a(z)c(z) < 1/4 in remark 7.4.4 is satisﬁed over the compact
set / (shown in Figure 72). Based on proposition 7.4.7, the state feedback β(z) can
be explicitly and globally obtained from (7.97).
7.4.4 Slightly Nontriangular Nonlinear Systems
In this section, we consider stabilization of a perturbed version of a nonlinear system
in strict feedback form such that the overall system is nontriangular. We call this
class of systems slightly nontriangular systems.
Theorem 7.4.4. Consider the following nonlinear system
˙ z = f
0
(z, ξ
1
) + g
0
(z, ξ
1
, ξ
2
, ),
˙
ξ
1
= ξ
2
,
˙
ξ
2
= u,
(7.101)
where f
0
, g
0
are smooth functions with f
0
(0) = 0, g
0
(0) = 0, g
0
(z, ξ
1
, 0, ) = 0,
g
0
(z, ξ
1
, ξ
2
, 0) = 0, and ∂g
0
/∂ξ
2
vanishes at ξ
2
= 0 (i.e.∃p
0
∈ C
1
: g
0
= p
0
(z, ξ
1
, ξ
2
, )ξ
2
2
).
Suppose there exists a state feedback law ξ
1
= α
1
(z) with α
1
(0) = 0 such that for
˙ z = f
0
(z, α
1
(z))
z = 0 is globally asymptotically and locally exponentially stable. Let V (z) be the
smooth positive deﬁnite proper Lyapunov function associated with this system satis
fying the property ∇V (z) f
0
(z, α
1
(z)) < 0, ∀z = 0. Deﬁne
ψ
2
(z, ξ
1
, ξ
2
, ) = −σ(ξ
1
−α
1
(z)) +
∂α
1
(z)
∂z
f(z, ξ
1
, ξ
2
, )
ψ
3
(z, ξ
1
, ξ
2
, u, ) = −σ(ξ
2
−ψ
2
) +
∂ψ
2
∂z
f(z, ξ
1
, ξ
2
, ) +
∂ψ
2
∂ξ
1
ξ
2
+
∂ψ
2
∂ξ
2
u
(7.102)
where f = f
0
(z, ξ
1
) + g
0
(z, ξ
1
, ξ
2
, ) and σ() is a sigmoidal function. In addition,
assume for any l
0
> 0, there exist k > 0, r
0
≥ 0 and λ ∈ (0, 1] such that
∂V (z)
∂z
h
0
(z, α
1
(z), δ)
≤ kV (z)
λ
, ∀[z[ > r
0
, ∀[δ[ ≤ l
0
(7.103)
where h
0
is given by
h
0
(z, ξ
1
, δ) =
1
0
∂f
0
∂ξ
1
(z, ξ
1
+ sδ)ds
253
and satisﬁes
f
0
(z, ξ
1
+ δ) = f
0
(z, ξ
1
) + h
0
(z, ξ
1
, δ)δ
Then, there exists an
∗
> 0 such that for all 0 < <
∗
the following hold:
i) The change of coordinates µ = T(z, ξ) as
µ
1
= ξ
1
−α
1
(z)
µ
2
= ξ
2
−ψ
2
(z, ξ
1
, ξ
2
, )
is invertible over a compact domain /
z
/
ξ
with a smooth inverse ξ = T
−1
(z, µ)
given by
ξ
1
= β
1
(z, µ
1
) := µ
1
+ α
1
(z)
ξ
2
= β
2
(z, µ
1
, µ
2
, )
with β
1
(0, 0) = 0 and β
2
(0, 0, 0, ) = 0.
ii) The ﬁxed point equations
ξ
2
= ψ
2
(z, ξ
1
, ξ
2
, )
u = ψ
3
(z, ξ
1
, ξ
2
, u, )
have unique smooth solutions
ξ
∗
2
= α
2
(z, ξ
1
, )
u
∗
= α
3
(z, ξ
1
, ξ
2
, )
with α
2
(0, 0, ) = 0 and α
3
(0, 0, 0, ) = 0 over a compact domain /
z
/
ξ
.
iii) The state feedback ξ
2
= α
2
(z, ξ
1
, ) semiglobally asymptotically and locally expo
nentially stabilizes (z, ξ
1
) = 0 for the (z, ξ
1
)subsystem of (7.101).
iv) The state feedback u = α
3
(z, ξ
1
, ξ
2
, ) semiglobally asymptotically and locally
exponentially stabilizes (z, ξ
1
, ξ
2
) = 0 for the composite system in (7.101).
Proof. See section C.4 in Appendix C.
The proof of the existence of unique smooth ﬁxed points in theorem 7.4.4 relies
on the following lemma.
Lemma 7.4.4. Consider the following ﬁxed point equation in y
y = φ
0
(x) + φ
1
(x, y, ) =: φ(x, y, ) (7.104)
where φ : R
n
R
m
R →R
m
is continuously diﬀerentiable with φ(0, 0, ) = 0. Then,
for any compact set /
x
and the interval I
= [0,
1
], there exists a compact set /
y
and an 0 <
∗
∈ I
such that for all 0 < <
∗
the ﬁxed point equation (7.104) has a
unique smooth solution
y
∗
= α(x, )
that belongs to /
y
and satisﬁes α(0, ) = 0 uniformly in .
254
Proof. See section C.3 in Appendix C.
The next theorem is a Lyapunovbased ﬁxed point backstepping theorem for sta
bilization of slightly nontriangular nonlinear systems. First, we need to make a deﬁ
nition.
Deﬁnition 7.4.3. (residual of a function w.r.t. its ith argument) Consider a contin
uously diﬀerentiable function f(x
1
, . . . , x
m
) : R
n
1
. . . R
nm
→R
p
. Deﬁne
h(x
1
, . . . , x
m
, a, δ) =
1
0
D
x
i
f(x
1
, . . . , x
m
)[
x
i
=a+sδ
ds
and denote
h(x
1
, . . . , x
m
, a, δ) = res
(i,a,δ)
(f(x
1
, . . . , x
m
))
Read res
(i,a,δ)
(f(x
1
, . . . , x
m
)), the residual of f w.r.t. its ith argument with perturba
tion δ around x
i
= a. Then, the following decomposition of f is always possible
f(x
1
, . . . , x
i
+ δ, . . . , x
m
) = f(x
1
, . . . , x
m
) +h(x
1
, . . . , x
m
, δ)δ
Based on this notation, we have f(x
1
, x
2
) = f(x
1
, 0) + h(x
1
, x
2
)x
2
with h(x
1
, x
2
) =
res
(2,0,x
2
)
(f(x
1
, x
2
)).
Theorem 7.4.5. Consider the perturbed nonlinear system in (7.101) where f
0
, g
0
are
smooth functions satisfying f
0
(0) = 0, g
0
(0) = 0, g
0
(z, ξ
1
, 0, ) = 0, and g
0
(z, ξ
1
, ξ
2
, 0) =
0 (i.e.∃p
0
∈ C
1
: g
0
= p
0
(z, ξ
1
, ξ
2
, )ξ
2
). Suppose there exists a state feedback
ξ
1
= α
1
(z) with α
1
(0) = 0 such that for
˙ z = f
0
(z, α
1
(z))
z = 0 is globally asymptotically and locally exponentially stable. Let V (z) be the
Lyapunov function for this system with
∂V (z)
∂z
f
0
(z, α
1
(z)) < 0 , ∀z = 0
Decompose f
0
as the following
f
0
(z, ξ
1
+ δ) = f
0
(z, ξ
1
) + h
0
(z, ξ
1
, δ)δ
and deﬁne
ψ
2
(z, ξ
1
, ξ
2
, ) = −σ
1
(ξ
1
−α
1
(z)) +
∂α
1
(z)
∂z
(f
0
(z, ξ
1
) + g
0
(z, ξ
1
, ξ
2
, ))
−
∂V (z)
∂z
h
0
(z, α
1
(z), ξ
1
−α
1
(z))
(7.105)
Then, there exists an
∗
> 0 such that for all 0 < <
∗
the following hold
i) The ﬁxed point equation
ξ
2
= ψ
2
(z, ξ
1
, ξ
2
, ) (7.106)
255
for all (z, ξ
1
) ∈ /
z
/
ξ
1
has a unique smooth solution ξ
∗
2
= α
2
(z, ξ
1
, ) with
α
2
(0, 0, ) = 0.
ii) For a suﬃciently large c
1
= σ
1
(0), the state feedback ξ
2
= α
2
(z, ξ
1
, ) semiglob
ally asymptotically stabilizes the origin for the (z, ξ
1
)subsystem of (7.101).
iii) Fix an > 0 valid for (ii). Then, there exists a static state feedback u =
α
3
(z, ξ
1
, ξ
2
, ) in the following form
u = −σ
2
(ξ
2
−α
2
(¯ z, )) +
∂α
2
(¯ z, )
∂¯ z
f
2
(¯ z, ξ
2
, ) −
∂W(¯ z)
∂¯ z
h
2
(¯ z, α
2
(z, ξ
1
, ), , ξ
2
−α
2
)
(7.107)
that semiglobally asymptotically stabilizes the origin for the composite system
in (7.101) for a suﬃciently large c
2
= σ
2
(0) > 0 (σ
2
is a sigmoidal function)
where
¯ z = col(z, ξ
1
)
W(¯ z) = V (z) +
1
2
(ξ
1
−α
1
(z))
2
f
2
(z, ξ
1
, ξ
2
, ) = col(f, ξ
2
)
h
2
(¯ z, ξ
2
, , δ) = res
(3,ξ
2
,δ)
(f
2
(z, ξ
1
, ξ
2
, ))
Proof. See section C.5 in Appendix C.
Now, we present some examples of nontriangular nonlinear systems that can be
stabilized using theorems 7.4.4 and 7.4.5.
Example 7.4.5. (perturbed chain of integrators) Consider the following perturbed
chain of integrators
˙ x = Ax +Bu + p(x, u, v)
˙ u = v
(7.108)
where p(x, u, v) is smooth and p(x, u, 0) = 0. In addition, A = ¦a
ij
¦ with a
ij
= 1 for
j = i + 1 and zero otherwise, and B = (0, . . . , 0, 1)
T
. Deﬁne the following change of
scale and timescale
z
i
=
(n+1−i)
x
i
, i = 1, . . . n
ξ
1
= u
ξ
2
= v/
and denote
ζ
:= (
−n
z
1
, . . . ,
−1
z
n
, ξ
1
, ξ
2
)
Deﬁne a new perturbation ¯ p = (¯ p
1
, . . . ¯ p
n
)
T
by its elements as the following
¯ p
i
(z, ξ
1
, ξ
2
, ) :=
n−i
p
i
(ζ
).
256
In proposition 7.4.8, we show that if the perturbation p satisﬁes the following growth
condition, then the nonlinear system in (7.108) can be semiglobally asymptotically
stabilized.
Assumption 7.4.10. The perturbation p satisﬁes the following condition
n−i
p
i
(ζ
)[
=0
= 0 , ∀(z, ξ
1
, ξ
2
) ∈ R
n+2
, i = 1, . . . , n
i.e. ¯ p(z, ξ
1
, ξ
2
, 0) = 0.
Proposition 7.4.8. Suppose Assumption 7.4.10 holds. Then, there exists a static
state feedback law in the form of a ﬁxed point controller that semiglobally asymptoti
cally stabilizes (x, u) = 0 for (7.108).
Proof. Apply the aforementioned change of scale and timescale τ = t. From now
on, (with a little abuse of notation) let (˙ ) denote d/dτ in new scale and timescale.
The dynamics of the system in new coordinates can be expressed as
˙ z = Az + Bξ
1
+ ¯ p(z, ξ
1
, ξ
2
, )
˙
ξ
1
= ξ
2
(7.109)
Clearly, ¯ p(z, ξ
1
, 0, ) = 0 and ¯ p(z, ξ
1
, ξ
2
, 0) = 0. Thus, system (7.109) is in the form
(7.101). Noting that a linear state feedback ξ
1
= Kz globally asymptotically stabilizes
z = 0 with a quadratic Lyapunov function V (z) = z
T
Pz where P is the solution of
the Lyapunov equation
(A + BK)
T
P +P(A +BK) = −I
a semiglobally asymptotically stabilizing feedback for (7.108) can be obtained from
part (ii) of theorem 7.4.5.
Remark 7.4.5. The problem of global asymptotic stabilization of a perturbed chain of
integrators with nontriangular nonlinear perturbation p in (7.108) that satisﬁes the
condition
∂p
∂v
(x, u, 0) = 0
is addressed in [79] using a state feedback in the form of sum of saturations with
relatively small magnitudes. The work in [79] is an extension of the method of nested
saturations due to Teel [102]. The main diﬀerence between the result of proposition
7.4.8 and the work in [79] is that the condition that p(x, u, v) = ˜ p(x, u, v)v
2
is not
needed in here. The next example clariﬁes this diﬀerence.
Example 7.4.6. Consider the following nonlinear system nonaﬃne in control
˙ x
1
= x
2
+ x
1
v
2
+ uv + x
2
v
˙ x
2
= u + x
2
v
4
3
+x
1
v
3
+ uv + x
1
x
2
uv
4
˙ u = v
(7.110)
257
the system is in the form of a chain of two integrators with a perturbation p = (p
1
, p
2
)
T
where p
1
(x, u, v) = x
1
v
2
+ uv + x
2
v and p
2
(x, u, v) = x
2
v
4
3
+ x
1
v
3
+ uv + x
1
x
2
uv
4
.
This system is not in feedforward form and the perturbation term does not satisfy
∂p
∂v
(x, u, 0) = 0 for all (x, u). Thus, it can neither be controlled using Teel’s nested
saturations [102] , nor the method in [79]. Applying the change of scale and timescale
z
1
=
2
x
1
, z
2
= x
2
, ξ
1
= u, ξ
2
= v/, τ = t
we get
˙ z
1
= z
2
+ z
1
ξ
2
2
+
2
ξ
1
ξ
2
+ z
2
ξ
2
˙ x
2
= ξ
1
+
1
3
z
2
ξ
4
3
2
+ z
1
ξ
3
2
+ξ
1
ξ
2
+z
1
z
2
ξ
1
ξ
4
2
˙
ξ
1
= ξ
2
Apparently, this system satisﬁes all the conditions of proposition 7.4.8 and can be
semiglobally stabilized using a static state feedback law in the form of a ﬁxed point
controller.
7.4.5 Notion of Partial Semiglobal Asymptotic Stabilization
In this section, we introduce a new notion of stabilizability that is in a sense close to
semiglobal asymptotic stabilizability but rather weaker than that. A reason behind
deﬁning this notion is that it is not always possible to transform any perturbed chain
of integrators with an nontriangular structure into the form (7.101) using a change of
scale and timescale as described in proposition 7.4.8. The following example clariﬁes
this issue.
Example 7.4.7. Consider the following nonlinear system in nontriangular form
˙ x
1
= x
2
˙ x
2
= η
1
+ η
1
x
3
2
η
2
2
˙ η
1
= η
2
(7.111)
applying a change of scale and timescale as
z
1
=
2
x
1
, z
2
= x
2
, ξ
1
= η
1
, ξ
2
= η
2
/, τ = t
one obtains
˙ z
1
= z
2
˙ z
2
= ξ
1
+
1
ξ
1
z
3
2
ξ
2
2
˙
ξ
1
= ξ
2
Apparently, as → 0 the perturbation term
1
ξ
1
z
3
2
ξ
2
2
goes unbounded and does not
vanish. A modiﬁcation of the change of scale and timescale as the following
z
1
= x
1
, z
2
= x
2
, ξ
1
= η
1
/, ξ
2
= η
2
/
2
, τ = t
258
resolves this problem and we get
˙ z
1
= z
2
˙ z
2
= ξ
1
+
2
ξ
1
z
3
2
ξ
2
2
˙
ξ
1
= ξ
2
(7.112)
which is clearly in the form (7.101). Now, the problem is that semiglobal stabilization
of (7.112) does not guarantee semiglobal stabilization of the original system in (7.111).
The obtained region of attraction for (7.111) is in the form /
x
/
η
1
where /
x
is
larger than the set of initial conditions x(0) and /
η
1
is smaller than the set of initial
conditions η
1
(0) due to the fact that η
1
= ξ
1
for a relatively small > 0.
This motivates us to deﬁne a new stabilizability notion that is weaker than
semiglobal stabilizability but still is rather similar to it.
Deﬁnition 7.4.4. (partial semiglobal stabilizability) Consider the following system
˙ x
1
= f
1
(x
1
, x
2
, u)
˙ x
2
= f
2
(x
1
, x
2
, u)
(7.113)
We say the origin (x
1
, x
2
) = 0 for (7.113) is partially semiglobally asymptotically
stabilizable with respect to x
1
if for any compact set of (partial) initial conditions
/
1
÷ x
1
(0), there exist a neighborhood /
2
of x
2
= 0 and a control law u such that
the closedloop system in (7.113) has a region of attraction /
1
/
2
with /
1
⊃ /
1
.
Remark 7.4.6. Apparently, partial semiglobal asymptotic stabilization is stronger
than local asymptotic stabilization.
Now, based on theorem 7.4.5, (x, η
1
) = 0 is partially semiglobally stabilizable
w.r.t. x for (7.111) using a static state feedback.
7.5 Applications
In this section, we present applications of ﬁxed point control laws to stabilization of
a number of challenging benchmark systems in robotics.
7.5.1 The CartPole System with Small Length/Strong Grav
ity Eﬀects
In this section, we analyze the eﬀects of relatively small lengths or large gravity
constants in stabilization of the CartPole system depicted in Figure 73. In both
cases, we provide semiglobally stabilizing ﬁxed point state feedback laws for the Cart
Pole system. The cascade normal form for the CartPole system is given in equation
259
m
F
m , l , I
1
2 2 2
q
q
1
2
Figure 73: The CartPole System
(5.21) as the following
˙ z
1
= z
2
˙ z
2
= ξ
1
(k
1
+k
2
ξ
2
2
(1 + ξ
2
1
)
3
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(7.114)
where k
1
, k
2
are the following positive constants
k
1
= g ≥ 1, k
2
=
m
2
l
2
2
+ I
2
m
2
l
2
Assume the link l
2
has uniform mass density. Thus, I
2
= m
2
l
2
2
/3 and k
2
=
4
3
l
2
. Denote
= k
2
/k
1
> 0 and apply the following change of scale
y
1
=
1
k
1
z
1
, y
2
=
1
k
1
z
2
Since k
1
≥ 1, under this change of scale, semiglobal stabilization in new coordinates
and original coordinates are equivalent (i.e. one implies the other). The dynamics of
the CartPole system in new scale can be rewritten as
˙ y
1
= y
2
˙ y
2
= ξ
1
(1 +
ξ
2
2
(1 + ξ
2
1
)
3
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(7.115)
Under either of the following conditions:
i) g 1, i.e. relatively high gravity.
ii) l
2
<1, i.e. relatively small lengths.
260
we obtain an that is relatively small, i.e. = k
2
/k
1
< 1. Clearly, this is in
the form (7.101) and the origin for this system can be semiglobally asymptotically
stabilized according to theorem 7.4.5. Here, we show that theorem 7.4.4 can be also
used to obtain a stabilizing state feedback for this CartPole system. Setting ξ
2
= 0
in the ysubsystem, the remaining system is a doubleintegrator that can be globally
asymptotically stabilized using a bounded control ξ
1
= −σ(y
1
+y
2
) (σ is a sigmoidal
function) with a positive deﬁnite Lyapunov function
V (y) = φ(y
1
) +φ(y
1
+ y
2
) + y
2
2
where φ(x) =
x
0
σ(s)ds and
˙
V < 0 for all y = 0. To prove the boundedness of the
solutions of the ysubsystem, one can use a simpler Lyapunov function
V
1
(y) = φ(y
1
) +
1
2
y
2
2
Because f
0
(y, ξ
1
) = (y
2
, ξ
1
)
T
, f
0
(y, ξ
1
+ δ) = f
0
(y, ξ
1
) + h
0
δ with h
0
= (0, 1)
T
. Thus,
we have
∂V
1
(y)
∂y
h
0
=
∂V
1
(y)
∂y
2
= y
2
or (independent of δ)
[
∂V
1
(y)
∂y
h
0
[ ≤ [y
2
[ <
√
2V
1
(y)
1
2
, ∀y = 0
and condition (7.103) of theorem 7.4.4 is satisﬁed with k =
√
2, any r
0
> 0, and
λ =
1
2
. Thus, a semiglobally stabilizing state feedback that makes no explicit use of
any Lyapunov functions can be obtained using the cascade backstepping procedure
in theorem 7.4.4). Figure 74, shows the simulation results for the cartpole system
starting at the initial state (1, 1, π/3, 0) for the choice of ξ
1
= −tanh(y
1
+ y
2
).
0 2 4 6 8 10 12 14 16 18 20
−2
0
2
4
6
8
10
12
time (sec)
p
o
s
i
t
i
o
n
/
a
n
g
l
e
Figure 74: The state trajectories of the CartPole system.
261
7.5.2 The CartPole System with Large Length/Weak Grav
ity Eﬀects
Consider the normal form of the CartPole system in (7.114) and apply the following
change of scale and timescale given in [67]
y
1
=
2
z
1
, y
2
= z
2
, η
1
= ξ
1
, η
2
= ξ
2
/, w = u/
2
, τ = t
The dynamics of the system in new coordinates and timescale is in the form
˙ y
1
= y
2
,
˙ y
2
= η
1
(k
1
+ k
2
2
η
2
2
(1 + η
2
1
)
3
2
),
˙ η
1
= η
2
,
˙ η
2
= w
(7.116)
Semiglobal stabilization for (7.116) is very similar to the one presented for system
(7.115) and will be repeated here.
Now, without loss of generality assume I
2
= 0. Thus, k
2
= l
2
. This case has an
interesting physical interpretation. Normalizing the units of y
1
, y
2
by k
1
= g as
x
1
=
1
g
y
1
, x
2
=
1
g
y
2
and setting
=
k
1
k
2
=
g
l
2
= ω
0
(where ω
0
is the natural frequency of a pendulum of length l
2
) we obtain the normal
form of the CartPole system with normalized units as
˙ x
1
= x
2
,
˙ x
2
= η
1
(1 +
η
2
2
(1 + η
2
1
)
3
2
),
˙ η
1
= η
2
,
˙ η
2
= w
(7.117)
In either of the following cases:
i) g <1, i.e. relatively weak gravity,
ii) l
2
1, i.e. relatively large length;
the obtained (or the natural frequency ω
0
) is relatively small, i.e. ω
0
= <1. This
means the solutions in scaled coordinates and time scale are relatively slow. Observe
that the conditions of large length/weak gravity are in complete opposition with the
case considered in section 7.5.1 with small length/large gravity eﬀects.
262
7.5.3 The Rotating Pendulum
Consider the Rotating Pendulum as shown in Figure 75. The nontriangular normal
form for the Rotating Pendulum is given in (5.24) as the following
˙ z
1
= z
2
˙ z
2
= ξ
1
(
g
L
1
+
l
2
L
1
1
1 + ξ
2
1
(z
2
−
m
22
m
2
l
2
L
1
ξ
2
1 + ξ
2
1
)
2
+
m
22
m
2
l
2
L
1
ξ
2
2
(1 + ξ
2
1
)
3
2
)
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(7.118)
where m
22
= m
2
l
2
2
+ I
2
. Without loss of generality assume I
2
= 0, or m
22
= m
2
l
2
2
x
y
z
1
2
q
m , L , I
m , L , I
2
1 1
2
q
1
2
Figure 75: The Rotating Pendulum
(otherwise, I
2
= k
0
m
2
l
2
2
with some k
0
> 0 and the line of analysis is very similar).
Thus, equation (7.118) can be rewritten as
˙ z
1
= z
2
˙ z
2
= ξ
1
[k
1
+ k
2
1
(1 + ξ
2
1
)
1
2
(z
2
−k
2
ξ
2
(1 +ξ
2
1
)
1
2
)
2
+ k
2
ξ
2
2
(1 + ξ
2
1
)
3
2
]
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(7.119)
where
k
1
=
g
L
1
, k
2
=
l
2
L
1
Remark 7.5.1. Let us deﬁne
y
1
=
1
k
1
z
1
, y
2
=
1
k
1
z
2
, =
k
2
k
1
where k
2
is constant and k
1
1. This change of scale transforms (7.119) into the
263
following nonlinear system
˙ y
1
= y
2
˙ y
2
= ξ
1
[1 +
k
2
1
(1 +ξ
2
1
)
1
2
z
2
2
−2
2
z
2
ξ
2
(1 + ξ
2
1
)
+ (1 + k
2
2
)
ξ
2
2
(1 + ξ
2
1
)
3
2
]
˙
ξ
1
= ξ
2
˙
ξ
2
= v
Apparently, the last equation is not a welldeﬁned slightly nontriangular system. Since
as →0, the coeﬃcient of the term containing z
2
2
goes unbounded, i.e. k
2
/ →∞.
Now, consider the following change of scale and time scale
y
1
= z
1
, y
2
= z
2
, η
1
= ξ
1
/, η
2
= ξ
2
/
2
, w = v/
3
, τ = t (7.120)
From (7.119), we obtain
˙ y
1
= y
2
˙ y
2
= η
1
[k
1
+
k
2
(1 +
2
η
2
1
)
1
2
(y
2
−
2
k
2
η
2
(1 +
2
η
2
1
)
1
2
)
2
+
4
k
2
η
2
2
(1 +
2
η
2
1
)
3
2
]
˙ η
1
= η
2
˙ η
2
= w
(7.121)
but
k
2
(1 +
2
η
2
1
)
1
2
= k
2
(1 +
1
(1 +
2
η
2
1
)
1
2
−1) = k
2
−
k
2
2
(1 +
2
η
2
1
)
1
2
[1 + (1 +
2
η
2
1
)
1
2
]
Hence
˙ y
1
= y
2
˙ y
2
= η
1
(k
1
+k
2
y
2
2
) + g(y
2
, η
1
, η
2
, ) + ∆(y
2
, η
1
, )
˙ η
1
= η
2
˙ η
2
= w
(7.122)
where
g(y
2
, η
1
, η
2
, ) = −
2
2
k
2
2
y
2
η
1
η
2
(1 +
2
η
2
1
)
+
4
k
2
(1 + k
2
2
)η
2
2
(1 +
2
η
2
1
)
3
2
∆(y
2
, η
1
, ) = −
2
η
1
y
2
2
(1 +
2
η
2
1
)
1
2
Notice that g(y
2
, η
1
, η
2
, ) and ∆(y
2
, η
1
, ) vanish at = 0. In addition, the perturba
tion ∆ has an upper bound that vanishes uniformly in η
1
due to
[∆(y
2
, η
1
, )[ ≤ k
2
y
2
2
The nonlinear system in (7.122) without the perturbation ∆ is in the form (7.101) and
satisﬁes the conditions of theorem 7.4.5. Thus, it can be semiglobally asymptotically
264
stabilized using the choice of η
1
= α(y) = −σ(y
1
+ y
2
) (σ is a sigmoidal function).
This state feedback renders y = 0 globally asymptotically stable for
˙ y
1
= y
2
˙ y
2
= η
1
(k
1
+ k
2
y
2
2
)
(7.123)
The following Lyapunov function
V (y) = φ(y
1
) + φ(y
1
+ y
2
) +
1
k
2
log(1 +
k2
k
1
y
2
2
) (7.124)
with φ(x) =
x
0
σ(s)ds is a valid Lyapunov function for the closedloop system in
(7.123) satisfying
˙
V < 0 for all y = 0. Now, semiglobal asymptotic stabilization
of (7.122) with the highorder perturbation ∆ can be shown using an elementary
argument. It is important to notice that semiglobal asymptotic stabilization of the
origin for (7.122) implies partial semiglobal asymptotic stabilization of the origin for
the dynamics of the Rotating Pendulum in the original coordinates. Figure 76 shows
the simulation results for the rotating pendulum with initial condition (π/3, 0, π/4, 0).
0 2 4 6 8 10 12 14 16 18 20
−2
0
2
4
6
8
10
12
q
1
/
q
2
time (sec)
q1
q2
Figure 76: The state trajectories for the Rotating Pendulum.
7.5.4 The Generalized BeamandBall System
The generalized beamandball system is a partiallylinear nontriangular cascade non
linear system in the following form
˙ x
1
= x
2
˙ x
2
= g sin(η
1
) + x
ν
1
η
2
2
, ν ≥ 1
˙ η
1
= η
2
˙ η
2
= u
(7.125)
265
for ν = 1, this system is the same as the famous beamandball system that has been
extensively studied and used as an example by many researchers in the past decade
[34, 102, 80, 79]. Here, we consider two cases: i) 1 ≤ ν < 3, and ii) 3 ≤ ν < ∞. For
the sake of simplicity assume g = 1.
Case (i): Applying the change of scale and time scale
z
1
= x
1
, z
2
= x
2
, ξ
1
= η
1
/, ξ
2
= η
2
/
2
, w = v/
3
, τ = t
In the new scale, we obtain
˙ z
1
= z
2
˙ z
2
= ξ
1
+
2
O([ξ
1
[
3
) +
(3−ν)
z
ν
1
ξ
2
2
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(7.126)
where
ξ
1
+
2
O([ξ
1
[
3
) =
sin(ξ
1
)
, > 0
Clearly, the origin for system (7.126) can be semiglobally asymptotically stabilized
using a state feedback in the form of a ﬁxed point controller. This way, the origin for
system (7.125) can be only partially semiglobally stabilized w.r.t. (x
1
, x
2
).
Case (ii): It is easy to see that using the change of scale and timescale
z
1
= x
1
, z
2
= x
2
/, ξ
1
= η
1
/
2
, ξ
2
= η
2
/
3
, w = v/
4
, τ = t
the dynamics of the system transforms into
˙ z
1
= z
2
˙ z
2
= ξ
1
+
5
O([ξ
1
[
3
) +
5
z
ν
1
ξ
2
2
˙
ξ
1
= ξ
2
˙
ξ
2
= v
(7.127)
and the origin for the system (7.125) can be partially semiglobally stabilized w.r.t.
x
1
. The simulation results for the generalized beamandball system are shown in
Figure 77.
266
0 5 10 15
−0.6
−0.4
−0.2
0
0.2
0.4
a
n
g
l
e
o
f
t
h
e
b
e
a
m
time (sec)
v=1
v=2
v=3
0 5 10 15
−1
0
1
2
3
4
time (sec)
p
o
s
i
t
i
o
n
o
f
t
h
e
b
a
l
l
v=1
v=2
v=3
Figure 77: The state trajectories of the generalized beamandball system with initial
condition (1, 0, −π/6, 0) for g = −9.8 and ν = 1, 2, 3.
267
Chapter 8
Conclusions
In this chapter, we summarize the results of this thesis and make concluding remarks.
This thesis creates a bridge between control of mechanical systems and nonlinear con
trol theory by providing transformations in explicit forms that take an underactuated
mechanical system to a cascade nonlinear system with structural properties. Due to
the special partiallylinear cascade structure of the obtained normal forms for under
actuated systems, control design for the original (possibly) highorder underactuated
system reduces to control of a loworder nonlinear part of the transformed system.
The main focus of this thesis was on identifying broad classes of underactuated sys
tems with kinetic symmetry where such a transformation for them could be found in
closedform. It turned out that the majority of obtained normal forms for underac
tuated systems are in nontriangular forms. Since control of nonlinear systems with
nontriangular structures was an important open problem, a great deal of eﬀort in
this thesis was devoted to the development of control design tools for stabilization of
nontriangular nonlinear systems in chapter 7.
Reduction of underactuated systems with nonholonomic velocity constraints and
symmetry is addressed in chapter 6. As a consequence, the corresponding class of an
underactuated system with nonholonomic velocity constraints is determined by the
class of its reduced system which is a welldeﬁned underactuated (or fullyactuated)
mechanical system itself.
The theoretical results of this thesis are applied to challenging control problems
for a variety of robotics and aerospace systems. Almost for all examples, detailed
control design and simulation results are provided. Among these examples are tra
jectory tracking for a ﬂexiblelink arm, automatic calculation of the diﬀerentially
ﬂat outputs of the VTOL aircraft, (almost) global asymptotic/exponential tracking
of feasible trajectories of an autonomous helicopter, global exponential tracking
and stabilization for a twowheeled mobile robot, global stabilization of the Acrobot,
global asymptotic stabilization of the 2D and 3D CartPole systems to an equilib
rium point, and semiglobal asymptotic stabilization of the Rotating Pendulum to an
equilibrium point.
268
8.1 Summary
The summary of this thesis is as follows.
In chapter 1, an introduction was provided to control of mechanical systems, non
linear control theory, and underactuated mechanical systems with their applications.
In chapter 2, some background on diﬀerent types of mechanical systems and basic
notions in mechanics are given. These mechanical notions include classical symmetry,
kinetic symmetry, shape variables, conservation laws, nonholonomicity, and ﬂatness.
In chapter 3, the dynamics of underactuated mechanical system is deﬁned as the
forced EulerLagrange equations of motion. Several examples of underactuated sys
tems were presented in chapter 3. For each example, only the inertia matrix and the
potential energy of the system are given. This allows the reader to clearly observe
the kinetic symmetry property in each example and identify its corresponding shape
variables. In addition, diﬀerent methods for partial feedback linearization of underac
tuated mechanical systems are introduced. This includes collocated and noncollocated
feedback linearization methods as special cases. Finally, the problem of decoupling of
the dynamics of the actuated and unactuated subsystems of a class of underactuated
systems after partial feedback linearization is addressed. The decoupling is achieved
using a global change of coordinates so that after transformation, the obtained cascade
system was in ByrnesIsidori normal form with a set of doubleintegrator linear part.
This result then was applied to complete classiﬁcation of underactuated mechanical
systems with two degrees of freedom and kinetic symmetry to three classes. By “com
plete”, we mean “all” such underactuated systems. It was proven that these three
classes of underactuated systems could be transformed into three types of cascade
normal forms. Namely, nonlinear systems in strict feedback form, strict feedforward
form, and nontriangular quadratic form. All the transformations are performed using
change of coordinates in closedform.
In chapter 4, some of the main contributions of this thesis are presented. Re
duction and classiﬁcation of highorder underactuated systems is certainly the most
important result of this thesis. The number of obtained classes are eight with certain
classes which contain two subclasses. In addition to classiﬁcation of underactuated
systems a correspondence is established between classes of underactuated systems
and cascade systems with triangular and nontriangular structural properties. This
correspondence allows control design for the original underactuated system. The
main tools in stabilization of underactuated systems with triangular normal forms
are backstepping and forwarding methods. In contrast, the main approach in con
trol of nontriangular nonlinear systems is a ﬁxedpoint based backstepping procedure
that is introduced in chapter 7. To deal with reduction of underactuated systems
with nonintegrable momentums, we introduced a procedure call momentum decom
position. This procedure represents a nonintegrable momentum as the sum of an
integrable momentum and an error momentum term which is nonintegrable. Mo
mentum decomposition was applied to reduction of multilink underactuated robots
and calculation of weakly ﬂat outputs for an accurate model of an autonomous heli
copter in chapter 5. For ClassI,II,III underactuated systems both the transformations
and the obtained normal forms are physically meaningful. This allows construction
269
of physically meaningful Lyapunov functions for stability and robustness analysis of
these normal forms. The fact that “the core reduced system for (almost) all classes of
underactuated systems is itself a Lagrangian mechanical system with shape variables
as the control”, creates a unity among diﬀerent classes of underactuated systems.
Global asymptotic stabilization of this core reduced system is addressed in chapter 4.
In chapter 5, several examples of underactuated systems are provided with detailed
control design and simulation results. A method is introduced for aggressive swingup
control using bounded input for special classes of underactuated systems that involve
inverted pendulums. Furthermore, trajectory tracking for a ﬂexible onelink robot
and an autonomous helicopter is addressed in chapter 5. The connection between
diﬀerential ﬂatness and a decoupling change of coordinates for underactuated systems
with input coupling are demonstrated for the case of the VTOL aircraft and an
accurate model of an autonomous helicopter.
In chapter 6, it was shown that underactuated systems with nonholonomic velocity
constraints and symmetry can be represented as the cascade of the constraint equation
and a lowerorder underactuated (or fullyactuated) mechanical system. Depending
on the speciﬁc class of the obtained reduced subsystem, the associated class of the
original nonholonomic system is determined. This framework allowed us to present
a systematic reduction of a snaketype robot called the snakeboard. It was demon
strated that the kinematic model of the snakeboard with antisymmetric heading
angles of the front and back wheels is diﬀeomorphic to the kinematic model of a car.
Moreover, the kinematics of the snakeboard can be transformed into a ﬁrstorder
chainedform nonholonomic system. The notations of stabilization and tracking
are formally deﬁned in chapter 6. Furthermore, global exponential tracking and
stabilization is achieved for a twowheeled mobile robot.
In chapter 7, the problem of stabilization of nonlinear systems in nontriangular
normal forms is addressed for several important special cases. This includes vector
ﬁelds that are aﬃne, quadratic, linearquadratic, and nonaﬃne in variables (ξ
1
, ξ
2
).
In addition, stabilization of slightly nontriangular systems that are small nonlinear
perturbations of systems in strict feedback forms is addressed. This has applications
in stabilization/tracking for the accurate model of a helicopter, ﬂexible linkrobots,
and in general underactuated systems with unactuated shape variables and/or input
coupling.
8.2 Future Work
This section provides potential future directions of research in continuation of this
work.
8.2.1 Hybrid Lagrangian Systems
From normal form of diﬀerent classes of underactuated systems, it is clear that the
shape variables have a fundamental role as the main control inputs of the reduced
subsystems. For example, in the case of classI underactuated systems, the reduced
270
system is itself a Lagrangian system where its Lagrangian function is parameterized
with the shape variables. One potential way to control the system is by switching
between a ﬁnite set of the values of the shape vector q
s
. Let Θ := ¦θ
1
, . . . , θ
k
¦ denote
a ﬁnite set of shape values θ
j
∈ Q
s
, j = 1, . . . , k where Q
s
denotes the shape manifold.
Now, consider the following nonlinear switching system
˙ q
r
= m
−1
r
(q
s
)p
r
˙ p
r
= g
r
(q
r
, q
s
)
(8.1)
where q
s
= v(t) and v(t) : R
≥0
→ Θ is a piecewise constant control input. A stabi
lizing control v(t) is a switching signal that appropriately puts one of the following
nonlinear dynamics in eﬀect at any time t ≥ 0
˙ q
r
= m
−1
r
(θ
j
)p
r
˙ p
r
= g
r
(q
r
, θ
j
)
(8.2)
such that (q
r
, p
r
) = 0 is asymptotically stable for (8.1). Setting z = col(q
r
, p
r
), this
nonlinear switching system can be rewritten as
˙ z = f
j
(z), z = col(q
r
, p
r
) (8.3)
According to [52], if there exists a common Lyapunov function V (z) such that
∇V (z) f
j
(z) < 0, ∀z = 0, j ∈ ¦1, . . . , k¦
Then, given any arbitrary switching signal v(t) : R
≥0
→Θ, z = 0 is globally asymp
totically stable for the closedloop system in (8.1). The existence of such a common
Lyapunov function is an extremely restrictive condition.
Consider a second scenario and assume g
r
(q
r
, q
s
) = g
r
(q
s
) and m
r
(q
s
) = m
r
is
a constant. In addition, g
r
(−q
s
) = −g
r
(q
s
) (i.e. g
r
(0) = 0) and ∃θ
0
: g
r
(θ
0
) > 0.
Setting Θ = ¦0, ±θ
0
¦, one possible stabilizing controller for the doubleintegrator
˙ q
r
= m
−1
r
p
r
˙ p
r
= u
(8.4)
with u = g
r
(v) is a timeoptimal controller that takes values in the set  = ¦0, ±¯ u¦
where ¯ u = g
r
(θ
0
) is a positive constant. Roughly speaking, stabilization using such a
piecewise constant control law is equivalent to stabilization of the reduced system by
switching between a ﬁnite set of possible shapes which the system could take. We call
this approach control via changing the shape. The method of “control via changing
the shape” is in complete agreement with natural locomotion in a swimming ﬁsh, a
ﬂying bird, and a walking human being . In the sense that in all the aforementioned
cases, locomotion is achieved via changing the physical shape of the system. A formal
and systematic analysis of the method of “changing the shape” for locomotion and
control purposes is needed.
271
8.2.2 Uncertainty and Robustness
Control of underactuated systems is rather challenging even in lack of any paramet
ric uncertainties. However, it is important to formulate and address robust stabi
lization/tracking problems for classes of underactuated systems where a satisfactory
nonlinear control design is available for them in lack of uncertainties (e.g. ClassI
systems). Normal forms for underactuated systems oﬀer a set of structured nonlinear
systems with a common core reduced system that can be exploited for construction
of Lyapunov functions. The obtained Lyapunov functions could be later employed in
robustness analysis of underactuated systems with parametric uncertainties.
8.2.3 Output Feedback Stabilization and Tracking
The majority of control laws presented in this work are in the form of full state feed
back. The exceptions to this statement include trajectory tracking for an autonomous
helicopter, a mobile robot, and a ﬂexiblelink robot. From theoretical point of view, it
is important to design nonlinear observers and develop output feedback stabilization
laws for control of underactuated systems (wherever possible). For example, in the
normal form of ClassI underactuated systems, the equation of the zerodynamics is
globally linear in the velocity (or momentum) z
2
. This promises the existence of an
output stabilization feedback for ClassI underactuated systems (under appropriate
technical conditions). The same argument applies to ClassIII systems and ClassII
systems without quadratic terms in z
2
.
8.2.4 The Eﬀects of Bounded Control Inputs
In reallife most of actuators are subject to limitations in their actuation power. A
formal analysis of nonlinear systems with saturated nonlinear state feedback similar
to the ones presented in sections 5.4.2 and 5.5.1 has yet to be developed.
8.2.5 Small Underactuated Mechanical Systems (SUMS)
Due to the availability of Microelectromechanical systems (MEMS), it is possible
to build underactuated microrobots and physically test the eﬀects of the relatively
small scale of the components of a robot in control design for underactuated systems.
This leads to control design for nonlinear systems with slightly nontriangular normal
forms. Control of SUMS is very challenging because of both the limitation of the
actuation power and the presence of parametric uncertainties that are inevitable in
dealing with semiconductor devices.
8.2.6 Underactuated Systems in Strong/Weak Fields
The behavior of the zerodynamics of underactuated systems in a strong potential
ﬁeld is dominant with the gravity terms, i.e. any nontriangular normal forms for such
underactuated systems are slightly nontriangular. In contrast, in zerogravity the
272
behavior of a ClassI underactuated system is the same as a mechanical system with
ﬁrstorder nonholonomic constraints. This suggests a future analysis for possible ex
tensions of some of the existing nonlinear control design approaches for nonholonomic
systems to control methods applicable to underactuated/nonholonomic mechanical
systems in lowgravity ﬁelds. This is of great importance in space exploration appli
cations.
273
Appendix A
Dynamics of Mechanical Systems
A.1 Underactuated Mechanical Systems with Two
DOF and Kinetic Symmetry
In this section, we present the closedform EulerLagrange equations of motion for a
mechanical system with two degrees of freedom (q
1
, q
2
) and kinetic symmetry with
respect to q
1
. This class of simple Lagrangian systems can be charachterize with the
following Lagrangian function
L(q, ˙ q) =
1
2
¸
˙ q
1
˙ q
2
T
¸
m
11
(q
2
) m
12
(q
2
)
m
21
(q
2
) m
22
(q
2
)
¸
˙ q
1
˙ q
2
−V (q
1
, q
2
) (A.1)
where V (q
1
, q
2
) is the potential energy of the system. From forced EulerLagrange
equations of motion
d
dt
∂L
∂ ˙ q
−
∂L
∂q
=
¸
F
1
(q)
F
2
(q)
τ (A.2)
where τ ∈ R is the control input and for all q ∈ Q either F
1
(q) = 0, or F
2
(q) = 0. For
the special cases where F(q) = (1, 0)
T
or F(q) = (0, 1), we say the system has non
interacting inputs. Otherwise, we say the system has inputcoupling. The equtions
of motion in (A.2) can be explicitly written as
m
11
(q
2
)¨ q
1
+m
12
(q
2
)¨ q
2
+ m
11
(q
2
) ˙ q
1
˙ q
2
+ m
12
(q
2
) ˙ q
2
2
+g
1
(q
1
, q
2
) = F
1
(q)τ
m
21
(q
2
)¨ q
1
+m
21
(q
2
)¨ q
2
−
1
2
m
11
(q
2
) ˙ q
2
1
+
1
2
m
22
(q
2
) ˙ q
2
2
+g
2
(q
1
, q
2
) = F
2
(q)τ
(A.3)
where
denotes d/dq
2
and the gravity terms are given by
g
i
(q
1
, q
2
) =
∂V (q
1
, q
2
)
∂q
i
, i = 1, 2
274
A.2 A Threelink Planar Robot
Consider a planar threelink robot arm with revolute joins depicted in Figure A1.
The mass, inertia, length, and length of the center of mass of the ith link is denoted
by m
i
, I
i
, L
i
, l
i
, respectively. To calculate the kinetic energy of this system we follow
a general method used in [95] which decomposes the kinetic energy as the sum of
translational and rotational kinetic energies, i.e. K = K
t
+ K
r
. The translational
kinetic energy is given by
K
t
=
1
2
v
T
mv
where v = (v
1
, v
2
, v
3
)
T
is the vector of linear velocities of the center of mass of the
links and m = diag(m
1
, m
2
, m
3
). The coordinates of the center of mass of three links
x
y
l
L
1
1
θ
θ
3
2
1
θ
, m ,
1
I
1
τ
τ
3
2
τ
1
Figure A1: A planar threelink robot arm.
are as follows
x
1
=
¸
l
1
sin θ
1
l
1
cos θ
1
x
2
=
¸
L
1
sin θ
1
+l
2
sin(θ
1
+ θ
2
)
L
1
cos θ
1
+l
2
cos(θ
1
+ θ
2
)
x
3
=
¸
L
1
sin θ
1
+ L
2
sin(θ
1
+θ
2
) + l
3
sin(θ
1
+ θ
2
+θ
3
)
L
1
cos θ
1
+L
2
cos(θ
1
+ θ
2
) +l
3
cos(θ
1
+θ
2
+ θ
3
)
275
Thus, denoting
˙
θ = (
˙
θ
1
,
˙
θ
2
,
˙
θ
3
)
T
, we have
v
1
= J
1
˙
θ =
¸
l
1
cos θ
1
0 0
−l
1
sin θ
1
0 0
˙
θ
1
˙
θ
2
˙
θ
3
¸
¸
v
2
= J
2
˙
θ =
¸
L
1
cos θ
1
+l
2
cos(θ
1
+ θ
2
) l
2
cos(θ
1
+ θ
2
) 0
−L
1
sin θ
1
−l
2
sin(θ
1
+ θ
2
) −l
2
sin(θ
1
+ θ
2
) 0
˙
θ
1
˙
θ
2
˙
θ
3
¸
¸
v
3
= J
3
˙
θ =
¸
L
1
cos θ
1
+ L
2
cos(θ
1
+ θ
2
) + l
3
cos(θ
1
+ θ
2
+ θ
3
)
−L
1
sin θ
1
−L
2
sin(θ
1
+ θ
2
) −l
3
sin(θ
1
+ θ
2
+ θ
3
)
L
2
cos(θ
1
+ θ
2
) + l
3
cos(θ
1
+ θ
2
+ θ
3
) l
3
cos(θ
1
+ θ
2
+ θ
3
)
−L
2
sin(θ
1
+ θ
2
) −l
3
sin(θ
1
+θ
2
+θ
3
) −l
3
sin(θ
1
+ θ
2
+ θ
3
)
˙
θ
1
˙
θ
2
˙
θ
3
¸
¸
This gives
K
t
=
1
2
˙
θ
T
(J
T
1
m
1
J
1
+ J
T
2
m
2
J
2
+ J
T
3
m
3
J
3
)
˙
θ
Now, let us calculate the rotational kinetic energy K
r
. Denoting the unit vector
orthogonal to the (x, y)frame by e
3
= (0, 0, 1)
T
, the angular velocities of the links in
reference frame are as follows
ω
1
=
˙
θ
1
e
3
ω
2
= (
˙
θ
1
+
˙
θ
2
)e
3
ω
3
= (
˙
θ
1
+
˙
θ
2
+
˙
θ
3
)e
3
Hence
K
r
=
1
2
ω
T
1
I
1
ω
1
+
1
2
ω
T
2
I
2
ω
2
+
1
2
ω
T
3
I
3
ω
3
=
1
2
˙
θ
T
I
˙
θ
where
I =
I
1
+ I
2
+I
3
I
2
+ I
3
I
3
I
2
+ I
3
I
2
+ I
3
I
3
I
3
I
3
I
3
¸
¸
Therefore, the total kinetic energy is given by
K =
1
2
˙
θ
T
(I + J
T
1
m
1
J
1
+ J
T
2
m
2
J
2
+ J
T
3
m
3
J
3
)
˙
θ
The inertia matrix of the system takes the following form
M = M(θ
2
, θ
3
) =
m
11
(θ
2
, θ
3
) m
12
(θ
2
, θ
3
) m
13
(θ
2
, θ
3
)
m
21
(θ
2
, θ
3
) m
22
(θ
3
) m
23
(θ
3
)
m
31
(θ
2
, θ
3
) m
32
(θ
3
) m
33
¸
¸
276
with elements
m
11
(θ
2
, θ
3
) = I
1
+ I
2
+ I
3
+ m
1
l
2
1
+ m
2
(L
2
1
+ l
2
2
) + m
3
(L
2
1
+ L
2
2
+ l
2
3
)
+ 2(m
2
L
1
l
2
+ m
3
L
1
L
2
) cos θ
2
+ 2m
3
L
1
l
3
cos(θ
2
+ θ
3
) + 2m
3
L
2
l
3
cos θ
3
m
12
(θ
2
, θ
3
) = m
21
= I
2
+I
3
+ m
2
l
2
2
+ m
3
(L
2
2
+ l
2
3
) + (m
2
L
1
l
2
+m
3
L
1
L
2
) cos θ
2
+ 2m
3
L
2
l
3
cos θ
3
+ m
3
L
1
l
3
cos(θ
2
+θ
3
)
m
13
(θ
2
, θ
3
) = m
31
= I
3
+m
3
l
2
3
+ m
3
L
1
l
3
cos(θ
2
+θ
3
) + m
3
L
2
l
3
cos θ
3
m
22
(θ
3
) = I
2
+ I
3
+ m
2
l
2
2
+ m
3
(L
2
2
+ l
2
3
) + 2m
3
L
2
l
3
cos θ
3
m
23
(θ
3
) = m
32
= I
3
+m
3
l
2
3
+ m
3
L
2
l
3
cos θ
3
m
33
= I
3
+ m
3
l
2
3
(A.4)
The potential energy of this threelink robot is given by
V (θ) = (m
1
l
1
+m
2
L
1
+ m
3
L
1
)g cos θ
1
+ (m
2
l
2
+ m
3
L
2
)g cos(θ
2
+ θ
3
)
+ m
3
l
3
g cos(θ
1
+ θ
2
+ θ
3
)
Notice that the potential energy V (θ) is identically zero for a threelink robot in a
horizontal plane. The EulerLagrange equations of motion for this system are as the
following
d
dt
∂L
∂
˙
θ
−
∂L
∂θ
= τ
where τ = (τ
1
, τ
2
, τ
3
)
T
is the input torque (control) and the Lagrangian L is given by
L(θ,
˙
θ) =
1
2
˙
θ
T
M(θ
2
, θ
3
)
˙
θ −V (θ)
Clearly, θ
1
is the external variable and θ
2
, θ
3
are the shape variables of this robot.
The following lemma is useful and shows that an threelink robot with one or two
actuators is an example of an underactuated system with nonintegrable normalized
momentums. It is important to notice that this property does not hold for a planar
twolink robot and both normalized momentums of the Acrobot (or the Pendubot)
are integrable.
Lemma A.2.1. All three normalized momentums
π
i
= m
−1
ii
(θ)
∂L
∂
˙
θ
i
, i = 1, 2, 3
of a planar threelink robot are nonintegrable, i.e. ∃h :
˙
h = π
1
.
Proof. First, we prove
π
1
=
˙
θ
1
+
m
12
(θ
2
, θ
3
)
m
11
(θ
2
, θ
3
)
˙
θ
2
+
m
13
(θ
2
, θ
3
)
m
11
(θ
2
, θ
3
)
˙
θ
3
is nonintegrable. For the purpose of contradiction, assume π
1
is integrable. Then,
277
π
1
−
˙
θ
1
is trivially integrable and the following oneform must be exact
ω
1
=
m
12
(θ
2
, θ
3
)
m
11
(θ
2
, θ
3
)
dθ
2
+
m
13
(θ
2
, θ
3
)
m
11
(θ
2
, θ
3
)
dθ
3
A suﬃcient and necessary condition that ω
1
is exact is
∂
∂θ
3
m
12
(θ
2
, θ
3
)
m
11
(θ
2
, θ
3
)
=
∂
∂θ
2
m
13
(θ
2
, θ
3
)
m
11
(θ
2
, θ
3
)
for all (θ
2
, θ
3
) ∈ S
1
S
1
. A rather lengthy but straightforward calculation shows that
the last equality does not hold for all θ
2
, θ
3
and therefore π
1
is nonintegrable. In a
similar way but after a short calculation, it can be shown that π
2
, π
3
are nonintegrable
normalized momentums as well.
Proposition A.2.1. A threelink robot in a horizontal plane or zero gravity (i.e.
V (θ) = 0) with two actuators at (θ
2
, θ
3
), a passive joint θ
1
, and zero initial velocity is
an underactuated mechanical system with a nonholonomic velocity constraint π
1
= 0.
In addition, there is an invertible change of control
v = α(θ)(τ
2
, τ
3
)
T
+ β(θ,
˙
θ)
that transforms the dynamics of the system into a driftless nonlinear system aug
mented with an integrator
˙
θ
1
= g
2
(θ
2
, θ
3
)u
2
+ g
3
(θ
2
, θ
3
)u
3
˙
θ
2
= u
2
˙
θ
3
= u
3
˙ u = v
(A.5)
where u = (u
2
, u
3
)
T
and
g
i
= −
m
1i
(θ
2
, θ
3
)
m
11
(θ
2
, θ
3
)
, i = 2, 3
Proof. Assume τ
1
= 0, then from EulerLagrange equations of motion
d
dt
∂L
∂
˙
θ
1
=
∂L
∂θ
1
= 0
Thus m
11
π
1
is a conserved quantity, i.e. m
11
π
1
= 0, ∀t ≥ 0. Since m
11
(θ
2
, θ
3
) >
0, based on lemma A.2.1, π
1
= 0 is a nonholonomic (i.e. nonintegrable) velocity
constraint. The normal form in (A.5) is a direct result of collocated partial feedback
linearization and deﬁnition of π
1
.
Theorem A.2.1. The inertia matrix of an nlink planar robot with n revolute joins
278
(θ
1
, . . . , θ
n
) (as shown in Figure A1 for n = 3) has the following structure
M(θ) =
m
11
(θ
2
, . . . , θ
n
) m
12
(θ
2
, . . . , θ
n
) . . . m
1n
(θ
2
, . . . , θ
n
)
m
21
(θ
2
, . . . , θ
n
) m
22
(θ
3
, . . . , θ
n
) . . . m
2n
(θ
3
, . . . , θ
n
)
.
.
.
.
.
.
.
.
.
.
.
.
m
n1
(θ
2
, . . . , θ
n
) m
n2
(θ
3
, . . . , θ
n
) . . . m
nn
¸
¸
¸
¸
¸
In other words, M(θ) = ¦m
ij
(θ
s
, . . . , θ
n
)¦ where s = min¦i, j¦ + 1 for i, j < n and
m
nn
is constant.
Proof. The proof is by direct calculation similar to the one applied to a threelink
robot. However, it is easier to prove this by induction. One assumes that the theorem
is true for k = 2 and k = n > 2 links and adds an (n+1)th link to the end point of the
nth link. It is rather straightforward to prove that the theorem holds for k = n + 1.
This proves that the theorem holds for and n ≥ 2.
A.3 Twolink Planar Robots: the Acrobot and the
Pendubot
A twolink planar robot with joint angles θ
1
, θ
2
can be viewed as a special case of a
threelink planar robot with l
3
= 0 (and thus m
3
= I
3
= 0)(see Figure A1). In other
words, the inertia matrix for a twolink planar robot with revolute joins is given by
m
11
(θ
2
) = I
1
+ I
2
+ m
1
l
2
1
+ m
2
(L
2
1
+ l
2
2
) + 2m
2
L
1
l
2
cos(θ
2
)
m
12
(θ
2
) = I
2
+ m
2
l
2
2
+ m
2
L
1
l
2
cos(θ
2
)
m
22
= I
2
+ m
2
l
2
2
(A.6)
and m
21
(θ
2
) = m
12
(θ
2
). The potential energy of a twolink robot in a vertical plane
is in the form
V (θ
1
, θ
2
) = (m
1
l
1
+ m
2
L
1
)g cos(θ
1
) + m
2
l
2
g cos(θ
1
+θ
2
) (A.7)
Based on equation (A.3), the EulerLagrange equations of motion for this twolink
robot can be expressed as
m
11
(θ
2
)
¨
θ
1
+ m
12
(θ
2
)
˙
θ
2
+ h
1
(θ,
˙
θ) = F
1
τ
m
21
(θ
2
)
¨
θ
1
+ m
22
(θ
2
)
˙
θ
2
+ h
2
(θ,
˙
θ) = F
2
τ
(A.8)
where
h
1
(θ,
˙
θ) = −m
2
L
1
l
2
sin(θ
2
)(2
˙
θ
1
˙
θ
2
+
˙
θ
2
2
) + g
1
(θ
1
, θ
2
)
h
2
(θ,
˙
θ) = m
2
L
1
l
2
sin(θ
2
)
˙
θ
2
1
+ g
2
(θ
1
, θ
2
)
g
1
(θ
1
, θ
2
) = −(m
1
l
1
+ m
2
L
1
)g sin(θ
1
) −m
2
gl
2
sin(θ
1
+ θ
2
)
g
2
(θ
1
, θ
2
) = −m
2
gl
2
sin(θ
1
+ θ
2
)
279
Depending on whether θ
1
is actuated and θ
2
is unactuated, or vice versa; two impor
tant examples of underactuated systems with two degrees of freedom arise.
Deﬁnition A.3.1. (Acrobot and Pendubot) The underactuated system in (A.8) is
called Acrobot if (F
1
, F
2
) = (0, 1). It is called Pendubot if (F
1
, F
2
) = (1, 0).
Deﬁning the state vector x = (q
1
, ˙ q
1
, q
2
, ˙ q
2
), the dynamics of the the system in
(A.8) can be expressed as a nonlinear system aﬃne in control
˙ x = f(x) +g
i
(x)u (A.9)
where u = τ and i = a, p corresspond to the Acrobot and the Pendubot, respectively.
Apprently, the drift vector ﬁeld f(x) is the same for the Acrobot and the Pendubot.
However, the control vector ﬁelds g
1
, g
2
are diﬀerent and can be expressed as
g
a
(q
2
) =
1
D(q
2
)
0
−m
12
(q
2
)
0
m
11
(q
2
)
¸
¸
¸
¸
, g
p
(q
2
) =
1
D(q
2
)
0
m
22
(q
2
)
0
−m
21
(q
2
)
¸
¸
¸
¸
where D(q
2
) = det(M(q
2
)) = m
11
m
22
−m
12
m
21
> 0.
Remark A.3.1. The distribution generated by nested Lie brackets of f(x), g
a
(x) is
not the same as the distribution generated by f(x), g
p
(x); unless g
p
(x) = λg
a
(x) for
a constant λ ∈ R. But this means m
21
= −λm
11
and m
22
= −λm
12
, or equivalently
D(q
2
) = 0 which contradicts the property D(q
2
) > 0. Therefore, the controllability
properties of the Acrobot and the Pendubot are diﬀerent.
A.4 The BeamandBall System
The beamandball system consists of a beam and a ball with radius r on it (see
Figure A2). Let d ≥ 0 denote the distance between the center of the mass of the
ball and the beam (d = r in Figure 524). In the current literature [34, 102, 80, 79],
what is known as the beamandball system corresponds to the special case where
d = 0. We refer to this case as the conventional beamandball system and to the case
where d > 0 as the beamandball system. The following treatment applies to the
both cases of d = 0 and d > 0. Let I
1
, m, I
2
denote the inertia of the beam, the mass
of the ball , and the inertia of the ball, respectively. The position and velocity of the
center of mass of the ball is
x
c
= (xcos(θ) −d sin(θ), xsin(θ) + d cos(θ))
v
c
= (v cos(θ) −xsin(θ)ω −d cos(θ)ω, v sin(θ) + xcos(θ)ω −d sin(θ)ω)
where ˙ x = v and
˙
θ = ω. Thus, the translational part of the kinetic energy is given by
K
trans
=
1
2
mv
2
c
=
1
2
m(v
2
+ (x
2
+ d
2
)ω
2
−2dωv)
280
d
x
r
I
1
θ
τ
m , I
2
Figure A2: The BeamandBall System.
The rotational kinetic energy of the beamandball system is
K
rot
=
1
2
I
1
ω
2
+
1
2
I
2
(
v
r
)
2
Hence, the total kinetic energy K = K
trans
+ K
rot
equals
K =
1
2
(I
1
+ m(x
2
+d
2
))ω
2
−mdωv +
1
2
λmv
2
where
λ = 1 +
I
2
mr
2
> 1
is a constant. The potential energy of the beamandball system is given by
V (x, θ) = mg(xsin θ + d cos θ)
Therefore, the Lagrangian of the beamandball system is as the following
L =
1
2
¸
˙
θ
˙ x
T
¸
I
1
+ m(x
2
+ d
2
) −md
−md mλ
¸
˙
θ
˙ x
−mg(xsin(θ) + d cos(θ)) (A.10)
Based on (A.10), the elements of the inertia matrix of the beamandball system are
m
11
(x) = I
1
+ m(x
2
+d
2
)
m
12
= m
21
= −md
m
22
= mλ
inertia matrix of the beamandball system only depends on the position of the ball
x. This means that the position of the ball is the shape variable of the beamand
ball system which is unactuated. Therefore, the beamandball system is a Class–II
underactuated system. The EulerLagrange equations of motion for the beamandball
system with d > 0 is given by
[I
1
+ m(x
2
+ d
2
)]
¨
θ − md¨ x + 2mx ˙ x
˙
θ +mg(xcos(θ) −d sin(θ)) = τ
d
−md
¨
θ + mλ¨ x −mx
˙
θ
2
+ mg sin(θ) = 0
(A.11)
281
While for the conventional beamandball system with d = 0, the EulerLagrange
equations take the following form
(I
1
+ mx
2
)
¨
θ + 2mx ˙ x
˙
θ + mgxcos(θ) = τ
0
mλ¨ x −mx
˙
θ
2
+ mg sin(θ) = 0
(A.12)
A.5 The CartPole System
The CartPole system depicted in Figure A3 consists of a pendulum on a cart. Let
(q
1
, q
2
) denote the conﬁguration vector of the CartPole system. The coordinates of
m
F
m , l , I
1
2 2 2
q
q
1
2
Figure A3: The CartPole system.
the center of mass of the cart and pendulum are given by
x
1
=
¸
q
1
0
, x
2
=
¸
q
1
+ l
2
sin(q
2
)
l
2
cos(q
2
)
with linear velocities
˙ x
1
=
¸
˙ q
1
0
, ˙ x
2
=
¸
˙ q
1
+ l
2
cos(q
2
) ˙ q
2
−l
2
sin(q
2
) ˙ q
2
Thus, the total kinetic energy of the system is
K =
1
2
m
1
[ ˙ x
1
[
2
+
1
2
m
2
[ ˙ x
2
[
2
+
1
2
I
2
˙ q
2
2
After simpliﬁcation, we obtain
K =
1
2
(m
1
+ m
2
) ˙ q
2
1
+ m
2
l
2
cos(q
2
) ˙ q
1
˙ q
2
+
1
2
(m
2
l
2
2
+ I
2
) ˙ q
2
2
The potential energy of the CartPole system is in the form
V (q) = V (q
2
) = m
2
gl
2
cos(q
2
)
282
Therefore, the Lagrangian of the CartPole system can be expressed as
L(q, ˙ q) =
1
2
¸
˙ q
1
˙ q
2
T
¸
m
1
+m
2
m
2
l
2
cos(q
2
)
m
2
l
2
cos(q
2
) m
2
l
2
2
+I
2
¸
˙ q
1
˙ q
2
−m
2
gl
2
cos(q
2
) (A.13)
and the elements of the inertia matrix for the CartPole system are given by
m
11
= m
1
+ m
2
m
12
(q
2
) = m
21
(q
2
) = m
2
l
2
cos(q
2
)
m
22
= m
2
l
2
2
+ I
2
(A.14)
The EulerLagrange equations of motion for the CartPole system is in the form
d
dt
∂L
∂ ˙ q
−
∂L
∂q
=
¸
F
1
0
In other words, the CartPole system is an underactuated system with two DOF and
an unactuated shape variable q
2
.
A.6 The Rotating Pendulum
The Rotating Pendulum consists of an inverted pendulum on a rotating arm as shown
in Figure A4. This system was ﬁrst introduced in [6]. The coordinates of the center
x
y
z
1
2
q
m , L , I
m , L , I
2
1 1
2
q
1
2
Figure A4: The Rotating Pendulum.
of mass of both links are given by
x
1
=
l
1
cos(q
1
)
l
1
sin(q
1
)
0
¸
¸
, x
2
=
L
1
cos(q
1
) −l
2
sin(q
1
) sin(q
2
)
L
1
sin(q
1
) +l
2
cos(q
1
) sin(q
2
)
l
2
cos(q
2
)
¸
¸
283
with linear velocities
˙ x
1
=
−l
1
sin(q
1
) ˙ q
1
l
1
cos(q
1
) ˙ q
1
0
¸
¸
, ˙ x
2
=
−L
1
sin(q
1
) ˙ q
1
−l
2
cos(q
1
) sin(q
2
) ˙ q
1
−l
2
sin(q
1
) cos(q
2
) ˙ q
2
L
1
cos(q
1
) ˙ q
1
−l
2
sin(q
1
) sin(q
2
) ˙ q
1
+ l
2
cos(q
1
) cos(q
2
) ˙ q
2
−l
2
sin(q
2
) ˙ q
2
¸
¸
The kinetic energy of the Rotating Pendulum is
K =
1
2
m
1
[ ˙ x
1
[
2
+
1
2
m
2
[ ˙ x
2
[
2
+
1
2
I
1
˙ q
2
1
+
1
2
I
2
˙ q
2
which can be simpliﬁed as
K =
1
2
(m
1
l
2
1
+ I
1
+ m
2
L
2
1
+m
2
l
2
2
sin
2
(q
2
)) ˙ q
2
1
+ m
2
l
2
cos(q
2
) ˙ q
1
˙ q
2
+
1
2
(m
2
l
2
2
+ I
2
) ˙ q
2
2
The potential energy of the Rotating Pendulum is given by
V (q) = V (q
2
) = m
2
gl
2
cos(q
2
)
Thus, the Lagrangian of the Rotating Pendulum can be written as
L =
1
2
¸
˙ q
1
˙ q
2
T
¸
m
1
l
2
1
+ I
1
+ m
2
L
2
1
+ m
2
l
2
2
sin
2
(q
2
) m
2
l
2
cos(q
2
)
m
2
l
2
cos(q
2
) m
2
l
2
2
+ I
2
¸
˙ q
1
˙ q
2
−m
2
gl
2
cos(q
2
)
(A.15)
Therefore, the elements of the inertia matrix for the Rotating Pendulum are given by
m
11
(q
2
) = m
1
l
2
1
+ I
1
+ m
2
L
2
1
+ m
2
l
2
2
sin
2
(q
2
)
m
12
(q
2
) = m
21
(q
2
) = m
2
l
2
cos(q
2
)
m
22
= m
2
l
2
2
+ I
2
(A.16)
The EulerLagrange equations of motion for the Rotating Pendulum is as the following
d
dt
∂L
∂ ˙ q
−
∂L
∂q
=
¸
τ
1
0
which means the Rotating Pendulum is an underactuated system with two DOF and
one unactuated shape variable q
2
.
A.7 The 3D CartPole System with Unactuated
Roll and Pitch Angles
Consider an inverted pendulum mounted on a moving platform via an unactuated 2
DOF joint (θ, φ) (i.e. the roll and pitch angles) with two external forces as shown in
Figure A5. The angles θ, φ denote rotation around x
2
axis and x
1
axis, respectively.
Denote the positions of the centers of mass of M and m as the following by x
c
1
and
284
F
F
2
1
x
3
x
1
2
x
M
m
a
φ
θ
l
φ
Figure A5: The 3D CartPole system with unactuated roll and pitch angles.
x
c
2
, respectively.
x
c
1
=
x
1
x
2
0
¸
¸
, x
c
2
=
x
1
+ l sin(θ)
x
2
+ l cos(θ) sin(φ)
a +l cos(θ) cos(φ)
¸
¸
The linear velocities associated with (x
c
1
, x
c
2
) are given by
˙ x
c
1
=
˙ x
1
˙ x
2
0
¸
¸
, ˙ x
c
2
=
˙ x
1
+ l cos(θ)
˙
θ
˙ x
2
−l sin(θ) sin(φ)
˙
θ + l cos(θ) cos(φ)
˙
φ
−l sin(θ) cos(φ)
˙
θ −l cos(θ) sin(φ)
˙
φ
¸
¸
The kinetic and potential energy of the system are
K =
1
2
M ˙ x
T
c
1
˙ x
c
1
+
1
2
m˙ x
T
c
2
˙ x
c
2
, V (q) = mgl cos θ cos φ
The expression for the kinetic energy can be simpliﬁed as
K =
1
2
˙ x
1
˙ x
2
˙
θ
˙
φ
¸
¸
¸
¸
T
M + m 0 ml cos θ 0
0 M + m −ml sin θ sin φ ml cos θ cos φ
ml cos θ −ml sin θ sin φ ml
2
0
0 ml cos θ cos φ 0 ml
2
cos
2
θ
¸
¸
¸
¸
˙ x
1
˙ x
2
˙
θ
˙
φ
¸
¸
¸
¸
(A.17)
Clearly, the inertia matrix for this system only depends on (θ, φ). Therefore, (θ, φ)
are the shape variables and (x
1
, x
2
) are the external variables of the 3D CartPole
system. The EulerLagrange equations of motion for the 3D CartPole system with
q = (x
1
, x
2
, θ, φ)
T
is as the following
d
dt
∂L
∂ ˙ q
−
∂L
∂q
=
F
1
F
2
0
0
¸
¸
¸
¸
285
In other words, the 3D CartPole System is an underactuated system with four DOF
and two unactuated shape variables. The equations of motion for the 3D CartPole
system can be explicitly written as
(M +m)¨ x
1
+ ml cos θ
¨
θ = ml sin θ
˙
θ
2
+ F
1
(M +m)¨ x
2
−ml sin θ sin φ
¨
θ +ml cos θ cos φ
¨
φ = ml cos θ sin φ
˙
θ
2
+ 2ml sin θ cos φ
˙
θ
˙
φ
+ ml cos θ sin φ
˙
φ
2
+F
2
ml cos θ¨ x
1
−ml sin θ sin φ¨ x
2
+ ml
2
¨
θ = mgl sin θ cos φ −ml
2
sin θ cos θ
˙
φ
2
ml cos θ cos φ¨ x
2
+ ml
2
cos
2
θ
¨
φ = mgl cos θ sin φ + 2ml
2
sin θ cos θ
˙
θ
˙
φ
(A.18)
A.8 An Underactuated Surface Vessel
A surface vessel with no side thruster is illustrated in Figure A6. Thus, it is an
underactuated system with three degrees of freedom and two actuators. The conﬁg
uration vector for this vessel is η = (x, y, θ) where (x, y) denotes the position and θ
determines the orientation of the vessel (i.e. the yaw angle). The velocity vector in
the body frame is ν = (v
1
, v
2
, ω) where (v
1
, v
2
) are the linear velocities and ω is the
angular velocity. The complete dynamics of an underactuated surface vessel can be
θ
Y
X
Figure A6: An underactuated surface vessel.
found in [30, 111]. Deﬁning the following rotation matrix
R(θ) =
cos(θ) −sin(θ) 0
sin(θ) cos(θ) 0
0 0 1
¸
¸
we have
˙ η = R(θ)ν
The kinetic energy of this system is in the form
K =
1
2
ν
T
Jν
286
with J = diag(J
1
, J
2
, J
3
). Thus, the kinetic energy in the reference frame is
K =
1
2
˙ η
T
M ˙ η
which means the inertia matrix of a surface vessel is
M = R(θ)JR
T
(θ) = J
or the surface vessel is a ﬂat underactuated mechanical system. The potential energy
of a surface vessel is identically zero. The rotation matrix R satisﬁes
˙
R = Rˆ ω
where ˆ ω is the following skew symmetric matrix
ˆ ω =
0 ω 0
−ω 0 0
0 0 0
¸
¸
The dynamics of this surface vessel can be expressed as
˙ η = R(θ)ν
J ˙ ν = −ˆ ωJν −Dν +
τ
1
0
τ
3
¸
¸
(A.19)
where Dν is a damping term with D = diag(d
1
, d
2
, d
3
) and d
i
> 0. The term ˆ ωJν
has the following explicit form
ˆ ωJν =
0 −J
2
v
2
ω 0
J
1
v
1
ω 0 0
0 0 0
¸
¸
= C(ν)ν
where
C(ν) =
0 0 J
2
v
2
0 0 −J
1
v
1
−J
2
v
2
J
1
v
1
0
¸
¸
The overall dynamics of the surface vessel can be rewritten as
˙ η = R(θ)ν
J ˙ ν + C(ν)ν + Dν =
τ
1
0
τ
3
¸
¸
(A.20)
From the second line of the dynamic equation in (A.20), we have
J
2
˙ v
2
+ J
1
v
1
ω + d
2
v
2
= 0
287
which is a nonholonomic secondorder constraint (i.e. a constraint that involves
secondorder time derivatives of the conﬁguration variables) that holds regardless
of the control inputs τ
1
, τ
3
.
288
Appendix B
Symbolic Calculation of the Flat
Outputs of the Helicopter
This appendix provides a MATLAB code for symbolic calculations required to prove
that the vector of oneforms ω
xs
= R(η)EJΨ(η)dη in proposition 5.11.2 has no ex
act elements. In addition, using momentum decomposition, an exact part of ω
xs
is
obtained that is denoted by C(η)dη in the following.
% (a,b,c)=(φ, θ, ψ)=(roll,pitch,yaw) are three Euler angles.
syms a b c;
J1=sym(’J1’);J2=sym(’J2’);J3=sym(’J3’); J=diag([J1,J2,J3]);
e1=sym(’e1’);e2=sym(’e2’);e3=sym(’e3’); E=[0 e2 0; e1 0 e3; 0 0 0];
% A=Ψ(η) is the inverse of the Euler matrix.
A=[1 0 sin(b); 0 cos(a) cos(b)*sin(a); 0 sin(a) cos(b)*cos(a)];
R1=[1 0 0; 0 cos(a) sin(a); 0 sin(a) cos(a)];
R2=[cos(b) 0 sin(b); 0 1 0 ; sin(b) 0 cos(b)];
R3=[cos(c) sin(c) 0; sin(c) cos(c) 0; 0 0 1];
R=simplify(R3*R2*R1); G=simplify(R*E*J*A);
% if d1,d2,d3 are not identically zero, G(η)dη is not exact.
d1=simplify(diff(G(:,2),’c’)diff(G(:,3),’b’));
d2=simplify(diff(G(:,1),’c’)diff(G(:,3),’a’));
d3=simplify(diff(G(:,1),’b’)diff(G(:,2),’a’));
% Observation: none of d1,d2,d3 are identically zero
% A=A(a,b); A1=A(a,0); A2=A(0,b); A3=A(0,0);
A1=[1 0 0; 0 cos(a) sin(a); 0 sin(a) cos(a)];
A2=[1 0 sin(b); 0 1 0; 0 0 cos(b)];
A3=eye(3);
G1=simplify(R1*E*J*A1); C1=G1(:,1); G2=simplify(R2*E*J*A2); C2=G2(:,2);
G3=simplify(R3*E*J*A3); C3=G3(:,3);
% The obtained locked vector of oneforms ω
l
= C(η)dη has exact elements
C=[C1 C2 C3];
C = [ 0, cos(b)*e2*J2, sin(c)*e3*J3] [ cos(a)*e1*J1, 0, cos(c)*e3*J3] [
sin(a)*e1*J1, sin(b)*e2*J2, 0]
289
Appendix C
Proofs of Theorems in Chapter 7
C.1 Proof of Theorem 7.4.3:
Equation (7.93) can be rewritten as
ξ
2
= D
z
1
α
κ
(z) z
2
+D
z
2
α
κ
(z) f(z, α
κ
(z), ξ
2
) =: ψ(z, ξ
2
)
Since D
z
2
α
κ
(z) = 0, ∀z
2
∈ /
0
, the ﬁxed point equation (7.93) has a (trivial) smooth
unique solution as the following
β(z) = D
z
1
α
κ
(z) z
2
for all z
2
∈ /
0
. For the rest of the proof, assume z
2
∈ /
0
and set r
0
= max
z
2
∈K
0
z
2
.
Denote
/
ξ
1
= B(0, L
0
) ⊂ R
/
ξ
2
= B(0, 1) ⊂ R
(where B(x
0
, r) = ¦x ∈ R
nx
: [x − x
0
[ ≤ r¦) and set S := R
n
/
0
/
ξ
2
. Based on
Assumption (7.4.9), over S we have
[ψ(z, ξ
2
)[ ≤ D
z
1
α
κ
(z) z
2
 +D
z
2
α
κ
(z) (f
0
(z, ξ
1
) +g
0
(z, ξ
1
, ξ
2
) ξ
2

2
)
≤ (L
0
+ L
2
)r
0
+ L
1
((a
1
+ a
2
)κ(z
1
) + (b
1
+ b
2
)[z
1
κ(z
1
)[)
≤ (L
0
+ L
2
)r
0
+ L
1
(a
1
+ a
2
+ b
1
+ b
2
) ≤ 1
where the last inequality can be achieved by taking
L
0
, L
2
≤
1
3
r
0
−1
, L
1
≤
1
3
(a
1
+ a
2
+ b
1
+ b
2
)
−1
This guarantees that ψ(S) ⊆ /
ξ
2
. Calculating D
ξ
2
ψ(z, ξ
2
) for all (z, ξ
2
) ∈ S, we get
[D
ξ
2
ψ(z, ξ
2
)[ ≤ D
z
2
α
κ
(z) [D
ξ
2
f(z, α
κ
(z), ξ
2
)[
≤ L
1
(a
3
κ(z
1
) + b
3
z
1
κ(z
1
))
≤ L
1
(a
3
+ b
3
) < 1
290
where the last inequality can be satisﬁed for L
1
< (a
3
+ b
3
)
−1
that guarantees the
mapping ψ(z, ξ
2
) is contractive. Thus, after taking
L
∗
0
= L
∗
2
=
1
3
r
0
−1
, L
∗
1
= min¦
1
3
(a
1
+ a
2
+b
1
+ b
2
)
−1
, (a
3
+ b
3
)
−1
¦
the result follows from the contraction mapping theorem. The property β(0) = 0 is
due to the uniqueness of the solution ξ
∗
2
= 0 of (7.93) at z = 0.
C.2 Proof of the Stability of µSubsystem in The
orem 7.4.4
Lemma C.2.1. The origin (µ
1
, µ
2
) = 0 for
˙ µ
1
= −σ(µ
1
) + µ
2
˙ µ
2
= −σ(µ
2
)
(C.1)
is globally asymptotically and locally exponentially stable.
Proof. First, in the linearization of this system ˙ µ = Aµ, A is a Hurwitz matrix
and µ = 0 is locally exponentially stable. On the other hand, µ
2
= 0 is globally
asymptotically stable for ˙ µ
2
= −σ(µ
2
) (because V
2
(µ
2
) = µ
2
2
is a valid Lyapunov
function that satisﬁes
˙
V
2
< 0 for µ
2
= 0). Thus, for any solution µ
2
(t) of ˙ µ
2
= −σ(µ
2
)
with µ
2
(0) ∈ R, [µ
2
(t)[ ≤ [µ
2
(0)[ for all t ≥ 0. Taking V
1
(µ
1
) =
1
2
µ
2
1
, we have
˙
V
1
= −µ
1
σ(µ
1
) + µ
1
µ
2
≤ −[µ
1
[σ([µ
1
[) +[µ
1
[ [µ
2
[
≤ −[µ
1
[(σ([µ
1
[) −[µ
2
(0)[)
Let r
µ
1
:= σ
−1
([µ
2
(0)[) and observe that for [µ
1
[ > r
µ
1
,
˙
V
1
< 0 and any solution
starting out of the closed ball B(0, r
µ
1
), enters B(0, r
µ
1
) after some ﬁnite time T
0
>
0 and remains in it thereafter. Meaning that the solutions of system (C.1) stay
uniformly bounded for all t > 0 (with a bound that depends on µ(0)). But given
µ
2
= 0 for ˙ µ
1
= −σ(µ
1
), µ
1
= 0 is GAS. Therefore, based on Sontag’s theorem on
stability of cascade nonlinear systems [83], (µ
1
, µ
2
) = 0 is GAS for the cascade (C.1).
C.3 Proof of Lemma 7.4.4:
First, we construct K
y
. Take r
0
= max
x∈Kx
[φ
0
(x)[,
˜
/
y
= B(0, 2r
0
), and r
1
=
max [φ
1
(x, y, )[ over /
x
˜
/
y
I
. We have
[φ(x, y, )[ ≤ [φ
0
(x)[ + [φ
1
(x, y, )[ ≤ r
2
:= 2r
0
291
for all 0 < ≤
2
:= min¦r
0
/r
1
,
1
¦. Thus, setting /
y
=
˜
/
y
, φ maps /
x
/
y
˜
I
into /
y
where
˜
I
= [0,
2
]. Deﬁne
L
y
= max¦[
∂φ
1
∂y
(x, y, )[, 1¦ ≥ 1
over /
x
/
y
˜
I
and set
∗
= min¦
2
, 1/L
y
¦. Then, for all for all 0 < <
∗
the
following holds
[
∂φ
∂y
[ = [
∂φ
1
∂y
[ ≤ L
y
< 1
over /
x
/
y
(0,
∗
] and the result follows from contraction mapping theorem (or
Banach ﬁxed point theorem). In addition, due to uniqueness of the ﬁxed point,
α(0, ) = 0 independent of .
C.4 Proof of Theorem 7.4.4:
Noting that ∂ψ
3
/∂u and ∂ψ
2
/∂ξ
2
are equal and both vanish at = 0, the proof of
(i) and (ii) follows directly from lemma 7.4.4 with appropriate choices of x, y. To
prove (iii), ﬁrst we show that α
2
locally asymptotically stabilizes the origin (z, ξ
1
) =
0 for the (z, ξ
1
)subsystem in (7.101). Then, we prove any solution of the system
starting in a compact set of initial conditions for a suﬃciently small > 0 stays
uniformly bounded and asymptotically converges to the origin. For doing so, take
µ
1
= ξ
1
− α
1
(z), we have ˙ µ
1
= −σ(µ
1
) and immediately it follows that µ
1
= 0
is globally asymptotically and locally exponentially stable for this system (with a
Lyapunov function V
1
(µ
1
) = µ
2
1
). The dynamics of the system in new coordinates is
a cascade nonlinear system as the following
˙ z = f
0
(z, α
1
(z) + µ
1
) + g
0
(z, α
1
(z) + µ
1
, α
2
(z, α
1
(z) + µ
1
, ), )
˙ µ
1
= −σ(µ
1
)
(C.2)
or
˙ z = f
0
(z, α
1
(z)) + h
0
(z, α
1
(z), µ
1
)µ
1
+ ˜ g
0
(z, µ
1
, )
˙ µ
1
= −σ(µ
1
)
(C.3)
where
˜ g
0
(z, µ
1
, ) = g
0
(z, α
1
(z) +µ
1
, α
2
(z, α
1
(z) + µ
1
, ), )
Note that g
0
can be rewritten as the following
g
0
(z, ξ
1
, ξ
2
, ) = p
0
(z, ξ
1
, ξ
2
, )ξ
2
2
where p
0
is a smooth function. Thus
∂˜ g
0
∂z
(z,µ
1
)=0
=
¸
∂p
0
∂z
α
2
2
+
∂p
0
∂ξ
1
∂α
1
(z)
∂z
α
2
2
+ 2α
2
∂α
2
∂z
p
0
(z,µ
1
)=0
= 0
292
where in the last equation ξ
1
= α
1
(z) + µ
1
, ξ
2
= α
2
(z, ξ
1
, ), and α
2
(0, 0, ) = 0.
Calculating the Jacobian matrix of (C.3) at the origin (z, µ
1
) = 0, we obtain
A
z,µ
1
=
∂f
0
∂z
(z, α
1
(z)) +
∂f
0
∂ξ
1
(z, α
1
(z))
∂α
1
∂z
+ (
∂h
0
∂z
+
∂h
0
∂ξ
1
∂α
1
∂z
)µ
1
+
∂˜ g
0
∂z
∗
0 −
dσ(µ
1
)
dµ
1
¸
¸
¸
(z,µ
1
)=(0,0)
or
A
z,µ
1
=
¸
A
z
∗
0 −σ
(0)
where
A
z
=
∂f
0
∂z
(z, α
1
(z)) +
∂f
0
∂ξ
1
(z, α
1
(z))
∂α
1
(z)
∂z
z=0
is a Hurwitz matrix (by assumption, ξ
1
= α
1
(z) locally exponentially stabilizes z = 0
for ˙ z = f
0
(z, ξ
1
)) and −σ
(0) < 0. Therefore, (z, µ
1
) = (0, 0) is locally exponentially
stable for (C.3) and there exists a local region of attraction U
0
as a small open
neighborhood of the origin (z, µ
1
) = 0 that is forward invariant in time. Now, we
show the solutions of (C.3) starting in a compact set stay uniformly bounded. To do
so, consider the following Lyapunov function candidate
W(z, µ
1
) = V (z) +
1
2
µ
2
1
Calculating
˙
W along the solutions of the closedloop system, we get
˙
W =
∂V (z)
∂z
f
0
(z, α
1
(z)) +
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)µ
1
+
∂V (z)
∂z
˜ g
0
(z, µ
1
, ) −µ
1
σ(µ
1
)
Take r
1
≥ r
0
to be suﬃciently large such that B(0, r
1
) ⊃ U
0
. Consider the compact
set Ω(c
1
) = ¦(z, µ
1
)[W(z, µ
1
) ≤ c
1
¦ (where c
1
> 0 to be determined later) with
c
1
> c
2
> 0 such that Ω(c
2
) ⊃ (/
0
∪ B(0, r
1
) (Ω(c
2
) is ﬁxed). Then, there exists
1
(c
1
) > 0 such that for all 0 < <
1
(c
1
), we have
∂V (z)
∂z
˜ g
0
(z, µ
1
, )
< −
∂V (z)
∂z
f
0
(z, α
1
(z)) + µ
1
σ(µ
1
)
over Ω(c
1
) ` U
0
. Thus, we have
˙
W <
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)µ
1
(t)
≤
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)
[µ
1
(t)[
≤ kV (z)
λ
[µ
1
(t)[ , ∀[z[ > r
1
, ∀t ≥ 0
293
First, assume λ = 1. It follows that
˙
W < kV (z)[µ
1
(t)[ ≤ kW[µ
1
(t)[
but µ
1
(t) is an L
1
function and ∃l
1
> 0 :
∞
0
[µ
1
(t)[dt ≤ l
1
. This means that
W(z(t), µ
1
(t)) < W(z(0), µ
1
(0)) exp(kl
1
) =: c
3
where c
3
> 0 is a constant that only depends on k, l
1
, W(z(0), µ
1
(0)). In other words,
c
3
is independent of > 0. Thus, taking a ﬁxed c
1
> max¦c
2
, c
3
¦ determines a
compact set /
1
:= Ω(c
1
) ⊃ /
0
such that the solutions (z(t), µ
1
(t)) ∈ /
1
for all t ≥ 0.
The case 0 < λ < 1 is rather similar, we get
˙
W < kV (z)
λ
≤ kW(z, µ
1
)
λ
Thus, c
3
can be deﬁned in the following way
d
dt
(W
(1−λ)
) < (1 −λ)kµ
1
(t)
or
W(z, µ
1
) < (W(z(0), µ
1
(0))
(1−λ)
+ (1 −λ)kl
1
)
1
(1−λ)
=: c
3
(λ) , ∀t > 0
and the boundedness result follows (also, observe that lim
λ→1
− c
3
(λ) = c
3
for λ = 1).
It remains to prove that the solutions starting in /
0
asymptotically converge to the
origin. Over the compact set /
1
, take
l
2
= max
(z,µ
1
)∈K
1
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)
and
0
= max
(z,µ
1
)∈K
1
\U
0
−
∂V (z)
∂z
f
0
(z, α
1
(z)) +µ
1
σ(µ
1
) > 0
Then, there exists
2
> 0 such that for all 0 < <
2
∂V (z)
∂z
˜ g
0
(z, µ
1
, )
<
0
3
Also, because µ
1
(t) →0 as t →∞, there exists T
0
> 0 such that for all t > T
0
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)µ
1
≤ l
2
[µ
1
(t)[ <
0
3
294
Therefore, over /
1
` U
0
, we have
˙
W ≤
∂V (z)
∂z
f
0
(z, α
1
(z)) −µ
1
σ(µ
1
) +
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)µ
1
+
∂V (z)
∂z
˜ g
0
(z, µ
1
, )
< −
0
+
0
3
+
0
3
= −
0
3
< 0
This means that any solution of the system starting in /
0
stays in /
1
for all t > 0
and enters U
0
after some ﬁnite time T
1
≥ T
0
> 0 and then asymptotically converges
to (z, µ
1
) = 0, or (z, ξ
1
) = 0. This ﬁnishes the proof of (iii). The proof of (iv) is very
similar to (iii) and is omitted.
C.5 Proof of Theorem 7.4.5:
The proof of (i) is based on contraction mapping theorem (see lemma 7.4.4). To prove
(ii), ﬁrst we establish local asymptotic stability of the origin (z, ξ
1
) = 0 for the closed
loop (z, ξ
1
)subsystem. Over a small open neighborhood of the origin U, ﬁx
1
> 0
such that for all 0 < ≤
1
(i) holds. Let ξ
2
= α
2
(z, ξ
1
, ) be the unique ﬁxed point
of (7.106) over U. Around (z, ξ
1
) = 0, f
0
and g
0
can be expressed as
f
0
(z, α
1
(z)) = A
0
z +
˜
f
0
(z)
g
0
(z, ξ
1
, ξ
2
, ) = ˜ g
0
(z, ξ
1
, ξ
2
, )ξ
2
where
A
0
=
¸
∂f
0
∂z
(z, α
1
(z)) +
∂f
0
∂ξ
1
(z, α
1
(z))
∂α
1
(z)
∂z
z=0
is a Hurwitz matrix by assumption and
˜
f
0
with all its ﬁrst order partial derivatives
vanish at z = 0. Consider the quadratic Lyapunov function
V
0
(z) =
1
2
z
T
Pz
where P is the solution of the following Lyapunov equation
A
T
0
P + PA
0
= −I
Applying the change of coordinates µ
1
= ξ
1
− α
1
(z), we show that the candidate
Lyapunov function
W
0
(z, µ
1
) = V
0
(z) +
1
2
µ
2
1
295
locally over U is a valid Lyapunov function for a suﬃciently large c
1
= σ
1
(0) > 0.
Note that locally σ
1
(µ
1
) = c
1
µ
1
+ O([µ
1
[
2
). Calculating
˙
W
0
, we have
˙
W
0
= −[z[
2
+z
T
P
˜
f
0
(z) + µ
1
z
T
Ph
0
(z, α
1
(z), µ
1
)
+ z
T
P˜ g
0
(z, α
1
(z) + µ
1
, α
2
) α
2
(z, α
1
(z) +µ
1
, )
− c
1
µ
2
1
+ O([µ
1
[
3
) + µ
1
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)
We need to determine a local approximation of each term over U in the preceding
expression. The second term z
T
P
˜
f
0
(z) ∼ O([z[
3
), and the third term satisﬁes
[µ
1
z
T
Ph
0
(z, α
1
(z), µ
1
)[ < l
1
[µ
1
[ [z[
for some l
1
> 0. Also, α
2
vanishes at the origin, thus [α
2
(z, α
1
(z) + µ
1
, )[ <
˜
l
2
[z[ +
˜
l
3
[µ
1
[. This means that
[z
T
P˜ g
0
(z, α
1
(z) + µ
1
, α
2
) α
2
(z, α
1
(z) +µ
1
, )[ < l
2
[z[
2
+ l
3
[µ
1
[ [z[
for some l
2
, l
3
> 0. Also, because ∂V (z)/∂z vanishes at z = 0, there exists
˜
l
4
> 0
such that
[
∂V (z)
∂z
[ ≤
˜
l
4
[z[
Thus
[µ
1
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
)[ < l
4
[µ
1
[ [z[
for some l
4
> 0. We obtain
˙
W
0
≤ −[z[
2
(1 −l
2
) + (l
3
+ l
4
)[µ
1
[ [z[ −c
1
[µ
1
[
2
+ O([z[
3
) + O([µ
1
[
3
)
Take
2
= min¦1/3l
2
,
1
¦ and a suﬃciently small open neighborhood U
0
⊂ U of the
origin in the form U
0
= U
0
(c
0
) = ¦(z, µ
1
)[W
0
(z, µ
1
) < c
0
¦ for a suﬃciently small
c
0
> 0 such that
O([z[
3
) <
1
3
[z[
2
, O([µ
1
[
3
) < b
1
[µ
1
[
2
for some b
1
> 0 (e.g. b
1
= 0.1). Then, for all 0 < ≤
2
, we have
˙
W
0
≤ −
1
3
[z[
2
+ (
l
3
3l
2
+l
4
)[µ
1
[ [z[ −(c
1
−b
1
)[µ
1
[
2
by completing the squares
˙
W
0
≤ −
1
3
([z[ +
3
2
b
2
[µ
1
[)
2
−(c
1
−b
1
−
3
4
b
2
2
)[µ
1
[
2
296
where b
2
= (l
3
/3l
2
) + l
4
. Therefore, by taking
c
1
> c
∗
1
:= b
1
+
3
4
b
2
2
> 0
there exists a positive deﬁnite matrix Q such that
˙
W
0
< −¯ z
T
Q¯ z < 0 , ∀¯ z = 0
where ¯ z = col(z, µ
1
) and thus the origin (z, µ
1
) = 0 is locally exponentially sta
ble. Moreover, U
0
is forward invariant in time. To establish semiglobal asymp
totic stabilization of (z, ξ
1
) = 0, consider the solutions starting in a compact set
/
0
:= /
0
z
/
0
µ
1
÷ (z(0), µ
1
(0)) that includes U
0
. Take the Lyapunov function
W(z, µ
1
) = V (z) +
1
2
µ
2
1
. Let Ω
0
= ¦(z, µ
1
)[W(z, µ
1
) ≤ c
0
¦ for a suﬃciently large
c
0
> 0 such that Ω
0
⊃ /
0
. Fix
3
≤
2
such that for all 0 < <
3
(i) holds over Ω
0
(note that W and thus Ω
0
are independent of ). Calculating
˙
W along the solutions
of the closed loop system, we get
˙
W =
∂V (z)
∂z
f
0
(z, α
1
(z)) +
∂V (z)
∂z
g
0
(z, ξ
1
, ξ
2
, )
+ µ
1
(ξ
2
−
∂α
1
(z)
∂z
f
0
(z, ξ
1
) −
∂α
1
(z)
∂z
g
0
(z, ξ
1
, ξ
2
, )
+
∂V (z)
∂z
h
0
(z, α
1
(z), µ
1
))
=
∂V (z)
∂z
f
0
(z, α
1
(z)) −µ
1
σ(µ
1
) +
∂V (z)
∂z
g
0
(z, ξ
1
, ξ
2
, )
=
∂V (z)
∂z
f
0
(z, α
1
(z)) −µ
1
σ(µ
1
) +
∂V (z)
∂z
g
0
(z, α
1
(z) + µ
1
, α
2
(z, α
1
(z) + µ
1
, ), )
Noting that Ω
c
0
:= Ω
0
` U
0
is a compact set and deﬁning
0
= min
z∈Ω
c
0
−
∂V (z)
∂z
f
0
(z, α
1
(z)) +µ
1
σ(µ
1
) > 0
there exists an
∗
> 0 such that for all 0 < <
∗
, [
∂V (z)
∂z
g
0
[ <
0
and
˙
W < 0 over
Ω
c
0
. Thus, any solution of the system enters U
0
after some ﬁnite time and exponen
tially converges to the origin. This guarantees semiglobal asymptotic stabilization
of (z, µ
1
) = 0 and therefore semiglobal asymptotic stabilization of (z, ξ
1
) = 0. It
remains to prove (iii). This follows from standard backstepping theorem applied to
the composite system (7.101) rewritten in the following form
˙
¯ z = f
2
(¯ z, ξ
2
, )
˙
ξ
2
= u
where z = col(z, ξ
1
) and f
2
= col(f, ξ
2
). Noting that based on (ii), there exists a
semiglobally asymptotically stabilizing feedback law ξ
2
= α
2
(¯ z, ) for the ¯ zsubsystem.
297
Then, given the control law (7.107), similar to the argument in the proof of (ii), it
can be shown that there exists a small neighborhood of the origin
˜
U
0
such that the
solution of the closed loop system is exponentially stable and the positive deﬁnite
proper function
Q(¯ z, ξ
2
, ) = W(¯ z) +
1
2
(ξ
2
−α
2
(¯ z, ))
2
is a valid Lyapunov function for the whole system with the property that
˙
Q < 0 over
˜
Ω `
˜
U
0
where
˜
Ω is an appropriate invariant set of the Lyapunov function Q. This
ﬁnishes the proof.
298
Bibliography
[1] R. Abraham and J. E. Marsden. Foundations of Mechanics. AddisonWesley,
1978.
[2] F. Albertini and E. D. Sontag. “Discretetime transitivity and accessibility:
Analytic systems”. SIAM J. Control and Opt., 31:1599–1622, 1993.
[3] F. Albertini and E. D. Sontag. “State observability in recurrent neural net
works”. Systems & Control Letters, 22:235–244, 1994.
[4] A. Astolﬁ. “Discontinuous control of nonholonomic systems”. Systems & Con
trol Letters, 27:37–45, 1996.
[5] A. Astolﬁ. “On the stabilization of the nonholonomic systems”. The 33rd Conf.
on Decision and Control, pages 3481–3484, Orlando, FL, 1994.
[6] K. J. Astrom and K. Furuta. “Swining up a pendulum by energy control”.
IFAC, San Francisco, 1996.
[7] B. E. Bishop and M. W. Spong. “Control of redundant manipulators using
logicbased switching”. Proc. IEEE Conf. on Decision and Control, Tampa,
FL, Dec., 1998.
[8] A. M. Bloch and P. E. Crouch. “Nonholonomic control systems on Riemannian
manifolds”. SIAM J. Control and Optimization, 33(1):126–148, January 1995.
[9] A. M. Bloch, P. S. Krishnaprasad, J. E. Marsden, and R. M. Murray. “Nonholo
nomic mechanical systems with symmetry”. Archives of Rational Mech. An.,
136:21–99, 1996.
[10] A. M. Bloch, N. E. Leonard, and J.E. Marsden. “Stabilization of the pendulum
on a rotor arm by the method of controlled lagrangians”. Proceedings of the
IEEE International Conference on Robotics and Automation, May 1999.
[11] A. M. Bloch, M. Reyhanoglu, and N. H. McClamroch. “Control and stabiliza
tion of nonholonomic dynamic systems”. IEEE Trans. on Automatic Control,
37:1746–1757, 1992.
[12] W. J. Book. “Controlled motion in an elastic world”. ASME Journal of Dynamic
Systems, Measurement and Control, 115(2B):252–261, 1993.
299
[13] S. Bortoﬀ and M. W. Spong. “Pseudolinearization of the acrobot using spline
functions”. Proc. IEEE Conf. on Decision and Control, pages 593–598, Tuscan,
AZ, 1992.
[14] S. A. Bortoﬀ. “Pseudolinearization Using Spline Functions with Application to
the Acrobot”. PhD thesis, University of Illionois at UrbanaChampain, Dept. of
Electrical and Computer Engineering, 1992.
[15] R. W. Brockett. “Asymptotic stability and feedback stabilization”. In R. W.
Brockett, R. S. Millman, and H. J. Sussmann, editors, Diﬀerential Geometric
Control Theorey, pages 181–191, Birkhauser, Boston, MA, 1983.
[16] F. Bullo. “Nonlinear Control of Mechanical Systems: A Riemannian Geome
try Approach”. PhD thesis, California Institute of Technology, Department of
Control and dynamical Systems, 1998.
[17] C. I. Byrnes and A. Isidori. “On the attitude stabilization of a rigid spacecraft”.
Automatica, 27(1):87–95, 1991.
[18] R. H. Cannon and E. Schmitz. “Initial experiments on the endpoint control of
a ﬂexible onelink robot”. Int. Journal of Robotics Research, 3(3):62–75, 1984.
[19] C. Canudas de Wit and O. J. Sordalen. “Exponential stabilization of mobile
robots with nonholonomic constraints”. The 30th Conf. on Decision and Con
trol, pages 692–697, 1991.
[20] P. A. Chudavarapu and M. W. Spong. “On noncollacated control of a single
ﬂexible link”. IEEE Int. Conf. on Robotics and Automation, MN, April, 1996.
[21] C. C. Chung and J. Hauser. “Nonlinear control of a wwinging pendulum”.
Automatica, 40:851–862, 1995.
[22] L. Crawford and S. Sastry. “Bilogical motor control approaches for a planar
diver”. Proc. IEEE Conf. on Decision and Control, 4:3881–3886, 1995.
[23] M. Dahleh and I. DiazBobillo. Control of Uncertain Systems: A Linear Pro
gramming Approach. PrenticeHall, 1995.
[24] A. De Luca. “Trajectory control of ﬂexible manipulators”. In B. Siciliano
and K. P. Valavanis, editor, Control Problems in Robotics and Automation,
SpringerVerlag, London, UK, 1997.
[25] A. De Luca and B. Siciliano. “Trajectory control of a nonlinear onelink ﬂexible
arm”. Int. Journal of Control, 50(5):1699–1715, 1989.
[26] M. P. Do Carmo. Diﬀerential Geometry of Curves and Surfaces. PrenticeHall,
1976.
300
[27] O. Egeland, M. Dalsmo, and O. J. Sordalen. “Feedback control of an nonholo
nomic underwater vehicle with constant desired conﬁguration”. Int. Journal of
Robotics Research, 15:24–35, 1996.
[28] I. Fantoni, R. Lozano, and M. Spong. “Passivity based control of the pendubot”.
To appear in IEEE Trans. on Automatic Control, 2000.
[29] M. Fliess, J. Levine, P. Martin, and P. Rouchon. “Flatness and defect of non
linear systems: Introductory theory and examples.”. Int. Journal of Control,
61(6):1327–1361, 1995.
[30] T. Fossen. Guidance and Control of Ocean Vehicles. John Willey & Sons Ltd.,
1994.
[31] E. Frazzoli, M. A. M.A. Dahleh, and E. Feron. “Trajectory tracking control
design for autonomous helicopters using a backstepping algorithm”. Proc. of
American Control Conference, pp. 4102–4107, Chicago, IL, 2000.
[32] J.M. Godhavn, A. Balluchi, L. Crawford, and S. Sastry. “Path planning for
nonholonomic systems with drift”. Proc. IEEE American Control Conference,
pages 532–536, July, 1997.
[33] Y.L. Gu. “A direct adaptive control scheme for underactuated dynamic sys
tems”. Proc. IEEE Conf. on Decision and Control, pages 1625–1627, San An
tonio, TX, 1993.
[34] J. Hauser, S. Sastry, and P. Kokotovi´c. “Nonlinear control via approximate
inputoutput linearization”. IEEE Trans. on Automatic Control, vol.37:pp.
392–398, 1992.
[35] J. Hauser, S. Sastry, and G. Meyer. “Nonlinear control design for slightly non
minimum phase systems”. Automatica, vol. 28:pp.665–679, 1992.
[36] A. Isidori. Nonlinear Control Systems. Springer, 1995.
[37] M. Jankovic, D. Fontaine, and P. V. Kokotovic. “TORA example: Cascade and
passivitybased control designs”. IEEE Trans. on Control Systems Technology,
4(3):292–297, 1996.
[38] Z. P. Jiang, D. J. Hill, and Y. Guo. “Stabilization and tracking via output
feedback for a nonlinear benchmark system”. Automatica, 34(7):907–915, 1998.
[39] Z. P. Jiang and I. Kanellakopoulos. “Global outputfeedback tracking for a
benchmark nonlinear system”. IEEE Trans. on Automatic Control, 45(5):1023–
1027, 2000.
[40] H. K. Khalil. Nonlinear Systems. PrenticeHall, Inc., 1996.
301
[41] I. Kolmanovsky and N. H. McClamroch. “Hybrid feedback stabilization of
rotational translational actuator (RTAC) system”. Int. Journal of Robust and
Nonlinear Control, 8:367–375, 1998.
[42] T. J. Koo and S. Sastry. “Diﬀerential ﬂatness based full authority helicopter
control design”. Proc. of the 38th IEEE Conf. on Decision and Control, pp.
1982–1987, Phoenix, AZ, 1999.
[43] T. J. Koo and S. Sastry. “Output tracking control design of a helicopter model
based on approximate linearization”. Proc. of the 37th IEEE Conf. on Decision
and Control, pp. 3635–3640, Tampa, FL, 1998.
[44] P. S. Krishnaprasad. “Eulerian manybody problems”. In J. E. Marsden and
P. S. Krishnaprasad and J. Simo, editor, Dynamics and Control of Multibody
Systems, vol. 97, Providence, RI:AMS, 1989.
[45] P. S. Krishnaprasad. “Geometric phases and optimal reconﬁguration for multi
body systems”. Proc. of American Control Conference, pp. 2440–2444, San
Diego, CA, 1990.
[46] P. S. Krishnaprasad and R. Yang. “Geometric phases, anholonomy, and optimal
movement”. Proc. of the Int. Conf. on Robotics and Automation, pp. 2185–2189,
Sacramento, CA, 1991.
[47] M.C. Laiou and A. Astolﬁ. “Quasismooth control of chained systems”. Proc.
of the American Control Conference, San Diego, CA, June 1999.
[48] A. D. Lewis and R. M. Murray. “Variational principles for constrained systems:
theory and experiment”. Int. J. of Nonlinear Mechanics, 30(6):793–815, 1995.
[49] A. D. Lewis and R. M. Murray. “Conﬁguration controllability of simple me
chanical control systems”. SIAM J. Control and Optimization, 35(3):766–790,
May 1997.
[50] A. D. Lewis, J. P. Ostrowski, R. M. Murray, and J. W. Burdick. “Noholonomic
mechanics and locomotion: the snakeboard example”. Proc. of the IEEE Int.
Conf. on Robotics and Automation, pages 2391–2400, San Diego, 1994.
[51] Z. Li and R. Montgomery. “Dynamics and optimal control of a legged robot in
ﬂight phase”. Proc. of the Int. Conf. on Robotics and Automation, pp. 1816–
1821, Cincinnati, OH, 1990.
[52] D. Liberzon and S. Morse. “Basic problems in stability and design of switched
systems”. IEEE Control System Magazine, 19(5):59–70, 1999.
[53] J. E. Marsden, P. S. Krishnaprasad, and J. Simo. Dynamics and Control of
Multibody Systems. vol. 97, Providence, RI: AMS, 1989.
302
[54] J. E. Marsden and T. S. Ratiu. Introduction to Mechanics and Symmetry.
Springer, 1999.
[55] P. Martin, S. Devasia, and B. Paden. “A diﬀerent look at output tracking:
control of a VTOL aircraft”. Proceedings of 33rd IEEE Conference on Decision
and Control, pages 2376–2381, 1994.
[56] F. Mazenc and L. Praly. “Adding integrations, saturated controls, and stabi
lization for feedforward Systems”. IEEE Trans. on Automatic Control, 40:1559–
1578, 1996.
[57] M. Krsti´c, I. Kanellakopoulos, and P. Kokotovi´c. Nonlinear and Adaptive Con
trol Design. John Wiley & Sons, 1995.
[58] N. H. McClamroch and I. Kolmakovsky. “A hybrid switched mode control for
V/STOL ﬂight control problems”. Proc. IEEE conf. on Decision and Control,
pages 2648–2653, Kobe, Japan, 1996.
[59] R. T. M’Closkey and R. M. Murray. “Exponential stabilization of driftless
nonlinear control systems using homogeneous feedback”. IEEE Trans. on Au
tomatic Control, pages 614–628, May 1997.
[60] R. T. M’Closkey and R. M. Murray. “Exponential stabilization of driftless
nonlinear control systems via timevarying homogeneous feedback”. Proc. IEEE
conf. on Decision and Control, pages 943–948, San Antonio, TX, 1993.
[61] A. Megretski and A. Rantzer. “System analysis via Integral Quadratic Con
straints”. IEEE Trans. on Automatic Control, 46(6):819–830, 1997.
[62] P. Morin and C. Samson. “Timevarying exponential stabilization of the at
titude of a rigid spacecraft with two controls”. IEEE Trans. on Automatic
Control, 42:528–533, 1997.
[63] R. Murray and S. Sastry. “Nonholonomic motion planning:steering using sinu
soids”. IEEE Trans. on Automatic Control, 38(5):700–716, May 1993.
[64] J. I. Neimark and F. A. Fufaev. Dynamics of nonholonomic systems, volume 33.
AMS, 1972.
[65] H. Nijmeijer and A. van der Schaft. Nonlinear Dynamical Control Systems.
Springer Verlag, 1990.
[66] R. OlfatiSaber. “Normal forms for underactuated mechanical systems with
symmetry”. submitted to IEEE Trans. on Automatic Control, Dec. 5, 1999.
[67] R. OlfatiSaber. “Fixed point controllers and stabilization of the cartpole sys
tem and the rotating pendulum”. Proc. of the 38th Conf. on Decision and
Control, pp. 1174–1181, Phoenix, AZ, Dec. 1999.
303
[68] R. OlfatiSaber. “Control of underactuated mechanical systems with two de
grees of freedom and symmetry”. Proc. of American Control Conference, pp.
4092–4096, Chicago, IL, June 2000.
[69] R. OlfatiSaber. “Cascade normal forms for underactuated mechanical sys
tems”. Proc. of the 39th Conf. on Decision and Control, Sydney, Australia,
Dec. 2000.
[70] R. OlfatiSaber. “Global conﬁguration stabilization for the VTOL aircraft with
strong input coupling”. Proc. of the 39th Conf. on Decision and Control, Syd
ney, Australia, Dec. 2000.
[71] R. OlfatiSaber. “Trajectory tracking for a ﬂexible onelink robot using a non
linear noncollocated output”. Proc. of the 39th Conf. on Decision and Control,
Sydney, Australia, Dec. 2000.
[72] R. OlfatiSaber and A. Megretski. “Controller design for the beamandball
System”. Proc. of the 37th Conf. on Decision and Control, pp. 4555–4560,
Tampa, FL, Dec. 1998.
[73] J. P. Ostrowski and J. W. Burdick. “Controllability tests for mechanical sys
tems with constraints and symmetry”. Journal of Applied Mathematics and
Computer Science, 7(2):101–127, 1997.
[74] J. P. Ostrowski and J. W. Burdick. “The mechanics and control of undulatory
locomotion”. Int. J. of Robtics Research, 177(7):683–701, 1998.
[75] K. Y. Pettersen and O. Egeland. “Exponential stabilization of an underactuated
surface vessel”. Proc. IEEE Conf. on Deision and Control, pages 1682–1687,
1996.
[76] J.P. Pomet. “Explicit design of timevarying stabilizing control laws for a class
of controllable systems without drift”. Systems & Control Letters, 18:147–158,
1992.
[77] M. Reyhanoglu, S. Cho, N. H. McClamroch, and I. Kolmanovsky. “Discon
tinuous feedback control of a planar rigid body with an unactuated internal
degree of freedom”. Proc. IEEE Conf. on Decision and Control, pages 433–438,
Tampa, FL, 1998.
[78] M. Reyhanoglu, A. van der Schaft, N. H. McClamroch, and I. Kolmanovsky.
“Dynamics and control of a class of underactuated mechanical systems”. IEEE
Trans. on Automatic Control, 44(9):1663–1671, 1999.
[79] R. Sepulchre. “Slow peaking and lowgain designs for global stabilization of
nonlinear systems”. submitted for IEEE TAC, 1999.
[80] R. Sepulchre, M. Jankovi´c, and P. Kokotovi´c. Constructive Nonlinear Control.
SpringerVerlag, 1997.
304
[81] B. Siciliano and W. J. Book. “A singular perturbation approach to control of
lightweight ﬂexible manipolators”. Int. Journal of Robotics Research, 7(4):79–
90, 1988.
[82] E. D. Sontag. “Smooth stabilization implies coprime factorization”. IEEE
Trans. on Automatic Control, 34:1376–1378, 1989.
[83] E. D. Sontag. “Remarks on stabilization and inputtostate stability”. Proceed
ings of the 28th Conference on Decision and Control, pages 1376–1378, Tampa,
FL, Dec. 1989.
[84] E. D. Sontag and Y. Wang. “On Characterizations of the inputtostate stability
property”. Systems & Control Letters, vol.24:pp.351–359, 1995.
[85] V. A. Spector and H. Flashner. “Modeling and design implications of non
collocated control in ﬂexible systems”. ASME Journal of Dynamic Systems,
Measurement and Control, 112:186–193, 1990.
[86] M. W. Spong. “The swing up control problem for the acrobot”. IEEE Control
Systems Magazine, 15:49–55, 1995.
[87] M. W. Spong. “Underactuated mechanical systems”. In B. Siciliano and K.
P. Valavanis, editor, Control Problems in Robotics and Automation, Springer
Verlag, London, UK, 1997.
[88] M. W. Spong. “Applications of switching control in robot locomotion”. Proc.
Workshop on Intelligent Control in Robotics and Automation, IFAC World
Congress, Beijing, China, July, 1999.
[89] M. W. Spong. “Passivitybased control of the compass gait biped”. IFAC World
Congress, Beijing, China, July, 1999.
[90] M. W. Spong. “Energy based control of a class of underactuated mechanical
systems”. 1996 IFAC World Congress, July 1996.
[91] M. W. Spong. “Bipedal locomotion, robot gymnastics, and robot air hockey: a
rapprochement”. SuperMechano Systems (SMS’99) Workshop, Tokyo, Japan,
February, 1999.
[92] M. W. Spong and D. J. Block. “The Pendubot: a mechatronic system for
control research and education”. Procedeings of the 34th IEEE Conference on
Decision and Control, pages 555–556, New Orleans, Dec. 1995.
[93] M. W. Spong, P. Corke, and R. Lozano. “Nonlinear control of the Inertia Wheel
Pendulum”. Automatica, submitted, September, 1999.
[94] M. W. Spong and L. Praly. “Control of underactuated mechanical systems using
switching and saturation”. Proc. of the Block Island Workshop on Control Using
Logic Based switching, 1996.
305
[95] M. W. Spong and M. Vidyasagar. Robot Dynamics and Control. John Wiley &
Sons, 1989.
[96] C.Y. Su and Y. Stepanenko. “Sliding mode control of nonholonomic systems:
Underactuated manipulator case”. Proc. IFAC Nonlinear Control Systems De
sign, pages 609–613, Tahoe City, CA, 1995.
[97] H. J. Sussmann. “A general theorem on local controllability”. SIAM Journal
of Control and Optimization, 25(1):158–194, 1987.
[98] H. J. Sussmann. “Limitations on stabilizability of globally minimum phase
systems”. IEEE Trans. on Automatic Control, vol.35:pp.117–119, 1990.
[99] H. J. Sussmann and P. Kokotovi´c. “The peaking phenomenon and the global
stabilization of nonlinear systems”. IEEE Trans. on Automatic Control,
vol.39:pp.2411–2425, 1991.
[100] H. J. Sussmann and Y. Yang. “On the stablizability of multiple integrators by
means of bounded feedback controls”. Proceedings of the 30th Conference on
Decision and Control, Bringhton, England, Dec 1991.
[101] A. R. Teel. “A nonlinear small gain theorem for the analysis of control systems
with saturation”. IEEE Trans. on Automatic Control, 40:1256–1270, 1996.
[102] A. R. Teel. “Using Saturation to stabilize a class of singleinput partially linear
composite systems”. IFAC NOLCOS’92 Symposium, pages 369–374, June 1992.
[103] A. R. Teel and L Praly. “Tools for semiglobal stabilization by partial state and
output feedback”. SIAM Journal of Control and Optimization, 33:1443–1488,
1995.
[104] D. Tilbury, R. Murray, and S. Sastry. “Trajectory generation for the Ntrailer
problem using Goursat normal forms”. IEEE Trans. on Automatic Control,
40(5):802–819, May 1995.
[105] M. van Nieuwstadt and R. Murray. “Realtime trajectory generation for diﬀer
entially ﬂat systems”. Int. Journal of Robust and Nonlinear Control, 8(11):995–
1020, 1998.
[106] M. van Nieuwstadt and R. Murray. “Approximate trajectory generation for
diﬀerentially ﬂat systems with zero dynamics”. Proc. of the 34th Conf. on
Decision and Control, New Orleans, LA, Dec. 1995.
[107] M. Vidyasagar and B. D. O. Anderson. “Approximation and stabilization of dis
tributed Systems by lumped Systems”. Systems and Control Letters, 12(2):95–
101, 1989.
[108] G. C. Walsh, R. Montgomery, and S. Sastry. “Orientation control of the dynamic
satellite”. Proc. American Control Conference, pages 138–142, 1994.
306
[109] C.J. Wan, D. S. Bernstein, and V. T. Coppola. “Global stabilization of the
oscillating eccenteric rotor”. Nonlinear Dynamics, 10:49–62, 1996.
[110] W. J. Wang, S. S. Lu, and C. F. Hsu. “Experiments on the position control
of a onelink ﬂexible robot arm”. IEEE Trans. on Robotics and Automation,
5(3):373–377, 1989.
[111] K. Y. Wichlund, O. J. Sordalen, and O. Egeland. “Control of vehicles with
secondorder nonholonomic constraints: Underactuated Vehicles”. European
Control Conf., pages 3086–3091, Rome, Italy, 1995.
[112] M. Wiklund, A. Kristenson, and K. J. Astrom. “A new strategy for swinging
up an inverted pendulum”. Proceedings of IFAC Symposium, pages 757–760,
Sydney, Australia, 1993.
307
Nonlinear Control of Underactuated Mechanical Systems with Application to Robotics and Aerospace Vehicles by Reza OlfatiSaber
Submitted to the Department of Electrical Engineering and Computer Science on January 15, 2000, in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy in Electrical Engineering and Computer Science
Abstract
This thesis is devoted to nonlinear control, reduction, and classiﬁcation of underactuated mechanical systems. Underactuated systems are mechanical control systems with fewer controls than the number of conﬁguration variables. Control of underactuated systems is currently an active ﬁeld of research due to their broad applications in Robotics, Aerospace Vehicles, and Marine Vehicles. The examples of underactuated systems include ﬂexiblelink robots, mobile robots, walking robots, robots on mobile platforms, cars, locomotive systems, snaketype and swimming robots, acrobatic robots, aircraft, spacecraft, helicopters, satellites, surface vessels, and underwater vehicles. Based on recent surveys, control of general underactuated systems is a major open problem. Almost all reallife mechanical systems possess kinetic symmetry properties, i.e. their kinetic energy does not depend on a subset of conﬁguration variables called external variables. In this work, I exploit such symmetry properties as a means of reducing the complexity of control design for underactuated systems. As a result, reduction and nonlinear control of highorder underactuated systems with kinetic symmetry is the main focus of this thesis. By “reduction”, we mean a procedure to reduce control design for the original underactuated system to control of a lowerorder nonlinear or mechanical system. One way to achieve such a reduction is by transforming an underactuated system to a cascade nonlinear system with structural properties. If all underactuated systems in a class can be transformed into a speciﬁc class of nonlinear systems, we refer to the transformed systems as the “normal form” of the corresponding class of underactuated systems. Our main contribution is to ﬁnd explicit change of coordinates and control that transform several classes of underactuated systems, which appear in robotics and aerospace applications, into cascade nonlinear systems with structural properties that are convenient for control design purposes. The obtained cascade normal forms are three classes of nonlinear systems, namely, systems in strict feedback form, feedforward form, and nontriangular linearquadratic form. The names of these three classes are due to the particular lowertriangular, uppertriangular, and nontriangular structure in which the state variables appear in the dynamics of the corresponding nonlinear systems. The triangular normal forms of underactuated systems can be controlled using existing backstepping and feedforwarding procedures. However, control of the 2
nontriangular normal forms is a major open problem. We address this problem for important classes of nontriangular systems of interest by introducing a new stabilization method based on the solutions of ﬁxedpoint equations as stabilizing nonlinear state feedback laws. This controller is obtained via a simple recursive method that is convenient for implementation. For special classes of nontriangular nonlinear systems, such ﬁxedpoint equations can be solved explicitly. As a result of the reduction process, one obtains a reduced nonlinear subsystem in cascade with a linear subsystem. For many classes of underactuated systems, this reduced nonlinear subsystem is physically meaningful. In fact, the reduced nonlinear subsystem is itself a Lagrangian system with a welldeﬁned lowerorder conﬁguration vector. In special cases, this allows construction of Hamiltoniantype Lyapunov functions for the nonlinear subsystem. Such Lyapunov functions can be then used for robustness analysis of normal forms of underactuated systems with perturbations. The Lagrangian of the reduced nonlinear subsystem is parameterized by the shape variables (i.e. the complement set of external variables). It turns out that “a control law that changes the shape variables” to achieve stabilization of the reduced nonlinear subsystem, plays the most fundamental role in control of underactuated systems. The key analytical tools that allow reduction of highorder underactuated systems using transformations in explicit forms are “normalized generalized momentums and their integrals” (whenever integrable). Both of them can be obtained from the Lagrangian of the system. The diﬃculty is that many reallife and benchmark examples do not possess integrable normalized momentums. For this reason, we introduce a new procedure called “momentum decomposition” which uniquely represents a nonintegrable momentum as a sum of an integrable momentum term and a nonintegrable momentumerror term. After this decomposition, the reduction methods for the integrable cases can be applied. The normal forms for underactuated systems with nonintegrable momentums are perturbed versions of the normal forms for the integrable cases. This perturbation is the timederivative of the momentumerror and only appears in the equation of the momentum of the reduced nonlinear subsystem. Based on some basic properties of underactuated systems as actuation/passivity of shape variables, integrability/nonintegrability of appropriate normalized momentums, and presence/lack of input coupling; I managed to classify underactuated systems to 8 classes. Examples of these 8 classes cover almost all major applications in robotics, aerospace systems, and benchmark systems. In all cases, either new control design methods for open problems are invented, or signiﬁcant improvements are achieved in terms of the performance of control design compared to the available methods. Some of the applications of our theoretical results are as the following: i) trajectory tracking for ﬂexiblelink robots, ii) (almost) global exponential tracking of feasible trajectories for an autonomous helicopter, iii) global attitude and position stabilization for the VTOL aircraft with strong input coupling, iv) automatic calculation of diﬀerentially ﬂat outputs for the VTOL aircraft, v) reduction of the stabilization of a multilink planar robot underactuated by one to the stabilization of the Acrobot, or the Pendubot, vi) semiglobal stabilization of the Rotating Pendulum, the BeamandBall system, using ﬁxedpoint state feedback, vii) global stabilization of the 2D and 3D CartPole systems to an equilibrium point, viii) global asymptotic 3
stabilization of the Acrobot and the InertiaWheel Pendulum. For underactuated systems with nonholonomic velocity constrains and symmetry, we obtained normal forms as the cascade of the constraint equation and a reducedorder Lagrangian control system which is underactuated or fullyactuated. This depends on whether the sum of the number of constraints and controls is less than, or equal to the number of conﬁguration variables. This result allows reduction of a complex locomotive system called the snakeboard. Another result is global exponential stabilization of a twowheeled mobile robot to an equilibrium point which is far from the origin ( 1), using a smooth dynamic state feedback. Thesis Supervisor: Alexandre Megretski Title: Associate Professor of Electrical Engineering
4
Acknowledgments
First and foremost, I would like to thank my advisor Alexandre Megretski. Alex’s unique view of control theory and his mathematical discipline taught me to see nonlinear control from a new perspective that enhanced my understanding of what constitutes an acceptable solution for certain fundamental problems in control theory. I can never thank him enough for giving me the freedom to choose the theoretical ﬁelds of my interest and helping me to pursue my goals. I have always valued his advice greatly. I would like to thank Munther Dahleh for our exciting discussions during the group seminars and for giving me many useful comments and advice on my work. I wish to give my very special thanks to both George Verghese and Bernard Lesieutre who have been a constant source of support and encouragement for me from the beginning of my studies at MIT. The past years at MIT were a wonderful opportunity for me to get engaged in interesting discussions with several MIT professors and scientists. Sanjoy Mitter has always amazed me with his brilliant view of systems and control theory. I am very grateful for his invaluable advice on many important occasions. I particularly would like to thank Eric Feron for his encouragement and for enthusiastically showing me simulation results and test models related to my work. I wish to thank Dimitri Bertsekas for being a great teacher and giving me important advice as my academic advisor during the past years. Many thanks goes to Nicola Elia and Ulf J¨nsson for our long hours of discussions o and their valuable technical comments. I would also like to thank Emilio Frazzoli for many discussions on helicopters and nonlinear control. I would like to take this opportunity to thank all my friends at LIDS for their encouragement, support, and the fun times we shared during our group seminars. I owe my sincere gratitude to Milena Levak, the Dean of the International Students Oﬃce at MIT, who has greatly helped me during very important situations and times. I am incredibly grateful to my family. My parents, Hassan and Soroor, and my sister, Negin, gave me their unconditional love and support during all these years. While I never had the opportunity to see my family during my ﬁve and a half years of studies at MIT, my heart has always been with them. They never stopped believing in me and did everything they could to help me achieve my goals. At last, my most special thanks goes to Manijeh. Her beautiful smile, emotional support, and strong encouragement were the driving force of my eﬀorts. With her creative mind, she gave me many examples of underactuated systems in nature that I had not thought of myself. Without her help and enthusiasm, this work certainly would not have been possible.
5
Contents
1 Introduction 1.1 Mechanics and Control Theory . . . . . . . . . . . . . . . . . . . . . 1.2 Nonlinear Control Systems . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Underactuated Systems . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Highly Nonlinear Systems . . . . . . . . . . . . . . . . . . . . 1.2.3 Cascade Nonlinear Systems . . . . . . . . . . . . . . . . . . . 1.3 Statement of the Contributions . . . . . . . . . . . . . . . . . . . . . 1.3.1 Normal Forms for Underactuated Systems . . . . . . . . . . . 1.3.2 Reduction and Control of LowOrder Underactuated Systems . 1.3.3 Reduction and Control of HighOrder Underactuated Systems 1.3.4 Reduction and Control of Underactuated Systems with Nonholonomic Velocity Constraints . . . . . . . . . . . . . . . . . 1.3.5 Applications in Robotics and Aerospace Vehicles . . . . . . . . 1.3.6 Control of Nontriangular Cascade Nonlinear Systems . . . . . 1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Mechanical Control Systems 2.1 Introduction . . . . . . . . . . . . . . . . . 2.2 Simple Lagrangian Systems . . . . . . . . 2.3 Fullyactuated Mechanical Systems . . . . 2.4 Underactuated Mechanical Systems . . . . 2.5 Flat Mechanical Systems . . . . . . . . . . 2.6 Nonholonomic Mechanical Systems . . . . 2.7 Symmetry in Mechanics . . . . . . . . . . 2.8 Normalized Momentums and Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 14 15 15 17 19 20 22 23 24 26 27 27 27 28 28 28 30 31 32 32 34 35 37 37 39 39 40 41 42 42 43
3 Normal Forms for Underactuated Systems 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Dynamics of Underactuated Systems . . . . . . . . . . . . 3.3 Examples of Underactuated Systems . . . . . . . . . . . . 3.3.1 The Acrobot and the Pendubot . . . . . . . . . . . 3.3.2 The CartPole System and the Rotating Pendulum 3.3.3 The BeamandBall System . . . . . . . . . . . . . 3.3.4 The TORA System . . . . . . . . . . . . . . . . . . 3.3.5 The InertiaWheel Pendulum . . . . . . . . . . . . 6
.5 3. . . . . . . . . . . . . . . . . . . . . . .9 3. . .4 3. . . . . . . . . . . . 5. . . . . . . . . . . 5. . . . . . 106 5 Applications to Robotics and Aerospace Vehicles 5. 5. .1 Underactuated Systems with Actuated Shape Variables . . . . . . . . . . . . . . . . . .11. . . . . . . . 5. .6 The Pendubot . . . . . . . . . . . . . .3 Tracking Control . . . 5. .4 Feedback Linearization of the Unperturbed Model of the Helicopter . . . 5. . . 5. . .11. 60 4.7 The BeamandBall System . . . . . . Noncollocated Partial Feedback Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . .4 The CartPole System . . . . 5. .1 Aggressive Swingup of the Rotating Pendulum . . . . Classes of Structured Nonlinear Systems . .1 Shape Variables and Kinetic Symmetry . . . Partial Feedback Linearization Under Input Coupling . . . . . . . 5. . . . . .4.11. . .6 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . .3 The InertiaWheel Pendulum . . . . . . . 44 45 46 47 49 51 53 4 Reduction and Control of HighOrder Underactuated Systems 58 4. . . .1 Global Stabilization to an Equilibrium Point . .2. . . . . . . . . . 5. 5. . . . . . . 5.4. 5. . . . . . . . . .5 Desired Control Attitude Rd for Position Tracking . . . . . . . . . . Collocated Partial Feedback Linearization . . . . . . . . . . . . . . .9. . . . . . . . . . . . .7 3. . . . .1 The Acrobot . . . . . 5. . .11. . . . . . . .8 Control of Multilink Underactuated Planar Robot Arms . . . . . . . . . . . . 86 4. . .2 Input Moment Decoupling for A Helicopter . . . . 5. . . . . . . . . . . . .3 Underactuated Systems with Input Coupling .11. . . . . . . .3. . . . . .1 Model of a Flexible Link . . . . . . 7 112 113 123 127 132 133 137 139 140 143 144 147 149 150 151 152 157 157 161 161 164 170 170 172 . . . 58 4. . . .6 The VTOL Aircraft .2 Aggressive Swingup of the Inverted Pendulum . . 60 4. . . . Normal Forms for Underactuated Systems . . . . . . . . . . . . . . .10 Conﬁguration Stabilization for the VTOL Aircraft . . . . . . . . . . . . . . . .4.9. . 63 4. . . . . . . . .2 Underactuated Systems with Noninteracting Inputs and Integrable Momentums .6 Classiﬁcation of Underactuated Systems . . .2 The Noncollocated Output . . . 100 4. . . . . . . . . . . . . . . . .3 Switchingbased Controller . . . . . . . . . 5.9. . . . . . . . .3 Nontriangular Normal Form of a Helicopter . . . . . . 74 4. . . . . . . . . . . . . . . .9 Trajectory Tracking for a Flexible OneLink Robot . . . . . . . .5. .5 Momentum Decomposition for Underactuated Systems . . . .7 Nonlinear Control of the Core Reduced System . . . . . . .8 3. . . . . . . . . . . .11 Trajectory Tracking and Stabilization of An Autonomous Helicopter . . . . . . . . 5. . . . . . . . . . . . . . . . . Normal Forms for Underactuated Systems with 2 DOF . . . . . . . . . . . .2. . . . . .2 Underactuated Systems with Unactuated Shape Variables . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Dynamic Model of A Helicopter . . . . 5. . . . . 5. . . . 5. . . . . . . . . . . 80 4. . . . . . .5 The Rotating Pendulum . . . . . . . . . . . . . . . . . . 5. . 5. . . . . .2 The TORA System . . . . . . . . . .4 Underactuated Systems with Nonintegrable Momentums . .
. . . . . . .4. . .3 Stabilization of Nonlinear Systems in Feedback Form . . . . . 6.2 A Nonholonomic Mobile Robot . . . . . . . . . . .4. .3.1 Summary . . . . .5. . . . . .4. . . .3 Output Feedback Stabilization and Tracking . . . . . . . . . . . . . . . . .5 Applications . .2 The CartPole System with Large Length/Weak Gravity Eﬀects 262 7. . .1 Nonholonomic Systems with Symmetry .4. . . . . . . . . . . . . .4. .4. . . . . . . . . . . . . . . . . . . . . . . ξ2 ) .2 NearIdentity Diﬀeomorphism . . . . . . . . 253 7. . .3. . . .4 The Eﬀects of Bounded Control Inputs . . . . . . . . . . . . . . . . . . . . . .7 Thrust for Exponential Attitude Stabilization . . . 173 5. . . . . . . . . .4 Global Exponential Stabilization/Tracking for a Mobile Robot . . . . .2.5 Notion of Partial Semiglobal Asymptotic Stabilization . . . . . .5. . . . . .1 Dynamics of a Mobile Robot in SE(2) .3 Notions of Stabilization and Tracking . . . . . . . . . . . . . . . . . . . . . . . . 231 7. . . . . .3 Global Existence of FixedPoint Control Laws . . . . . . . . .2. . . . 224 7. . 224 7. . . . . . . . . . . .1 Nontriangular Nonlinear Systems Aﬃne in (ξ1 . .2 Applications to Nonholonomic Robots . .2 Cascade Backstepping Procedure . . . . . . . . . . . . . . . . . . 181 6 Reduction and Control of Underactuated Nonholonomic Systems 6. 232 7. . . . . 6. . . . . . . . . . . . . . . . . . . .4. . . . . 265 8 Conclusions 8.2. . . . . . . . . .2. 226 7.2 Future Work . . . 6. . . . . . . .5. . . . . .4. 258 7. . . 8. .3 A Car . . . . . .6 Asymptotic Tracking of a Desired Attitude Rd ∈ SO(3) . . . 217 7. 263 7. . . . 8. . . . . . . . . . . . . . 6. . . . 259 7. . . . . . . . . . .1 Hybrid Lagrangian Systems . . . 8. . . . . . . . . . . . . . . . 6. . 6. 8.3 The Rotating Pendulum . 8. . . . 8. . . . . . . . . . . . . . .2. .2. . . . . . . . . .2 Uncertainty and Robustness . . . 242 7. .5.1 A Rolling Disk . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Stabilization of Nonlinear Systems in Nontriangular Forms . 183 184 187 188 189 190 194 205 207 207 208 210 216 7 Control of Nonlinear Systems in Nontriangular Normal Forms 7. . . . . . . . . . .3 Control Design for Stabilization/Tracking . . 246 7. . . . . . .1 The CartPole System with Small Length/Strong Gravity Eﬀects259 7. . . 6. . . .2 Nontriangular Nonlinear Systems Nonaﬃne in (ξ1 . . . . . ξ2 ) . . . . . . . . . .2. . .2 Structure of Nontriangular Normal Forms of Underactuated Systems 222 7. .1 Introduction .4 The Generalized BeamandBall System . . . . . . . . . . . .2. . . 6. . . . .11. 6. . 268 269 270 270 272 272 272 272 272 8 . . . . . . . . . . . .5 Small Underactuated Mechanical Systems (SUMS) 8. 6. . . . . . . . . . . . . . . . . . . . .4 The Snakeboard . . .1 Standard Backstepping Procedure . . . .11. . . .4 Slightly Nontriangular Nonlinear Systems . . . . . . . .5. . . .6 Underactuated Systems in Strong/Weak Fields . . . . . . . .2. . .2.
.4. . . . .4 Proof of Theorem 7. . . . . . . B Symbolic Calculation of the Flat Outputs of the Helicopter C Proofs of Theorems in Chapter 7 C.A Dynamics of Mechanical Systems A. . . . . . .4: . . . . . . . . . C. . .4 C. . . . . . .3: . . . . .5: .5 The CartPole System . . . . . . .4: . . . .6 The Rotating Pendulum . . . . . . . . . . .4. . A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. . . . . C. . . . . . .4. . . . A. . . .7 The 3D CartPole System with Unactuated Roll and Pitch Angles . . . . . . . . .5 Proof of Theorem 7. . . . . . . . . . . . . . . . . . A. . . . . . . . .4 The BeamandBall System . . . . . . . . . . . . . . . . . . . . . . . . . .4. . 274 274 275 279 280 282 283 284 286 289 290 290 291 291 292 295 9 . . . . . . . . . . . . . . . . . . A. . . . .4. . . . . . . . . . . . . . . C. . . . . . . . .1 Proof of Theorem 7. . . .2 A Threelink Planar Robot . . . . . . . .2 Proof of the Stability of µSubsystem in Theorem 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Underactuated Mechanical Systems with Two DOF and Kinetic Symmetry . . . . . . . . . . . .3 Proof of Lemma 7.3 Twolink Planar Robots: the Acrobot and the Pendubot . A. . .8 An Underactuated Surface Vessel . A. . . . .
0) and control input p2 . . . . . . Trajectory of (p1 . . . . . . . . . . . . . . . . . . . . π. . . . . . . 0. . . . . . . q2 . . q2 ) for the Acrobot with initial condition (q1 . . . . q2 ) for the Acrobot with initial condition (q1 . . . (b) The Rotating Pendulum. . . . . An inverted pendulum on a moving platform with 4 DOF and 2 controls. . . . . . . . . . The VTOL aircraft. The function γ(q2 ) for a = 20/3. . . . . Trajectory of (p1 . . . 0) and control input p2 . . The Inertia Wheel Pendulum . . . . . . . . . . . . . p1 ) = (0. . . . . . . . . . . . . . . . . . . . . q2 . . . π. . . . . . . . . . . 0. . . . . . . .List of Figures 31 32 33 34 35 36 41 42 43 44 45 46 47 51 52 53 54 55 56 57 58 59 510 511 (a) The Acrobot. . . . . . . . . . . . p2 ) = (1. . p2 ) for the Acrobot with initial condition (q1 . . . . . . . . . . 0. . . . . . . . . . . . . . . 0) and control input p2 . . . 40 41 42 43 44 44 73 77 81 84 90 94 96 114 115 119 120 120 121 121 122 122 123 A ﬂexible onelink robot arm. . . . . . . . . . . Three possible actuation conﬁguration for a planar threelink robot arm. . p1 ) = (0. . . . . . and (d) u. . . . . . . . . . . . . . 0) and control input p2 . . . . The Acrobot . . . . p1 ) = (π/3. . . . . . . . . . . . . . . A triplelink robot arm with two actuators. . . . . . . . . . . . . A planar threelink robot arm. . The function α(qr ) for m1 = m2 = 1 and l1 = l2 = 1. . . . . . . . . . . . A triplelink robot arm with actuated shape variables. . . . . . . . . 0. . The TORA system. . . . . . The TORA system. . . . . . . . . . . . . . . b = 4. 0. . . . q2 . . . . 0): (a) (q1 . . . . . p1 ) = (π. . . . . . . . . . . . . . 0) and control input p2 . . . . . . . . . . c = 4/3. . . . . . . . . . . . Trajectory of (q1 . . . . . . . . . . . . . . . . . . . . . . . p2 ) for the Acrobot with initial condition (q1 . q2 . . . . . . 512 127 128 10 . . . . . . A ﬂying bird. . . . . . . . Trajectory of (q1 . . . . . p2 ). . . . . 0) and control input p2 . . . 0. . . q2 . . . Trajectory of (p1 . . . . . . . . q2 . . . . . (b) The Pendubot. . . . p1 ) = (π. . . . . . . . p1 ) = (π/3. . . . . . . . . . . . . q2 . . . . . . State trajectories and control input of the TORA system for the initial condition (q1 . . (c) (qr . The InertiaWheel Pendulum . . . . . . . . . . . . . . . . (b) (q2 . . p2 ) for the Acrobot with initial condition (q1 . . . . . . . . . . . . . . . . . . . . . . p1 . . . . . q2 ) for the Acrobot with initial condition (q1 . (a) The CartPole system. . . . . . . . Trajectory of (q1 . . pr ). . . . . . . . . . . . . . p1 ). . . . . . . . . . . . . . . . . . . The BeamandBall System. . . . . . . . . . . . . . . . . . . . . .
. . . . (b) shows the state trajectory in (q1 . . . . . . . . (b) The force F and control input u. . . . . . . . 516 (a) State trajectory of the cartpole system for the initial condition (5. . . . . . . . . . . . . p2 ) = (π. . . . . . q1 ). . . . . . . . . . . . 526 The VTOL aircraft. . . . . . . . . . . . 520 (a) State trajectory of the cartpole system for the initial condition (0. . . . . . . . . . . 525 A ﬂexible onelink robot arm. . . . . 0. 195 The geometry of the paths of the rear and front wheels of the car with a radius of rotation ρ. . . . . . . . . . . . . . . . . . . . . . . . . . . . 527 (a) The state trajectories of the VTOL aircraft. . 0). . . 521 The Rotating Pendulum . . . y1 )plane. . . 528 A helicopter. . . . . .1 (·). . . 0. . . . . . . . (a) shows the state trajectories for (q1 . . . . . . . . . (c) shows q2 . . . . . . . . . . 0) and σi (·) = sat0. . . . . . . . . x2 )plane for initial position x = (2. . . . . . . . 514 For the initial condition (π/2. . . . 188 A mobile robot . . 3)T (v = 0) and orientation angles θ = 0. . . . . . . . . 3)T (v = 0) and orientation angles θ = π. . . . . . . . . . . . . (c) shows q2 . . . 190 The model of a car. . . . 0. . . .212 Trajectories of the nonholonomic mobile robot in (x1 . . (b) Trajectory of the pendulum in (q2 . . . . . . . . . . . . . . . . . q1 ) plane. . . . . 202 The geometry of the paths of the rear and front wheels and the center of mass of the snakeboard with a radius of rotation ρ. . . . . . . . (a) shows the state trajectories for (q1 . . (b) The input force F and the trajectory in (q2 . . . . . . . . . . . . 7π/4. 0. . . . . . . . . . 0). . . . . . . . . . . . 519 (a) Aggressive swingup of an inverted pendulum on a cart from the downward position (q2 . . . 515 The CartPole system. . . . . . . . . . . . . . 3)T (v = 0) with angle θ = π/4 . . . . . . . . . . . . . . . . . . . . . . 0) and σi (·) = sat1 (·). . . . q1 ) plane. . . . . . . . . . . . . . . . . . . . 0). . . π/3. . . . . . . . (b) The path of the pendulum in (q2 . . . (b) shows the state trajectory in (q1 . . . . . . . . 5π/4. . . . . . . . . . . . . . . . . . q1 ). . . . . 0. . π. . . . . . . . . . . . . . . . . . . . . . . . . (b) The force F and control input u. . p2 )plane. . . . . . . . . . . . . . . . p2 )plane. . . 209 Trajectories of the nonholonomic mobile robot in (x1 . . . . . 3π/2. . . x2 )plane starting at x = (2. . . . . . . . . .513 For the initial condition (π. . . . . . . . . . 191 The Snakeboard. . . . . and (b) The path of the center of mass of the aircraft in (x1 . . . . . . ˙ ˙ ˙ and (d) shows the control input τ . . . . . . . . . . . . . . . . . . p2 )plane. 522 (a) Aggressive swingup of an inverted pendulum on a rotating arm from the downward position (q2 . . . π/3. . . . . . . . . . . 0). . π/2. . ˙ ˙ ˙ and (d) shows the control input τ . 3π/4. . . . . . . . . . . . . . . . 0).211 Trajectories of the nonholonomic mobile robot in (x1 . . . x2 )plane for initial position x = (2. 524 The BeamandBall System. . . . . . . . . . . . 523 The Pendubot . . π/4. . . 517 (a) State trajectory of the cartpole system for the initial condition (5. 518 The deadzone nonlinearity. p2 ) = (π. . . . 213 11 . 61 62 63 64 65 66 67 68 69 610 131 132 133 136 136 138 140 141 141 145 145 147 151 157 160 161 A vertical rolling disk . . . 204 A near identity transformation with a parameter λ 1.
. . . π/2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Rotating Pendulum. . . . . . . . 4. . . . . . . . . 213 214 214 215 248 250 260 261 263 265 267 275 281 282 283 285 286 12 . .8 and ν = 1. . . . . . . . . . . . . . . . . . . . . . . The CartPole system. . 2)T (v = 0) and initial orientation angles θ = 0. . . . . 2)T (v = 0) with angle θ = 0 . . . . . . . . . . . . . . . . . . . . 3π/4. . . . . . . . . . . . . . . . . . 614 Position trajectory tracking for a mobile robot from the initial point (4. . . . . . x2 )plane starting at x = (0. . . . 612 Evolution of position and controls for the nonholonomic mobile robot in (x1 . . . . . . The state trajectories of the generalized beamandball system with initial condition (1. . . 3. . . . . .611 Trajectories of the nonholonomic mobile robot in (x1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613 Trace trajectories of a twowheel nonholonomic robot with initial position (5. . π/2). . . . . . 4) and heading angles (a) θ = 0. . . . . The CartPole System . . . . . . . . . . . . The state trajectories of the CartPole system. . . . . . . . . . x2 )plane for initial position x = (0. 71 72 73 74 75 76 77 A1 A2 A3 A4 A5 A6 A C 2 smooth sigmoidal function with ﬂat tails with its derivatives. . . . . (c) θ = π. . . . . . . 2. . . . . and (d) θ = 3π/2. . . . . . . . . . . . . . . . . . 0. . . . . The BeamandBall System. . . . . . . . . . The state trajectories for the Rotating Pendulum. . . . 0) for g = −9. . . . . . . . (b) θ = π/2. . . A planar threelink robot arm. . . . . . . . . . . . . . . . . . . . . . and pitch angles. The 3D CartPole system with unactuated roll An underactuated surface vessel. . . . . . . The Rotating Pendulum . . . . . . . π/4. . . −π/6. . . . . . . . . . . . . . Sigmoidal ribbon S and compact set K. . . . . . . .
. . . . . . . n0 ).List of Tables 4.1 5. . . . . . . . 1). . . . . . . 139 The Values of τ ∗ ( . . . . . . . . . . . . . . . . . . .2 Classiﬁcation of Underactuated Systems . . . . . . . . . . . . 101 The Values of F ∗ ( . . . . . 144 13 . . .1 5. . . . . . . . . . . n0 .
Optimal Control. humans managed to fulﬁll their longtime dream to travel in the space and land on the moon. the study of mechanical systems goes back to Euler and Lagrange in the 1700’s. 14 . All along the way. a series of scientiﬁc. Control of Multibody Systems [53] in mechanics and rigorous analysis (e. advances achieved by researchers in Mechanics and Nonlinear Control Theory mutually aﬀected and enhanced each other. it was not until 1850’s that mechanical control systems came to the picture in regulation of steam engines. Aerospace Vehicles and Robotics Systems with ability to explore the surface of other planets are still among the most complex machines built by humans. and military applications motivated rigorous analysis and control design for mechanical systems. stabilization and output tracking) for nonlinear systems aﬃne in control [36. On the other hand. Robust Control theory was born and added to the above picture because of an inevitable need to deal with the presence of uncertainties in reallife control systems.g. 65]. which shaped the ﬁeld of nonlinear control theory.g.Chapter 1 Introduction 1. the eﬀorts of engineers and scientists together led to creation of Linear Control. industrial. Though. were just two theoretical consequences of this eﬀort. theoretically challenging nature of analysis of the behavior of nonlinear dynamical systems attracted many mathematicians to study control systems. During the past century. This is one of the greatest accomplishments of control scientists and engineers of the ﬁrst half of the past century.1 Mechanics and Control Theory Control of mechanical systems is currently among one of the most active ﬁelds of research due to the diverse applications of mechanical systems in reallife. Aerospace and Robotics applications remained as some of the most inﬂuential sources of motivation for rigorous analysis and control of both mechanical systems and nonlinear systems. and Nonlinear Control theories. As a result. More recently. Before the creation of most of these theories. controllability and observability) and control design (e. Adaptive Control. For the past 50 years.
e. As a result. highly nonlinear systems.e. The main control methods applied to examples of invertedpendulum type underactuated systems is based on swingup of the pendulum from its downward position and then switching to a balancing controller that is designed using a linearization technique or gain scheduling to balance the pendulum [94]. Underactuated systems appear in a broad range of applications including Robotics. the main body of this thesis involves the following topics: i) Dynamics. satellites with two thrusters and ﬂexiblelink robots). Underactuated Systems are mechanical control systems with fewer actuators (i. spacecraft. aircraft. in a surface vessel or aircraft). a twolink planar robot 15 . we provide an overview of the past research on classes of nonlinear control systems and mechanical systems that are relevant to this work. This includes swing up control using energybased methods [90] for the Acrobot (i. and Control of Underactuated Systems ii) Control of Highly Nonlinear Systems iii) Nonlinear Control of Cascade Systems with Nontriangular Structures In the following.g. Mobile Systems. Marine Systems. controls) than conﬁguration variables. In this thesis. the BeamandBall system. Flexible Systems. the TORA system (all are described in chapter 3)). The “underactuation” property of underactuated systems is due to the following four reasons: i) dynamics of the system (e. One of the main contributions of this thesis is explicit transformation of highorder underactuated systems into cascade nonlinear systems with both triangular and nontriangular structural properties. This motivated us to develop new control design methods for both cascade nonlinear systems with nontriangular structures and classes of nonlinear systems nonaﬃne in control. our main focus in is on control design for nonlinear systems that arise from control of an important and broad class of mechanical systems known as underactuated systems.g. we present the stateoftheart of research in each of the above topics. 1. This transformation is performed using a global/semiglobal change of coordinates obtained from the Lagrangian of the system in closedform.1.2 Nonlinear Control Systems In this section. iii) actuator failure (e. the Acrobot. control of the original (possibly) highorder underactuated system.g. and Locomotive Systems.g.1 Underactuated Systems Underactuated mechanical systems are systems that have fewer control inputs than conﬁguration variables. Reduction.e. helicopters. After applying this transformation. i. the Pendubot. Aerospace Systems. ii) by design for reduction of the cost or some practical purposes (e. underwater vehicles. locomotive systems without wheels). reduces to control of lowerorder nonlinear systems (possibly) nonaﬃne in control. the Rotating Pendulum. iv) imposed artiﬁcially to create complex loworder nonlinear systems for the purpose of gaining insight in control of highorder underactuated systems (e. the CartPole system.2.
accessibility of classes of underactuated mechanical systems has been recently addressed in a recent important work by Reyhanoglu et al. 14]. Passivitybased method is mostly used for swingup control design of underactuated systems with invertedpendulums [28. The VTOL (vertical take oﬀ and landing) aircraft is another example of an underactuated system that has been extensively used as a testbed for diﬀerent methods of trajectory tracking and conﬁguration stabilization. This includes tracking for slightly nonminimum phase systems [35] and hybrid/switching based control methods [58]. the cartpole system [21] .e. [34]. This is done based on the original work due to Mazenc and Praly in [56]. a twolink planar robot with an actuator at the shoulder) [92]. In addition to this issue. This is mainly based on the famous condition of Brockett on stabilizability of nonlinear systems using timeinvariant continuously diﬀerentiable state feedbacks [15]. Trajectory tracking for ﬂexible robots is rather complicated and common measurements like the angle of rotation or the position of the tip. the Pendubot (i. using small nested saturations for stabilization of feedforward cascade nonlinear systems due to Teel [102] and stabilization by output feedback [103]. The role of secondorder nonholonomic constrains in the necessity of the use of discontinuous stabilizing feedback laws for stabilization of underactuated systems is discussed in [111]. and the rotating pendulum [6]. A truncated modal analysis can be used to obtain a ﬁnite dimensional statespace model for ﬂexible robots [25.e. 90]. a triplelink inverted pendulum [90]. An example of a discontinuous stabilizing feedback for a system with an internal unactuated degree of freedom is given in [77]. The EulerBernoulli model for a ﬂexible arm is an inﬁnite dimensional system [25]. global stabilization of the beamandball system is achieved in [79]. Flexiblelink robots are an important class of underactuated systems that are appropriate for space applications due to their lightweight and fast execution of commands. A similar type result for attitude control of an underactuated spacecraft is given in [62]. the TORA system) that was transformed to a cascade form [37]. Moreover. In fact. This is based on a framework applied to analysis of controllability of nonholonomic systems [11] and a famous controllability theorem due to Sussmann [97]. and Marine applications is known. Moreover. Adaptive control [33] and sliding mode control techniques [96] have been also applied to underactuated mechanical systems for rather limited applications. Aerospace. [78]. Exponential stabilization of examples of underwater vehicles and surface vessels that are underactuated was achieved in [27] and [75] using appropriate coordinate transformations and analysis of a timevaring linear system. to the best of my knowledge. the beamandball has been the focus of study among researchers with diverse interests including approximate feedback linearization methods by Hauser et al. lead to a poor performance and nonminimum phase zero dy16 .with an actuator at the elbow) [86]. Due to its complexity. The balancing controller for the Acrobot using spline functions can be found in [13. a passivitybased approach was employed for a special example of a pendulum on a cart (i. so far no applications of energybased methods in control of reallife underactuated systems in Robotics. The main drawback of these passivitybased methods is their narrow range of applications. 107]. stabilization of the beamandball by construction of Lyapunov functions is addressed in [80]. Recently. respectively.
hybrid and switchingbased control methods are ﬁnding their way in control of underactuated mechanical systems [7. y ∈ Rm . recently.2) 17 .namics (see [20. and tracking for this system using state or output feedback has been quite wellunderstood for a long time. However. consider the following linear system in feedback connection with a memoryless nonlinearity φ x = Ax + Bu .e.1) where x ∈ Rn . One can observe that minor deviations from the standard problem of stabilization of an LTI control system and additional constraints make the system rather complex. u ∈ Rp . We start with linear timeinvariant control systems in the form x = Ax + Bu . ﬁrst we present an evolution of control systems from linear systems to nonlinear systems. In addition to more traditional methods. 41] and bipedal locomotion of walking robots [88. Questions regarding controllability. a noncollocated minimumphase output is proposed based on an analysis of the initial inﬁnite dimensional model and then a ﬁniteorder compensator is deigned for trajectory tracking. A = A 0 + ∆ where A0 is known and ∆ is unknown but normbounded). is there any static or dynamic state feedback that asymptotically stabilizes the system? or what if A. we obtain a nonlinear noncollocated minimumphase output for a ﬂexible onelink robot arm from a ﬁniteorder EulerLagrangian equations of the system. C are known up to an uncertainty (i. y = φ(Cx) ˙ (1. so far this method has been unable to stabilize the Rotating Pendulum or more general underactuated systems to an equilibrium point. In [20].2. stabilization. 89. 1.1). applying a control input that preserves the Lagrangian structure of a mechanical system) has been recently applied to local stabilization of the cartpole system and the rotating pendulum to an equilibrium manifold [10]. 24] and references therein).2 Highly Nonlinear Systems To describe what we mean by “highly nonlinear systems”. The method of controlled Lagrangians (i. observability. B. The stateoftheart of research in control of underactuated systems is currently very far from our goal to ﬁnd control design methods for broad classes of highorder underactuated systems that are eﬀective for Robotics and Aerospace applications. Here. As a further step. y = Cx ˙ (1.e. apart from linearizationbased techniques. 91]. if the system is controllable with bounded control inputs. For example. adding constraints or further speciﬁcations to the description of the system might make the control design for the system rather complicated. However. None of these problems can be addressed as simple as the control problems for the original linear system (1. control of underactuated mechanical systems has been mainly along the line of stabilization of special examples of cascade nonlinear systems and using energybased methods combined with a supervisorybased switching control. In conclusion.
they are not applicable to truly nonlinear systems (i.1) is x = f (x) + g(x)u . By a highly nonlinear system. lie algebraic conditions are not robust to uncertainties in 18 .2) with diﬀerent types of nonlinearities led to Absolute Stability theory in 60’s. this notion of feedback interconnection of a linear system and a nonlinearity was generalized to feedback of an LTI system and an uncertainty (or operator) that has bounded gain. h are nonlinear smooth functions. the linearity of the output does not simplify analysis of the system due to the fact that timeevolution of the system follows a nonlinear law. a modiﬁcation of (1. Moreover. u) (regardless of linearity or nonlinearity of h(x)) such that there exist no diﬀeomorphism (z. ob˙ servability.5) where f is a nonlinear function of (x. ∀z ∈ R. sigmoidal) nonlinearity σ. We call these systems slightly nonlinear systems. This led to Robust Stability theory [23] and more recently Integral Quadratic Constraints (IQC’s) [61] due to Megretski and Rantzer.where φ(0) = 0 and ∃c2 > c1 ≥ 0 : c1 z 2 ≤ z · φ(z) ≤ c2 z 2 . The analysis method employed in the preceding work was a rather involved timedomain analysis. In mideighties. y = Cx (1.e. T2 (x. In addition. stabilization. An inputoutput stability approach or a frequency domain analysis are the dominant tools in dealing with slightly nonlinear systems.4) can be found in [36] (see references therein for further details). v) = (T1 (x). and disturbance decoupling for nonlinear systems in (1. B satisfying z = Az + Bv. u) . A rather standard form for nonlinear systems aﬃne in control in analogy to (1. A comprehensive local theory regarding controllability. g. Obviously. and slightly nonlinear systems. has no fundamental similarities in terms of controllability and observability to an LTI control system or a slightly nonlinear system. Control design and analysis for systems in the form (1. due to the fact that a frequency domain analysis can only deal with systems that have linear timeevolution of the state.3) called a recurrent neural network with a saturationtype (i. 3]. systems with nonlinear timeevolution of the state) that do not include any basic linear parts.2) as ˙ x = σ(Ax + Bu) . Though. Dynamic neural networks are examples of highly nonlinear systems and only recently around midnineties conditions for controllability (only the discretetime case) and observability of these systems were introduced by Albertini and Sontag [2.e. y = h(x) ˙ (1. To be more precise. y = h(x) ˙ (1. tracking. these methods are successful in dealing with linear systems. we mean a system in the following form x = f (x. u)) and matrices A. uncertain linear systems. The main tools to address these control problems were diﬀerentialgeometry and Lie theory that became very common in the literature around late 70’s and early 80’s. Though these methods were rather successful in local analysis of nonlinear systems aﬃne in control they usually fail to work for a global analysis and nonlinear systems that are nonaﬃne in control.4) where f.
40. This motivated us to consider global/semiglobal stabilization and analysis of highly nonlinear systems that arise from the study of underactuated mechanical systems and nonholonomic systems. InputtoState Stability (ISS) theory. The main tools in this theory for robustness analysis to disturbances are control Lyapunov functions (CLF’s). The backstepping procedure has proven to be successful for systems in strict feedback form [57. The problem is that in general it is not easy to construct CLF’s for highly nonlinear systems. This suggests developing new stabilization methods for cascade nonlinear systems with nontriangular structures. 36. However. Namely. 5. Control of feedforward cascade nonlinear systems is also wellstudied.2. The simplest example of a feedforward nonlinear system is a perturbed chain of integrators that can be stabilized using small nested saturations [102]. or due to design of an output feedback or a dynamic state feedback. u) (1. and 7. 19 . g. In many control applications. Then. we show that the Rotating Pendulum and the Pendubot are examples of underactuated systems that have nontriangular structures and they do not satisfy any of the aforementioned growth conditions. It turns out that if some coupling terms in the dynamics of cascade nonlinear systems satisfy appropriate growth conditions then it is possible to stabilize them using lowgain or highgain feedback laws [80]. 6. combines both Absolute Stability and Robust Stability theories in one for highly nonlinear systems in the general form (1. In addition. 1.5). ξ) ˙ ˙ ξ = g(ξ. control of cascade nonlinear systems with nontriangular structures has proven to be rather problematic. This stabilization method is one of the main contributions of this work which will be presented in chapter 7. This is due to a counterexample by Sussmann [98] and existence of the peaking phenomenon [100]. essentially developed and introduced by Sontag [82. [79]. nonlinear systems in strict feedback forms and feedforward forms. after applying certain nonlinear coordinate transformation to the dynamics of nonlinear systems aﬃne in control. the transformed system or its subsystems could be nonlinear systems that are nonaﬃne in control. 84].f. a global/semiglobal control design and analysis is required. This topic is covered in chapters 4. In general. they are in the following form z = f (z. h. 80].6) The most wellknown results for cascade systems are related to nonlinear systems in triangular forms. In chapter 5. more general classes of feedforward systems were either controlled using a nonlinear smallgain theorem [101] or construction of Lyapunov functions [56].3 Cascade Nonlinear Systems Cascade nonlinear systems arise in many control applications either naturally after some change of coordinates.
feedforward form [102]. 56.3 Statement of the Contributions In this section. More detailed statement of the contributions are given in the subsequent sections. We call this systematic translation procedure reduction of underactuated systems. The main focus in this thesis is on nonlinear control of underactuated mechanical systems. We introduce these explicit change of coordinates for broad classes of underactuated systems including both reallife examples in robotics and aerospace applications and benchmark nonlinear systems. Under such a diﬀeomorphism. iii) Control design for each reduced normal form (or class). and ﬁxedpoint state feedback laws [67] (presented in chapter 7). 80]. i. Of course. this thesis includes a complete package for control design and analysis of both loworder and highorder underactuated systems with symmetry by providing i) Classiﬁcation of underactuated systems. 101. since even stabilization of the general ByrnesIsidori normal form with a doubleintegrator linear part is currently an open problem in nonlinear control theory. 87].1. ii) Reduction of each class to its normal form using an explicit transformation. All three types of aforementioned cascade normal forms are special classes of ByrnesIsidori normal form with a doubleintegrator linear part [36]. This is motivated by broad applications of underactuated systems and theoretically challenging problems that they have to oﬀer. and nontriangular linearquadratic form [69] (see section 3. Though. In fact.e. Therefore. This in turn implies that these normal forms can be controlled using backstepping procedure [57].8 for deﬁnitions). control of mechanical systems with nonholonomic ﬁrstorder (or velocity) constraints is very wellstudied. This is very important. we mean invariance of the kinetic energy under the action of a cyclic group. However. the original underactuated system transforms into a nonlinear system that is referred to as a normal form. feedforwarding [102. Our main contribution in this thesis is to provide analytical tools that allow us to translate and solve control problems for broad classes of highorder underactuated systems to equivalent control problems for lowerorder nonlinear systems. respectively. These cascade normal forms are nonlinear systems in strict feedback form [57]. the special structure of each type of normal form allows eﬀective use of the existing and recently developed nonlinear control design methods. Nearly all current applications of underactuated systems possess certain kinetic symmetry properties. control of general underactuated mechanical systems is currently considered a major open problem. Such a reduction process requires an explicit transformation. By kinetic symmetry. there is very little known about control of underactuated systems with nonholonomic secondorder (or acceleration) constraints. the reader should notice that this 20 . a global change of coordinates (or diﬀeomorphism) in closedform. we brieﬂy present the main contributions of this thesis. It turns out that underactuated systems with symmetry properties can be reduced to normal forms that are structured cascade nonlinear systems. based on two recent surveys [78.
Then. A Lyapunov function candidate for this analysis can be calculated via the Lyapunov function of the core reduced system of the unperturbed system. This class of underactuated systems with nonintegrable momentums can be then reduced to cascade normal forms plus structured perturbation terms. The key tools in reduction of highorder underactuated systems are normalized generalized momentums conjugate to external or shape variables together with their integrals (see section 2. In chapter 5. The classical symmetry is deﬁned as the invariance of the Lagrangian of a system under a symmetry group action. In other words. It turns out that actuation or lack of actuation of the shape variables of an underactuated system and also integrability of the normalized momentums play the fundamental role in the structure of the corresponding normal form of the system. mechanical systems that their inertia matrix is independent of a subset of conﬁguration variables possess kinetic symmetry property w.t.is a diﬀerent notion than the famous classical symmetry in mechanics [1. after performing a Lyapunovbased Robustness Analysis. We call this subset of variables external variables. 54] which in a simpler form was familiar to Lagrange. We introduce a method called Momentum Decomposition which uniquely decomposes a nonintegrable normalized momentum as the summation of an integrable momentum and a nonintegrable momentum. this allows us to use the Hamiltonian/Lagrangian of the reduced system as a Lyapunov function. the stability of the perturbed system in closedloop with the nominal nonlinear controller can be veriﬁed. we provide its associated normal form. A possible way to deal with this problem is to design a nonlinear controller for the nominal unperturbed system that renders the desired equilibrium point of the underactuated system asymptotically stable and makes the perturbation term exponentially vanishing (the details are provided in chapter 4). this subset of variables. These two properties together with another two basic properties of underactuated systems led us to classiﬁcation of underactuated mechanical systems with kinetic symmetry to 8 classes. Several examples of reallife and benchmarktype highorder underactuated systems possess nonintegrable normalized momentums (e. I obtain ByrnesIsidori normal forms for underactuated systems with zerodynamics that are welldeﬁned lowerorder Lagrangian systems.8 for the deﬁnitions). we apply our theoretical results in reduction and control of highorder underactuated systems to a number of complex underactuated systems. Here is a list of contributions on these applications: – Global asymptotic stabilization of the Acrobot to any arbitrary equilibrium point of the system including the upright equilibrium using a nonlinear state feedback. and examples from reallife and benchmark nonlinear systems. shape variables are conﬁguration variables that appear in the inertia matrix of a (simple) mechanical system. A surprising result is that the reducedorder subsystem of the normal form. Thus. 21 . a ﬂexiblelink robot and a 3D CartPole system). control design method. For each class. In certain cases.r. In simple words.g. The compliment of the set of external variables are called shape variables. for several classes of underactuated systems is a Lagrangian system satisfying forced (or unforced) EulerLagrange equations.
– Semiglobal asymptotic stabilization of the CartPole system using a ﬁxedpoint state feedback law. In [90]. 1.1 Normal Forms for Underactuated Systems Underactuated systems as mechanical systems with fewer inputs than conﬁguration variables with a nonholonomic secondorder (or acceleration) constraint are not fully (i. This is called 22 . – Almost global exponential stabilization of the CartPole system. – Automatic calculation of diﬀerentially ﬂat outputs for the VTOL aircraft as a byproduct of the decoupling of the eﬀects of the body torque in the translational dynamics of the VTOL. – Global asymptotic and local exponential stabilization of the 2D and 3D CartPole systems to their upright equilibrium point. – Trajectory tracking for a ﬂexible onelink robot arm using a nonlinear noncollocated output.3. – Reduction of stabilization of a planar nlink robot underactuated by one to stabilization of the Acrobot or the Pendubot.e. and the Rotating Pendulum to their upright relative equilibrium point.4) can be globally linearized using an invertible change of control. – Automatic calculation of weakly diﬀerentially ﬂat outputs for a relatively accurate model of an autonomous helicopter as a byproduct of the approximate decoupling of the eﬀects of the body torque in the translational dynamics of the helicopter. – Partial Semiglobal asymptotic stabilization of the Rotating Pendulum and the BeamandBall system using a ﬁxedpoint state feedback law.– Global asymptotic stabilization of the TORA system using a nonlinear state feedback. – Global position and attitude stabilization for the VTOL aircraft with strong input coupling. exact) feedback linearizable. – Global exponential stabilization of a nonholonomic twowheeled mobile robot to a point 1 far from the origin using a smooth dynamic state feedback. – Almost global exponential tracking of the feasible trajectories of an autonomous helicopter. – Global asymptotic stabilization of the InertiaWheel Pendulum using a nonlinear state feedback. Spong showed that actuated conﬁguration variables of an underactuated system with noninteracting inputs (deﬁned in section 3.
Partial feedback linearization is an initial simplifying step for reduction and control of underactuated. reduces to solving a ﬁnite number of ODE’s which constitute the characterization equations of the PDE. and ii) linearization of actuated conﬁguration variables for underactuated systems with input coupling. we prove that stabilization of an nlink (n > 2) planar robot with (n − 1) controls reduces to stabilization of the Acrobot or the Pendubot regardless of n. 23 . As an example. in chapter 4. in chapter 4. The latter one is very useful in aerospace applications. one obtains a cascade normal form which is a special class of ByrnesIsidori normal form with a doubleintegrator linear part [36]. Derivation of explicit normal forms for loworder underactuated systems is important. 2 DOF) systems. based on Frobenius’s theorem. this PDE can be explicitly solved. For loworder underactuated systems with 2 DOF. This is a important initial step in some of our results and proves that a broad class of underactuated systems can be partially linearized.3. because we later show that control of special classes of highorder underactuated systems reduces to control of a number of loworder (e.g.t. or it is unactuated. After applying this transformation. respectively. we propose an appropriate structure for a global change of coordinates that decouples these two subsystems w. particularly. We call these two classes of underactuated systems ClassI and ClassII systems. In chapter 3.collocated partial feedback linearization. Since the new control appears both in the actuated and unactuated subsystems of the original underactuated system. 1. in turn. Solving this PDE. the main diﬃculty arises after applying this change of control. Either their shape variable is actuated. The former one is used for partial feedback linearization of underactuated systems with unactuated shape variables (e. The method mainly relays on the use of normalized generalized momentums and their integrals (wherever possible). However. for decoupling of the eﬀects of the body torque in the translational dynamics of a VTOL aircraft or a helicopter. we present a new method to resolve this issue. combination of a partial feedback linearization step and an explicit decoupling change of coordinates is used for reduction of underactuated systems with two degrees of freedom.2 Reduction and Control of LowOrder Underactuated Systems In chapter 3. Instead. The condition is a simple algebraic test on the inertia matrix of the system. a ﬂexiblelink robot or the Rotating Pendulum). The weakness of this result is that to obtain the change of coordinates. regardless of the method used for decoupling of the actuated and unactuated subsystems. Solving this PDE explicitly for highorder underactuated is not a feasible approach.g.r. For this purpose. we extended the partial feedback linearization result in [90] to two other cases: i) linearization of the dynamics of the unactuated conﬁguration variables which is called noncollocated partial feedback linearization. the new control. it is required to solve a ﬁrstorder partial diﬀerential equation (PDE). Suﬃcient conditions are provided for the existence of this diﬀeomorphism. There is a simple way to classify underactuated systems with 2 DOF and a single shape variable.
for n = 2.3.8.3 Reduction and Control of HighOrder Underactuated Systems Most of reallife underactuated systems are highorder systems. It turns out that the integrability of theses normalized momentums plays a fundamental role in the structure of the normal forms for highorder underactuated systems. n controls. Our key analytical tools here are the normalized generalized momentums conjugate to actuated/unactuated conﬁguration variables and their integrals (wherever possible). consider a class of underactuated systems with n degrees of freedom. and the BeamandBall system. An important property of normal forms for highorder underactuated systems is that they are physically meaningful. m actuated shape variables with noninteracting inputs. the CartPole system. We provide suﬃcient conditions so that under a second change of coordinates. This normal form is in fact a cascade nonlinear system in strict feedback form with a zerodynamics (associated with the shape variables as the output) which is a Lagrangian function 24 . We prove that there exists a global diﬀeomorphism obtained from the Lagrangian of the system that transforms the dynamics of a ClassI underactuated system into a reducedorder system with a new conﬁguration vector of dimension (n−m) and a welldeﬁned Lagrangian function parameterized by the shape variables which satisﬁes the unforced EulerLagrange equations. and nm unactuated degrees of freedom (e. Stabilization of general partiallylinear nontriangular cascade nonlinear systems is currently an open problem [36]. the Pendubot. the CartPole system is the only one that falls within this class and can be globally asymptotically stabilized using a state feedback in the form of nested saturations due to Teel [102].g. we present detailed control design for all the aforementioned examples of underactuated systems with 2 DOF. We address this problem for important classes of nontriangular linearquadratic normal forms and more general nontriangular forms in chapter 7. the Rotating Pendulum. This motivated us to focus on reduction of highorder underactuated systems (in chapter 4). the InertiaWheel Pendulum. Among all the stated examples. 1. m = 3 the system has 8 DOF and 2 controls). We proved that all ClassI underactuated systems can be transformed into a cascade nonlinear system in strict feedback form and (possibly) globally asymptotically stabilized using backstepping procedure. In chapter 5. The kinetic symmetry property of underactuated systems and the degree in which the shape variables are actuated are two main factors in reduction of highorder underactuated systems. Examples of ClassII systems include ﬂexiblelink robots. Assume the normalized momentums conjugate to the (n − m) external variables of this system are integrable. all ClassII underactuated systems can be globally/semiglobally transformed into cascade systems in nontriangular quadratic normal form (deﬁnition 3.5). and the TORA system. In contrast. ClassII systems can be reduced to cascade nonlinear systems in feedforward forms. We call this class of highorder underactuated systems ClassI systems. To be more precise. An aircraft or a helicopter has 6 DOF and 4 controls.Examples of ClassI systems are the Acrobot. A ﬂexible nlink robot with m deformation modes at each link has n(m + 1) DOF.
For the reduction of ClassI and ClassII underactuated systems. we provide a method for stabilization of this core reduced system using a nonlinear static state feedback. To address reduction underactuated systems with nonintegrable momentums. The answer to this question is positive. We provide simple algebraic suﬃcient conditions on the inertia matrix and potential energy of the system such that the nontriangular normal form of a ClassII underactuated system can be transformed into a feedforward system using a second explicit change of coordinates. particularly in aerospace applications and locomotive systems. In this case. the reduction is rather straightforward.e. These Lyapunov functions can be later used for robustness analysis of the perturbed normal forms of ClassI underactuated systems.itself. respectively. the zerodynamics) is a welldeﬁned Lagrangian system (under minor conditions). we obtain an energytype Lyapunov function for the closedloop core reduced system. consider the class of underactuated systems with integrable normalized momentums conjugate to shape variables and noninteracting inputs. In general. In chapter 4. For ﬂat underactuated systems including aircraft and helicopters. After the reduction process. is the presence of input coupling between actuated and unactuated subsystems. In chapter 4. Other examples include a swimming ﬁsh and a walking human being. The complexity of the structure of this input force is one of the main sources of extreme diﬃculties in control design for ClassII underactuated systems. We refer to this Lagrangian zerodynamics system as the core reduced system. This method uniquely decomposes a nonintegrable momentum as 25 . In certain cases. But it satisﬁes a forced EulerLagrange equation with a rather complex input force with a nonlinear dependence on both the conﬁguration and momentum of the reduced Lagrangian system. A natural question to ask is whether it is possible to reduce these systems if their normalized momentums are nonintegrable. As a byproduct. we introduce a new procedure called Momentum Decomposition. ClassII underactuated systems can be explicitly transformed into nontriangular quadratic normal forms. This class of reallife systems can potentially have very high degrees of freedom. The subclass of ClassII systems that satisﬁes these suﬃcient conditions are called ClassIII systems. The integrability property of the normalized momentums can be considered as the third most important factor in reduction of highorder underactuated systems. We call them ClassII underactuated systems. We call these new classes of nonﬂat and underactuated systems with input coupling ClassIV and ClassV systems. The second important factor in reduction of underactuated systems. many locomotive systems in the nature fall within this class. we address reduction of nonﬂat underactuated systems with input coupling. we assumed the normalized momentums are integrable. still the reducedorder subsystem (i. An example is a ﬂying bird as a threebody system with 14 DOF and 8 controls. Now. the physical property of the zerodynamics subsystem allows using Hamiltonian/Lagrangian of the core reduced system as a valid Lypunov function. All of these three examples have actuated shape variables and they locomote in their environment using the coupling between their translational dynamics and shape variables. Therefore. the presence of input coupling is very crucial for the purpose of locomotion of these underactuated systems.
The point x = x + v(θ) is called an nearby point of x. For each class. we prove that the kinematic model of a snakeboard and a car are diﬀeomorphic. We use a nearidentity change of coordinates for global exponential stabilization/tracking of a nonholonomic twowheeled mobile robot to an nearby point/trajectory of a desired position/trajectory using a smooth dynamic state feedback ( 1). Namely. 1. and ClassVIII systems. ClassI through ClassVIII systems which are precisely deﬁned with their subclasses in section 4. ClassVI. ψ is clearly a diﬀeomorphism which is equal to id(x) = x for = 0. ) = x + v(θ) where v(θ) is a unit vector and for a ﬁx θ and . Our reduction method. it is fullyactuated. Otherwise. An example of a nearidentity diﬀeomorphism is ψ(x.3.VIII underactuated systems with nonintegrable momentums are perturbed versions of the normal forms of ClassI. The result of classiﬁcation of underactuated systems is summarized in Table 4. and 2 nonholonomic contraints [50].III systems with a structured perturbation that only appears in the dynamics of the momentum of the reduced subsystem. We introduce a new class of parameterized diﬀeomorphisms that are called nearidentity diffeomorphisms.1. Based on Brockett’s result [15].II. 64) is a mobile robot with 6 degrees of freedom. aerospace systems.II. This motivated us to formally deﬁne the notions of stabilization and tracking for nonlinear systems. An example of this class is a simple twowheeled mobile robot. 26 .VII. In conclusion. The snakeboard example (Fig. θ.II. and benchmark nonlinear systems. 3 controls. Let us call the underactuated systems with nonintegrable normalized momentums corresponding to ClassI.4 Reduction and Control of Underactuated Systems with Nonholonomic Velocity Constraints In chapter 6. the reduced system is underactuated. related examples and an appropriate control design method are given. we focus on reduction of underactuated systems with kinetic symmetry and nonholonomic velocity constraints.6. the class of mechanical systems with ﬁrstorder nonholonomic constraints can not be stabilized to the origin using a smooth static state feedback. This implies that the snakeboard can be transformed into a ﬁrstorder chainedtype nonholonomic system (equation 2. We prove that normal forms of ClassVI. respectively. ClassVII. we obtain 8 classes of underactuated systems. Our main contribution is that the methods for reduction and control of highorder systems presented in chapter 4.8) using an explicit change of coordinates.the summation of an integrable momentum and a nonintegrable momentum. applies to the majority of important examples of robots. the reduced system can be represented as the cascade of the constraint equation and a lowerorder underactuated (or fullyactuated) Lagrangian system. If the sum of the number of controls and constraint equations is less than the dimension of the conﬁguration vector. provides a systematic way for calculation of the reduced system of the snakeboard. Furthermore. Our main result on reduction of nonholonomic underactuated systems with symmetry is that under certain assumptions.
systems in nontriangular linearquadratic normal form. Chapter 1. we address this stabilization problem for important classes of nontriangular cascade nonlinear systems. In chapter 8.4 Outline The outline of this thesis is as follows. classiﬁcation. we introduce ﬁxedpoint state feedback laws for stabilization of nontriangular cascade nonlinear systems. and control of highorder underactuated systems.3.3. This includes normal forms for ClassII.6 Control of Nontriangular Cascade Nonlinear Systems Stabilization of nontriangular cascade nonlinear systems is currently considered a major open problem [36]. 1.1. and classiﬁcation of loworder underactuated systems.3. we address reduction and control of underactuated systems with nonholonomic velocity constraints and symmetry properties. ClassIV. we provide detailed control design for several examples of loworder and highorder underactuated systems in robotics and aerospace applications. 1. Our main contributions are presented in chapter 4. This includes reduction. In chapter 5. we give some background on classes of mechanical control systems and deﬁne some basic notions in mechanics. In chapter 2. Chapter 3 is devoted to the dynamics of underactuated system. In chapter 7. In chapter 7.5 Applications in Robotics and Aerospace Vehicles In chapter 5. we make concluding remarks and state possible directions of future research. and ClassVII underactuated systems. we present detailed control design for several robotics and aerospace applications. A list of some of the examples and their associated control task is stated in the last part of the introduction of section 1. several examples of underactuated systems. normal forms for underactuated systems. provides an introduction and statement of the contributions. In chapter 6. partial feedback linearization techniques. particularly. 27 .
Therefore.Chapter 2 Mechanical Control Systems 2. We start by deﬁning simple mechanical control systems and introduce Legendre normal form of a simple mechanical system. we discuss fullyactuated mechanical systems and their trasformation into a set of doubleintegrators using a change of control. 2.1 Introduction This chapter provides some background on mechanical control systems. M (q) is the inertia matrix which is a positive deﬁnite symmetric 28 . We deﬁne underactuated mechanical systems in contrast to fullyactuated mechanical systems and demonstrate why control of underactuated systems is a challenging open problem. We brieﬂy review stateoftheart of research in modeling and control of mechnaical systems with nonholonomic velocity (i. q) = K − V = q T M (q)q − V (q) ˙ ˙ ˙ 2 where q ∈ Q denotes the conﬁguration vector that belongs to an ndimensional conﬁguration manifold Q. A special class of mechanical systems called ﬂat mechanical systems is described next. Simple mechanical systems are systems that their Lagrangian is in the form of the diﬀerence between a (positive semideﬁnite) kinetic energy and a potential energy 1 L(q. we state forced EulerLagrange equations of motion as the model of control mechanical systems and introduce Legendre normal form for simple mechanical systems. We present a result on controllability limitations of underactuated ﬂat mechnaical systems in lack of gravity terms based on the Legendre normal form of a ﬂat underactuated system.e. ﬁrstorder) constranits and clarify their close relations to underactuated mechanical systems that contain nonholonomic acceleration constraints. the role of symmetry and conservation laws in mechanics and their relation through Noether’s theorem is explained in the ﬁnal section. Then.2 Simple Lagrangian Systems In this section. Symmetry in mechanics plays a crucial role in some of the main results of this thesis regarding reduction of underactuated mechanical systems.
j Γk (q)qi qj + gk (q) = eT F (q)u. We call (2.2) is invertible and the dynamics of the mechanical system in (2. gk (q) = ∂qk V (q). i = {1. The EulerLagrange equation for this mechanical control system is as the following d ∂L ∂L − = F (q)u (2. In other words. . because M is symmetric. q). . p) = C(q. x2 = p. The equations of motion for this mechanical system can be derived as the following Σj mkj (q)¨j + Σi.2) can be rewritten in the canonical form q = M −1 (q)p ˙ ˜ p = −G(q) + C T (q. one gets ˙ M (q) = C(q.1) dt ∂ q ˙ ∂q where u ∈ Rm and F (q) = (f1 (q).matrix. . fm (q)) denotes the matrix of external forces. . The term C(q. . the Legendre transform with respect to q p= ∂L = M (q)q ˙ ∂q ˙ (2.5) (2. .3) can be rewritten as x1 = M −1 (x1 )x2 ˙ x2 = −G(x1 ) + xT Q(x1 )x2 + F (x1 )u ˙ 2 or x = f (x) + g(x)u ˙ 29 (2. Deﬁning the variables x1 = q. Γk (q) = ( ij 2 ∂qi ∂qj ∂qk in the vector form. Also.3) ˜ where C(q. q ˙ ˙ ij k k = 1. q). n where ek is the kth standard basis in Rn . The in˙ teresting relationship between the matrices M and C is that S0 = M (q) − 2C(q. p)M −1 (q)p + F (q)u ˙ (2. m} be m linearly independent external forces applied to the system. M −1 p). Let fi (q) : Q → Rn . . ˙ ˙ G(q) contains the gravity terms (see [95] for more details and examples). q) + C T (q. . q)q + G(q) = F (q)u q ˙ ˙ cij = Σn Γi (q)qk ˙ k=1 kj ˙ ˙ ˙ is an element of C(q. (2. and Γk (q) are called ij Christoﬀel symbols and can be deﬁned as 1 ∂mkj (q) ∂mki (q) ∂mij (q) + − ).4) .3) Legendre Normal Form of a simple mechanical system. . . K is the kinetic energy and V (q) is the potential energy of the system. we have M (q)¨ + C(q. . q)q contains two types of terms involving qi qj that are called Centrifugal terms (i = j) and Coriolis terms (i = j). q) ˙ is a skew symmetric matrix. . Taking this property into account and noting that M (q) ˙ ˙ ˙ is a positive deﬁnite matrix.
Remark 2.3) is equivqlent to the following forced Hamiltonian equations ∂H(q. i. q)q + G(q)) ˙ ˙ and redeﬁning the variables as x1 = q.t. x2 ) and vector ﬁelds f (x) = col(M −1 (x1 )x2 .5) is that many analytical tools are available for controllability and observability analysis and control design of nonlinear systems aﬃne in control [36]. In the following sections.2. Remark 2. For fullyactuated systems the number of control inputs is equal to the dimension of their conﬁguration manifold.1).3 Fullyactuated Mechanical Systems Consider the control mechanical system in (2. x2 = q.2) is a secondorder ODE.2. For mechanical systems.which is a nonlinear system aﬃne in control with state x = col(x1 . p) p = − ˙ + F (q)u ∂q However. Later. x 2 and g(x) is independent of x2 .r. However. Therefore. F (x1 )) with the property that f (x) has a mixed linearquadratic structure w. 2. F (q) is an invertible matrix. Deﬁning the Hamiltonian of a simple Lagrangian system as 1 H(q. the Hamiltonian form is not appropriate for direct controllability analysis and control design compared to the Legendre normal form.1. 2 The main advantage of the form (2.e. Note that σ is a cubic matrix such that xT Q(x1 )x2 ∈ Rn . The Legendre normal form is a ﬁrstorder ODE and (2.e.2. −G(x1 ) + xT σ(x1 )x2 ) 2 g(x) = col(0.3) are two equivalent forms. This can be proved by applying the following change of control u = F (q)−1 (M (q)v + C(q. fullyactuated mechanical systems are exact feedback linearizable (i. they have no zerodynamics [36]). We call (2. we describe several important classes of mechanical control systems.2) and (2.1) a fullyactuated mechanical system if m = rank F (q) = n. to obtain ˙ x1 = x2 ˙ x2 = v ˙ 30 . p) = pT M −1 (q)p + V (q) 2 the Legendre normal form in (2.3) is more appropriate and can be directly used for controllability analysis of a mechanical system. equations (2. p) q = ˙ ∂p ∂H(q. we will see that from control point of view (2.
q. C. One of the main contributions of this thesis is to decouple the actuated and unactuated subsystem of the underactuated system in (2. A comprehensive treatment of control of underactuated system will be presented in the subsequent chapters. the ﬁrst (m − n) equations in (2.7).7) is called a second˙ ¨ order nonholonomic constraint. The same is true if the control input of a mechanical system is required to be bounded. control and analysis of mechanical systems could potentially be rather complicated. For the special case of F (q) = (0.g.6) with respect to the new control (see section 3.1)). and gravity terms that can be highly nonlinear. Im )T can be linearized using an invertible change of control. or a nonholonomic acceleration constraint (e. In other words. parametric uncertainties in M. 2. Im )T .6) as the unactuated subsystem and to the last m equations of (2.4 Underactuated Mechanical Systems A control mechanical system with conﬁguration vector q ∈ Q and Lagrangian L(q. G). This procedure is called partial feedback linearization. under any holonomic or nonholomic constraint (both to be deﬁned). we refer to the ﬁrst (n − m) equations of (2. 31 . The main focus of this thesis is control design and analysis of underactuated mechanical systems. q ). assume F (q) = (0.8). the unactuated subsystem of (2. Im )T .7) which contains Coriolis.6) still remains as a nonlinear system that is coupled with the linearized actuated subsystem through both the new control input and other nonlinear terms. equation (2. q) ˙ satisfying the EulerLagrange equation d ∂L ∂L − = F (q)u dt ∂ q ˙ ∂q (2. This is why most of problems for fullyactuated mechanical systems can be reduced to equivalent problems for linear systems.6) can be expressed as a secondorder dynamic equation ϕ(q. see section A. q ) = 0 ˙ ¨ (2. Moreover. As a special case. underactuted systems are mechanical systems that have fewer actuators than conﬁguration variables. This fact suggests that control of mechanical systems is challenging either in the presence of uncertainties (e.6) is called an Underactuated Mechanical System (UMS) if m = rankF (q) < n = dim(Q) (see the deﬁnitions for equation (2. Spong has shown that the actuated subsystem of a general underactuated system with F (q) = (0. centerifugal. ¨ If there exists no function h such that h = ϕ(q. We assume all underactuated systems throughout this work have secondorder nonholonomic constraints unless otherwise is stated. This highly complicates control design for underactuated systems.g. Then. or when the number of control inputs is less than n which in both cases exact feedback linearization is not possible. after partial linearization. However.which is clearly a (vector) double integrator. In [90].6) as the actuated subsystem. q. This restriction of the control authority does not allow exact feedback linearization of underactuated systems.
pi (t) = ˙ pi (0). . Proposition 2. M −1 (q)p) = 0. .5. More recently.1. Bloch and Crouch addressed modeling of nonholonomic systems on Riemmanian manifolds [8]. gn (q))T . then the system is not controllable/stabilizable. i. p) = C(q. ∀t ≥ 0. If pi (0) = 0 then the system cannot be stabilized to any equilibrium point/manifold with p = 0.6 Nonholonomic Mechanical Systems W T (q)q = 0 ˙ A Lagrangian mechanical system with m < n velocity constraints ˙ (W is an m × n matrix) that are nonintegrable (i. pi = 0 and pi is a conserved quantity. The history of research in this ﬁeld goes back to early 1900’s. then we have z = M 1/2 q ˙ ˙ and the Lagrangian of the system can be written as 1 ˙ ˙ L(z.e.e. h(t) : h = W T (q)q) is called ˙ a Mechanical System with Firstorder Nonholonomic Constraints. z) = z T I z − V (ψ −1 (z)) ˙ 2 Thus. many advances have been made on controllability of nonholonomic systems by Lewis and Murray 32 . qi is unactuated and gi (q) = 0. Consider an underactuated ﬂat mechanical system with the gravity term G(q) = (g1 (q). Proof. Controllability and stabilization of nonholonomic systems is considered in [11].5 Flat Mechanical Systems Mechanical systems that their inertia matrix is constant are called Flat Mechanical Systems. 2. Control of nonholonomic mechanical systems with velocity constraints has been extensively studied in the recent years by numerous researchers. M (z) = I is constant and in new coordinates the mechanical system is ﬂat. One of the characters of a ﬂat mechanical system is that the Christoﬀel symbols associated with their inertia matrix vanishes identically and therefore the matrix C(q. q) vanishes as well. Now. In an important work. The Legendre normal form for a ﬂat mechanical system with B(q) = I is as the following q = M −1 p ˙ p = τ − G(q) ˙ ˜ This is due to the fact that C(q. Suppose for some 1 ≤ i ≤ n. assume that for a mechanical system with inertia ˙ matrix M (q) there exists a diﬀeomorphism z = ψ(q) such that dψ(q) = M 1/2 (q)dq. . The following proposition reveals a major controllability limitation of ﬂat underactuated mechanical systems. Formulation of the dynamics of nonholonomic systems can be found in [64]. Based on the assumptions.2. .
locomotion and control design for nonholonomic systems by Ostrowski in [74]. (kn ) xn (k ) = u1 = u2 = x 2 u1 = xn−1 u1 (2. q2 )T . In the literature [111]. i = 1. q2 )T . carlike vehicles.8) where ki ≥ 1. . then regardless of the control input we have d ∂L ∂L ( )− =0 dt ∂ q1 ˙ ∂q1 or m11 (q)¨1 + m21 (q)¨2 + h1 (q. Mobile robots.6. .. the theoretical results in [60] were applied to local exponential stabilization of the attitude of a spacecraft with nonholonomic constraints 33 . A discontinuous dynamicstate feedback was introduced for stabilization.8). Recently. . Following the line of formulation in [11] and based on [64]. in a an important work [47]. Traditionally. q) + G(q) = W (q)λ + B(q)u q ˙ W T (q)q = 0 ˙ where λ ∈ Rm is the vector of Lagrange multipliers. stabilization of chainedform systems with at most ﬁrstorder nonholonomic constraints has been done using sinusoids [63] or homogeneous timevarying state feedback [60]. (quasi) exponential stabilization of higherorder chainedtype systems was addressed by Laiou and Astolﬁ based on the original work of Astolﬁ [4]. [9].8) is nonholonomic with ﬁrstorder constraints. [78]. q) = 0 q q ˙ where mij ’s are block matrices of the inertia matrix associated with q = (q1 . This is due to the fact that if B(q) = (0. sometimes in contrast to nonholonomic systems with ﬁrstorder (or velocity) constraints. [59] due to M’Closkey and Murray.. I)T is partitioned according to q = (q1 .1. Remark 2. However. n are integers. system (2. this terminology is somewhat misleading since the Lagrangian of an underactuated mechanical system satisﬁes EulerLagrange equations of motions without any external diﬀerentialalgebraic constraints that requires use of Lagrangian multipliers in a variational setting as in [8] for nonholonomic systems with ﬁrstorder constraints.[49]. For ki = 1. and reduction of nonholonomic systems with symmetry due to Bloch et al. Later. . and multitrailer systems [104] are examples of chainedform systems in (2. Control design for broad classes of nonholonomic mechanical systems can be reduced to control of chainedtype systems as the following x1 1 (k ) x2 2 (k ) x3 3 . the dynamics of nonholonomic systems with velocity constraints can be described as the following: M (q)¨ + C(q. underactuated mechanical systems are called systems with secondorder (or acceleration) nonholonomic constraints.
This is due to the fact that the degree of the secondorder equation of motion (i.r. and the role of symmetry in reduction of the dynamics of a mechanical system. n} ∂qi Denote the ith generalized momentum by pi as pi = ∂L ∂ qi ˙ and consider the unforced EulerLagrange equations of motion d ∂L ∂L − =0 dt ∂ qi ∂qi ˙ It immediately follows that the symmetry of L w. a car.e. the condition pi = 0 is equivalent to a ﬁrstorder constraint ˙ W T (q)q = pi (0) ˙ where W (q) = (mi1 (q). Lagrange himself was aware of this fact and he called qi an ignorable coordinate.t. qi .e. left invariant Lie groups. i ∈ {1.7 Symmetry in Mechanics In this section. This gives rise to symmetries (i. the conﬁguration variable qi iﬀ ∂L = 0. Therefore. . the Lagrangian of a helicopter. or a Satellite is independent of their position.t. Hamiltonian form) for the mechanical system reduces by 2. qi imples that the ith momentum pi is conserved (i. The last theorem in its most general form ˙ is called Noether’s theorem [1] which establishes a onetoone equivalence between the existence of symmetries and conservation laws. For the special case of a simple mechanical systems that its Lagrangian is symmetric w. 2. the symmetry in mechanics is deﬁned as the invariance of the Lagrangian of a system under the action of a left (or right) invariant Lie group. We say the Lagrangian L(q. . q) is symmetric w.e.t. for deﬁnitions of Lie groups. min (q))T 34 . . Almost all reallife mechanical systems possess certain symmetry properties.t. the connection between symmetry and conservation laws (i. Formally. For a complete treatment of symmetry see [54] (also [1]). the existence of symmetry properties leads to reduction of the mechanical systems.[62] and conﬁguration stabilization of a nonholonomic underwater vehicle [27].r. we only consider symmetry of Lagrangian systems in a Euclidean space ˙ Q = Rn . . . and the action of Lie groups on manifolds see Warner and Boothby. For example. .r. Here.e. translation. . pi = 0) and vice versa. . we deﬁne symmetry in mechanics. invariance of the Lagrangian) w.r. Noether’s theorem).
we will show that the existence of kinetic symmetry properties leads to reduction of underactuated Lagrangian systems regardless of the lack of conserved quantities. By Kinetic Symmetry. Therefore.t. we mean the kinetic energy of the system is invariant w. i = 1.t.r. In this thesis. q1 are deﬁned as the following ˙ ˙ ˙ ˙ π1 := m−1 (q)p1 = q1 + m−1 (q)m12 (q)q2 . 2.e. However. normalized momentums are the most important analytical tools in reduction of underactuated systems. later. 2 ˙ ˙ ∂qi The normalized momentums conjugate to q1 and q2 w. We say π = π(q. the analysis of the system reduces to the analysis of a mechanical system with a ﬁrstorder nonholonomic constraint.e. qi .is the ith row of the inertia matrix M (q). q) is an integrable normalized momentum. If this constraint is nonintegrable. q2 ) and Lagrangian 1 L(q. ∂K =0 ∂qi The kinetic symmetry is equivqlent to classical symmetry in lack of potential energy. if there ˙ ˙ exists a generalized conﬁguration function h = h(q) such that h = π where the 35 . we will see that the fact that the generalized momentums are not conserved has a crucial role in controllability of large classes of underactuated mechanical systems. we use a diﬀerent notion of symmetry called Kinetic Symmetry that is rather similar to classical symmetry. in addition to the conservation laws. i. the existence of symmetries gives rise to holonomic/nonholonomic velocity constraints for mechanical systems.8 Normalized Momentums and Integrability In this work. It is important to notice that the existence of kinetic symmetries (in presence of a potential ﬁeld) does not lead to the existence of conserved quantities. the majority of systems that we consider here have nonzero (i. nonconstant) potential energies and the notions of kinetic symmetry and classical symmetry do not coincide. In addition.r. Consider an underactuated system with conﬁguration vector q = (q1 . q) = ˙ 2 q1 ˙ q2 ˙ T m11 (q) m12 (q) m21 (q) m22 (q) q1 ˙ q2 ˙ − V (q) The generalized momentum conjugate to qi can obtained via partial Lengendre transform ∂L pi := = mi1 (q)q1 + mi2 (q)q2 . π2 := m−1 (q)p2 = q1 + m−1 (q)m22 (q)q2 21 21 11 11 where π1 is always welldeﬁned and π2 is deﬁned wherever m21 (q) is an invertible square matrix.
Whenever h : h = π. ˙ h(q. q) = ˙ 36 . we say π is nonintegrable.˙ function h is deﬁned as h(q) · q ˙ ˙ We call h(q) the integral of π.
The structural properties of the new normal forms allow eﬀective use of many existing control design methods. spacecraft.g. 101. 57]. 80. 101. aircraft.1 Introduction Underactuated mechanical systems appear in a variety of reallife control applications including Robotics (e.Chapter 3 Normal Forms for Underactuated Systems 3. helicopters. Underactuated systems are not fully feedback linearizable. highgain/lowgain designs [80. The restriction of the control authority for underactuated systems causes major diﬃculties in control design for these systems. and feedforward forms (deﬁnition 3.14)) [36]. and satellites). During the past decade. In this chapter. and Surface Vessels.g. control of autonumous vehicles. Due to their broad range of applications.8. and walking robots). The classes of nonlinear systems of interest include ByrnesIsidori normal forms (equation (3. control of underactuated systems is extremely important. The reasons behind the complexity of control design for underactuated systems can be summarized as follows. lightweight ﬂexiblelink robots. 56].3) [102. 87]. nonholonomic mobile robots.2) [36. This is due to the fact that a method for transforming underactuated systems into cascade nonlinear systems with upper/lower triangular or nontriangular structural properties has not yet been discovered. and sliding mode control [40] are not directly applicable to underactuated systems with the exception of a few special cases (e. Control of general underactuated mechanical systems is currently considered a major open problem based on recent surveys [78. and control of ﬂexible structures. robotics and automation. 102]. strict feedback forms (deﬁnition 3. our main contribution is to obtain the aforementioned cascade normal forms for underactuated systems by applying global/semiglobal nonlinear change of coordinates. many recent and traditional methods of nonlinear control design including backstepping [36. forwarding [56. 79].g. the beamandball system and the cartpole system). a new class of underac 37 . Aerospace Vehicles (e. In addition. 57].8. the challenging nature of analysis and control of underactuated systems has attracted numerous researchers with interests in nonlinear control theory. Moreover. Underwater Vehicles.
In addition. a helicopter with 6 DOF and 4 controls). We call this procedure noncollocated partial feedback linearization. we focus on the class of underactuated systems with two degrees of freedom and kinetic symmetry (deﬁned in section 2. We prove that for this special class of underactuated systems both the decoupling change of coordinates and the corresponding cascade normal form can be obtained in explicit forms. This global change of coordinates transforms the dynamics of the system into a cascade nonlinear system in ByrnesIsidori normal form with a doubleintegrator linear part [36]. the input highly complicates control design for underactuated systems. This new partial linearization method is used in obtaining the nontriangular quadratic normal form for a class of underactuated systems with kinetic symmetry. and the TORA system (see section 3. We introduce a procedure to linearize the dynamics of the unactuated conﬁguration variables using a change of control. In chapter 7. 99]. Due to the nontriangular nature of this normal form.tuated systems is found that leads to a cascade normal form with a nontriangular quadratic structure (see deﬁnition 3. backstepping/forwarding procedures [80] are not applicable to it. As an initial step. 98.t.3). The structure of the obtained normal forms for underactuated systems with two degrees of freedom naturally leads to their classiﬁcation in two classes. Spong showed that the dynamics of the actuated conﬁguration variables of any underactuated mechanical system can be globally linearized by applying a change of control. and the BeamandBall system (see section 3.8.r. the CartPole system.7).r. 38 . the new control. To resolve this issue. the cartpole system can be transformed into a feedforward form after another transformation (deﬁnition 3. we address important special cases of this stabilization problem including the cases that appear in control of underactuated mechanical systems. For doing so.5).8. However.5). In addition. we introduce an appropriate structure for a global change of coordinates that decouples the linear actuated and nonlinear unactuated subsystems w. ﬁrst it is very useful to consider calculation of cascade normal forms for lowerorder underactuated systems. To attain insight in dealing with control design for higherorder underactuated systems (e. This coupling w.2). The second class includes the Pendubot. characterization of the cases where this normal form reduces to (strict) feedback/feedforward forms is provided. the new control appears both in the actuated subsystem and the unactuated subsystem of the original system. our main result on normal forms for underactuated systems partly relays on a method called collocated partial feedback linearization due to Spong [90]. Stabilization of this class of nontriangular nonlinear systems is equivalent to control of a special ByrnesIsidori normal form with a doubleintegrator linear part that is a major open problem in nonlinear control theory [36. This allowed application of passivitybased/energybased methods to solving swingup problem for special examples of underactuated systems that involve inverted pendulums.g. The ﬁrst class includes systems like the Acrobot.3 for the deﬁnitions) which can be transformed into cascade nonlinear systems in strict feedback form (deﬁnition 3. the Inertia Wheel Pendulum. the Rotating Pendulum.8. after this change of control.8.3 for the deﬁnitions) which can be transformed into a nontriangular quadratic normal form (deﬁnition 3. This is applicable under the condition that the number of unactuated variables is less than or equal to the number of actuated variables.t.
the InertiaWheel Pendulum. we present several examples of underactuated systems. the TORA (translational rotational actuator) system. it is still possible to partially linearize the system such that the dynamics of q2 transforms into a double integrator [90]. In section 3. These examples include the Acrobot. we explain collocated and noncollocated partial feedback linearization methods. Due to the lack of control in the ﬁrst equation of (3. Then.2 Dynamics of Underactuated Systems The EulerLagrange equations of motion for an underactuated system are as the following d ∂L ∂L − = F (q)τ dt ∂ q ˙ ∂q where τ ∈ Rm is the control and F (q) ∈ Rn×m is a nonsquare matrix of external forces with m < n and full column rank. and the VTOL aircraft (vertical takeoﬀ and landing). the Pendubot. the conﬁguration vector can be partitioned as q = (q1 . We brieﬂy introduce each example with its related control design task and 39 .4 and 3. the Rotating Pendulum. q) = 0 τ (3. We demonstrate this procedure in sections 3.2). Here. the BeamandBall system. For the case of simple Lagrangian systems.The outline of this chapter is as follows. we provide several examples of underactuated systems including both benchmark systems and reallife applications in section 3. q) ˙ ˙ h2 (q.9.3.1) Assuming F (q) = [0. All the examples are chosen due to the complexity of their control design and the fact that for analysis and control purposes they are of high interest in the literature. After partitioning the inertia matrix M (q) accordingly. the CartPole system. First. q)q + G(q) = F (q)τ q ˙ ˙ (3.2. 3. Im ]T .4 and 3. Our main results on normal forms for underactuated systems are presented in section 3. we present standard models for underactuated systems in section 3.7. the dynamics of an underactuated system takes the form m11 (q) m12 (q) m21 (q) m22 (q) q1 ¨ q2 ¨ + h1 (q. respectively [90]. the equations of motion for an underactuated mechanical system can be expressed as the following M (q)¨ + C(q. m denotes the number of control inputs that is less than the number of conﬁguration variables n.5.2) with τ ∈ Rm . 3. q2 ) ∈ R(n−m) × Rm according to F (q) where q1 and q2 denote the actuated and unactuated conﬁguration vectors. However. In sections 3.3 Examples of Underactuated Systems In this section.5. we focus on normal forms and classiﬁcation of underactuated systems with two degrees of freedom and kinetic symmetry. it is not possible to fully linearize this underactuated system using a change of control.
Though. (b) The Pendubot. we give examples of higherorder underactuated systems that appear in Robotics and Aerospace applications. Acrobot and the Pendubot graphically seem to be very similar (i. The inertia matrix for each system is given for the future use.e. A more complicated task is to stabilize the Acrobot or the Pendubot to any of their arbitrary inﬁnite equilibrium points.1 (a).3.1 The Acrobot and the Pendubot The Acrobot [13. In this work.later use them as applications of our theoretical results by providing control design and analysis for each case in chapter 5. The inertia matrix for both the Acrobot and the Pendubot is the same and has the following form M (q2 ) = 1 a1 + a2 cos(q2 ) a3 + 2 a2 cos(q2 ) a3 + 1 a2 cos(q2 ) a4 2 where the ai ’s are positive constants.3. Later in chapter 5. 86. 94.3. the former task has been done using swinging up these double inverted pendulums from their downward initial positions and bringing them close to their upright position. normal form) and control design. 28]. 86] is a twolink planar robot with revolute joints and one actuator at the elbow as shown in Figure 3. The examples of this section only include underactuated systems with two or three degrees of freedom. The Pendubot shown in Figure 3.1 (b) is also a twolink planar robot with revolute joints and an actuator at the shoulder. the share the exact same inertia matrix). The y q 2 y q m2 I2 2 m2 I2 q 1 q L1 l1 m1 I 1 x 1 L1 l1 m1 I 1 x (a) (b) Figure 31: (a) The Acrobot. 3. The latter task has never been done before to the best of our knowledge. A possible control task is to stabilize the upright equilibrium point for the Acrobot or the Pendubot from any initial condition. then switching to a linear controller around the upright equilibrium point [13. later we will see that the diﬀerence in the location of their single actuator causes a major diﬀerence in their standard representation (i. In the past.e. 90. we perform both of 40 . The stateoftheart of research in control design for each example is provided as well.
6] which is an inverted pendulum on a rotating arm. I2 q1 q2 m1. In addition. ci > 0 are constants.3.2 The CartPole System and the Rotating Pendulum The CartPole system shown in Figure 3. we introduce aggressive swingup control designs for both the CartPole system and the Rotating Pendulum using bounded control inputs. 41 .2 (a) consists of an inverted pendulum on a cart. In this work. Clearly. (b) The Rotating Pendulum. Another similarity between the CartPole system and the Rotating Pendulum is that both have the same form of the potential energy V (q2 ) = a4 cos(q2 ) with a4 > 0. M (q2 ) = a1 a2 cos(q2 ) a3 a2 cos(q2 ) and for the Rotating Pendulum the inertia matrix is M (q2 ) = b1 + c1 sin2 (q2 ) b2 cos(q2 ) b2 cos(q2 ) b3 where all the ai . Swingup control design for the inverted pendulum in the CartPole system [21.3. Figure 3. l2. the only diﬀerence between the inertia matrices of these two mechanical systems is in the ﬁrst element m11 (q2 ). l1.3. bi . 6] has been done by several researchers. The drawback of the existing swingup design methods is that they are relatively slow.2 (b) shows the Rotating Pendulum [112. J1 m2. 3. J2 y F q1 x m1 (a) (b) Figure 32: (a) The CartPole system. The inertia matrix for the CartPole system is z q2 m2. 90] and the Rotating pendulum [112.these tasks for the Acrobot and the former task for the Pendubot using a single state feedback. we address asymptotic stabilization of both of these systems to their upright equilibrium points using nonlinear state feedback. l2 .
stabilizam . global stabilization of the beamandball system with friction was achieved using a numeric version of the method of construction of Lyapunov functions with crossterms which is originally due to Mazenc and Praly [56]. 80. Semiglobal stabilization of the beamandball system using small nested saturations (i. state feedback) was addressed by Teel in [102]. 79]. In this work. Moreover. tracking for the beamandball system was considered using approximate inputoutput linearization. The task is to bring the ball to the center of the beam by applying a torque τ to the beam. 103.8.3. In all the aforementioned works.3. The vertical distance between the center of mass of the ball and the location of the external torque (i. 102. In [80].3 The BeamandBall System The BeamandBall system illustrated in Figure 33 consists of a beam and a ball on it. stabilization of this system using output feedback is due to Teel and Praly [103]. tion and tracking for the beamandball system using state or output feedback has been considered by many researchers [34. Figure 34 illustrates the TORA system consisting of a translational oscillating platform with mass m1 that is controlled via a rotational eccentric mass 42 . the model of the beamandball system is taken from [34] with the assumption that d = 0.e. I2 q2 d q1 r τ I1 Figure 33: The BeamandBall System.e. 3.3. control) is shown by d. Due to complexity of this system. Also. In [34]. The inertia matrix for the beam and ball system takes the following form M (q2 ) = 2 a1 + a2 q2 −a3 d −a3 d a4 where all the ai ’s are positive constants. global stabilization of the beamandball system as a perturbed chain of integrators with nontriangular (deﬁnition 3.4 The TORA System The TORA (Translational Oscillator with Rotational Actuator) system was ﬁrst introduced in [109].4) homogeneous higherorder perturbations was achieved in [79]. we consider aggressive stabilization of the beamandball system with d = 0 in chapter 7 and introduce the challenges in dealing with the case d = 0.
m2 . in [93]. i. we are interested in global stabilization of the TORA system in the presence of gravity. The pendulum is unactuated and the system has to be controlled via the rotating wheel. Here. g = 0. g = 0. it received a tremendous attention by several researchers [37. 43 . change of coordinates was available to transform it to a cascade nonlinear system [109]. 3. Here.e.e. In [93]. This is due to the constant inertia matrix of the InertiaWheel Pendulum. 38. 80. an energybased method is used for the swingup of the pendulum and then a supervisorybased switching strategy is employed to switch to a stabilizing local nonlinear controller with a relatively large region of attraction. i. The speciﬁc angle of rotation of the wheel is not important. The task is to stabilize the pendulum in its upright equilibrium point while the wheel stops rotating.5 The InertiaWheel Pendulum The InertiaWheel Pendulum was ﬁrst introduced by Spong et al.3. Figure 35 shows the inertia wheel pendulum that consists of a pendulum with a rotating uniform inertiawheel at its end. q2 )) = k1 q1 + m2 gr cos(q2 ) 2 Due to the fact that the TORA system is not fully feedback linearizable an an easy m1 k1 τ q r 2 I2 m2 q1 Figure 34: The TORA system. The inertia matrix for the system is in the form M (q2 ) = a1 a2 cos(q2 ) a2 cos(q2 ) a3 where the ai > 0 are constants and the potential energy is given by 1 2 V (q1 . we achieve global stabilization of the upright equilibrium point of the inertia wheel pendulum using a single nonlinear state feedback in analytically explicit form. The inertia wheel pendulum is the ﬁrst example of a ﬂat underactuated mechanical system with two degrees of freedom and a single actuator. In most of these works the TORA system was used as a benchmark example for passivitybased techniques and stabilization/tracking using output feedback in zero gravity. 39].
The dynamics of the VTOL aircraft is u1 ε u2 θ u2 y x Figure 36: The VTOL aircraft. bounded tracking is achieved. 55] as the following x1 ˙ x2 ˙ y1 ˙ y2 ˙ ˙ θ ω ˙ = = = = = = x2 −u1 sin(θ) + u2 cos(θ) y2 u1 cos(θ) + u2 sin(θ) − g ω u2 (3.6 The VTOL Aircraft The VTOL aircraft depicted in Figure 36 is a simpliﬁed planar model of a real vertical take oﬀ and landing plane (e.3) The VTOL aircraft is an underactuated system with three degrees of freedom and two control inputs. In section 5.g. given in [35. I 1 Figure 35: The Inertia Wheel Pendulum 3. it is assumed that   is relatively small and the VTOL aircraft is treated as a slightly nonminimum phase system. 44 . using approximate inputoutput linearization. the Harrier).e. Then. we discuss global conﬁguration stabilization for the VTOL aircraft with strong input coupling (i.10. In [35].3.τ q2 I2 q1 L 1.
This is one of the main sources of the complexity of control design for underactuated 45 . the system is strongly nonminimum phase and stabilization or tracking for system (3.2). deﬁning α(q) = m22 (q) − m21 (q)m−1 (q)m12 (q) 11 β(q. we restate this result as the following with our own phrasing due to applications in this work.1. Substituting this in the second line of (3. we get (m22 (q) − m21 (q)m−1 (q)m12 (q))¨2 + h2 (q.2).4) where α(q) is an m × m positive deﬁnite symmetric matrix and g0 (q) = −m−1 (q)m12 (q) 11 Proof.3) is a challenging problem.2) as the following q1 ˙ p1 ˙ q2 ˙ p2 ˙ = = = = p1 f0 (q. q) ˙ ˙ ˙ 11 and observing that α(q) is positive deﬁnite and symmetric ﬁnishes the proof. Here. The procedure of linearization of the dynamics of q2 is called collocated partial linearization which is due to Spong [90].e. p1 )subsystem (i.2) is q2 . In this case. q) − m−1 (q)m12 (q)¨2 ¨ ˙ q 11 11 which proves the expression for g0 (q). 1). nonlinear subsystem) and (q2 .4 Collocated Partial Feedback Linearization The actuated conﬁguration vector for the underactuated mechanical system in (3. linear subsystem). we have q1 = −m−1 (q)h1 (q. q) = h2 (q.4) is that after partial feedback linearization.e. the new control u appears in the dynamics of both (q1 . From the ﬁrst line of (3.2) can be globally partially linearized using a change of control. (Spong [90]) There exists a global invertible change of control in the form τ = α(q)u + β(q. q) = τ q ˙ ˙ 11 11 Thus.4. p2 )subsystem (i. q) − m21 (q)m−1 (q)h1 (q. Spong has shown that all underactuated systems in the form (3. Proposition 3. q) − m−1 (q)h1 (q. The main property of the underactuated system in (3. p) + g0 (q)u p2 u (3. q) ˙ that partiallylinearizes the dynamics of (3. 3.
τ1 ).5.7. Remark 3. p) + g0 (q)u0 + g2 (q)u2 p2 u2 (3. q) ˙ that partially linearizes the dynamics of (3. In [78]. we introduce a global change of coordinates that decouples these two subsystems while leaves the linear subsystem invariant.systems. Thus.1.4. u = col(u0 . using explicit expressions of f (x) and g(x).5) as the following q0 ˙ p0 ˙ q1 ˙ p1 ˙ q2 ˙ p2 ˙ = = = = = = p0 u0 p1 f0 (q.4) can be rewritten as x = f (x) + g(x)u ˙ with obvious deﬁnitions of f (x) and g(x). We call this procedure noncollocated partial feedback linearization. there exists a change of control in the form τ = α1 (q)u + β1 (q.5.5) 0 h2 (q. q) = τ1 ˙ ¨ (3. q2 ). 3. For the special case where there are equal number of actuated and unactuated variables n = 2m and n0 = 0.1. We show this is possible if the number of the control inputs is greater or equal to the number of unactuated conﬁguration variables. consider the following underactuated system τ0 h0 (q. In section 3. u1 ) and u1 = α0 (q)u0 + α2 (q)u2 + β2 (q.1. the dynamics of (3.5). q) ˙ where τ = col(τ0 .5 Noncollocated Partial Feedback Linearization In this section. Then. q2 . Deﬁning x = (q1 . p2 ). q) ˙ q0 ¨ m00 (q) m01 (q) m02 (q) m10 (q) m11 (q) m12 (q) q1 + h1 (q. q1 . Consider the underactuated mechanical system in (3. q) ˙ q2 ¨ m20 (q) m21 (q) m22 (q) where q = (q0 . q0 has dimension zero or q = (q1 . Remark 3. p1 . suﬃcient conditions are provided for smalltime local controllability (STLC) [97] of a special class of underactuated mechanical systems in the form (3. we present a partial feedback linearization procedure for underactuated systems that linearizes the dynamics of the unactuated conﬁguration variables.6) 46 . q2 ) ∈ Rn0 × Rn1 × Rn2 with n1 = n2 = m and n0 = n − 2m ≥ 0. To be more precise. Proposition 3.4).
By input coupling. after a second change of control as u1 = −m−1 h2 (q. p) = −m−1 (q)h2 (q. Apparently. q) − m−1 (q)m20 (q)u0 − m−1 (q)m22 (q)u2 ˙ 21 21 21 where u2 is the new control. we mean F1 (q) ≡ 0 for all q.5). we get q1 = −m−1 (q)h2 (q. Consider the following underactuated system with input coupling 47 . p) 21 g0 (q) = −m−1 (q)m20 (q) 21 g2 (q) = −m−1 (q)m22 (q) 21 Proof.1) the dynamics of the actuated conﬁguration vector (q0 . q1 ) can be linearized using a change of control τ = α1 (q)u + β1 (q. q) that gives ˙ ¨ q0 = u 0 q1 = u 1 ¨ Over the set U . the result follows. Proposition 3. Without loss of generality.6. m21 (q) is an invertible matrix and from the last equation in (3. we provide the conditions which allow partial feedback linearization for an underactuated (simple) Lagrangian system. 3. q) − m−1 (q)m20 (q)¨0 − m−1 (q)m22 (q)¨2 ¨ ˙ q q 21 21 21 Thus.6 Partial Feedback Linearization Under Input Coupling d ∂L ∂L − = F (q)τ dt ∂ q ˙ ∂q Consider the underactuated mechanical control system with conﬁguration q ∈ Rn and control τ ∈ Rm and rank F (q)