Professional Documents
Culture Documents
Application of H Infinity Theory To Robot Manipulator Control PDF
Application of H Infinity Theory To Robot Manipulator Control PDF
-
%ion 4 then highlights the methodology behind H-infinily
control. Last an example control design is presented along
This paper investigates the feasibility of H-infinity control with simulation results in &ion 5.
theoly for robol manipulator control. Given a model of
manipulator dynamics, a control methodology is developed
which utilize an inner loop linearizing, decoupling feedback
control and an H-infinity outer loop compensator. A The robol manipulator is modeled as a set of n rigid bodies
methodology for handling model simplifications in the connected in series. The equations of motion in joint space
linearizing feecback loop is also developed. An example are gven by
-
control design is presented along with simulation results.
M(q) 6 + qq,$ + G(q) = 7 (1 1
Robust tracking of trajectories is shown given bounds on
model parameters. where t k !!?? n x 1 joint torque vector (supplied by actuators
at each joint) and q, q and q are the joint positions,velocities
and accelerations, respectively. Also, M(q) is the n x n
inertia matrix (symmetric, positive definite), C(q,$
Since the development of state space solutions to Hinfinity represents the n x 1 centrifugal and Coridis terms vector and
control problems, Hinfinity has become a powerfultechnique G(q) is the gavity vector. Note (1) is a set of hi@ly coupled,
for multivariable control design. This power hinges on three nonlinear equations. Given this set of n equations the
issues: 1) it is a multivariable technique, 2) both performance control problem is to determine the necessary joint torques
and stability robustness can be addressed and 3) it is aMe to to follow a desired trajectory.
drectly handle uncertainty. However for the practice of H-
infinity controller design to permeate into the practicing
control community's box of tools, more experience and
insight into the methodology is needed. This papec attempts 4lwzEmd-
to address this issue by applying H-infinitycontrol theory to a Since Hinfinity control is a linear control technique the plant
robot manipulator as well as demonstrate the feasibility of H- equations of mdion must be linearizable over a reasonable
infinity theory for manipulator control. regon of the joint space or else a linearization scheme must
be applied to the plant in real time. Due to the nature of ( l ) ,
Control of robot manipulators is a challenging control task regional linearization of the EOM is nol possible so the well-
because of the nonlinear, coupled nature within the known technique of nonlinear feedback linearization is
equations of motion. This leads to uncertainly in the robol utilized. As will become apparent later, the nonlinear
model. The ability to accurately assess these uncertainties feedback forms an inner loop which produces linearized
and to push the bandwidhs of controllers (while maintaining behavior of the plant and inner loop combination [l]. The H-
stability w.r.1. the uncertainties) is very important for robol infinity controller is then applied to this plant.
performance. Because of H-infinity's ability to incorporate
uncertainty into the controller desi@, this is a good The linearizing feedback control essentially subtracts off the
framework to improve robot performance and to gain nonlinear terms in (i)and thus takes the form
experience with H-infinity techniques.
7 = m(q) uc + cced + 9(q) (2)
The contents of this paper are as follaws. Sedion 2 where U- becomes a new input and m(q), c(q,q) and as)
presents the structure of robot dynamic equations. Next, in represent approximations to the adual manipulator EOM.
section 3, the equations of motion are linearized using Inserting(2) into (1) yields
feedbadc linearization (or inverse dynamics). A technique for
representing uncertainty related to modeling errors in the
feedback linearization loop is also developed in this section.
149
This forms the model used in the desi@ of the H-infinily Kinfinity controller design requires converting the block
controller. dagam in figure 3 to a linear fractional famat. The
pciblem formulation then becomes a l i a r system in the
form
40 H-mHTY-IOGY x(t) = A ~ ( t+) 61 w(t) + 62 ~ ( t )
In this section the imporlant aspects of the H-infinily theory
z(t) = C1 X(t) + D11 W(t) + D l 2 ~ ( t ) (1 0)
and deslgn methodology are presented. No theory is
fit) = C2 X(t) + D21 W(t) + D22 U(t)
developed here as it is weH documented in the literature
(References [2], [3]and [4] are good places to start). where
w(t) E p1 is the external inpuUdsturbancevector
Technically, H-infinity encompasses 4 steps as follows: u(t) E flis the contrd input vector'
z(t) E RP1 is the vector d all signals used to characterize
slep 1: Develop an appropriate optimization format dosed loop system behavior
(using the linear fractional interconnection fit) E Fi@ is the measured (sensed) output vector
structure. x(t) E R" is the state vector
slep 2: Form weighting functions to meel desired
specifications (typically on S and T). The frequency dependent weights are included in this model.
slep 3: Design an H-infinity controller using the Glover A linear fractional representation Mock dagam is shown in
/ Doyle algorithm that meets the design figure 4. In the figure,
constraints.
Extmd Contrdld
slep 1. Verify the design using non-linear simulation.
Because of the nature of the H-infinity norm, plant weightings figure 4 Hinfinily Contrd Problem
must weight plant parameters that are desired to be kept
small. Wt= leads to weightings on the closed loop sensitivity
matrix, S, for low frequency weighting and on the dosed loop (1 1)
transfer function, T, for high frequency control. Figure 3
shows a generic block dagam set up for mixed performance With a linear controller K(s) connecting y to U, the closed
/ robustness oplimization. toop transfer fundion from w to z is @en by
OD
a
II FI(P,K) llinf = SUP [ FI(P,K) ] (13)
The solution to minimizing the M nom in the above
equation requires sdving two algebak Rsccati equations,
each of &gee n [2]. l k fallowing assum@ions are
required on the plant, P
L =G(s) K(s)
e = I I(I+L)w = S w i) (A,Bl) and (A,=) are SlabilitaMe
y =( I - S)w;: L/(I+L) W E T w ii) (C1,A) and (C2,A)are detectable
Ws(r)rwsetrsarrpmasnnvlly IUMlCn. S iii) D12TD12 = I
W s ) =wa@WQdn c o n p ( a " t w y 9dnlnnPj hlndicn. T iv) M 1 MlT=I
In most designs, transformations on the plant are required in
figure 3 W-lnfinily Mixed Performance / Robustness Block order to meel assumptions fii an8 iv. These transformations
I3iagam are detailed in [q. The H-infinity algorithm involves solving
,
two Ricalti equations and is detailed in References[2], [3] or These values are based on rough estimates and, while
141. hopefully realistic, are not intended to represent adual
estimates of uncertainty for a PUMA 560 robot arm. The
mass matrix p r o b d is assumed to be known Within 1o%,ie.,
5 - m
In this section the H-infinity theory of Sedion 4.0 is applied to
a two-link manipulator. The linearized robot model is of the
form developed in Sections 2.0 and 3.0. For a block diagram of the plant with uncertainty refer to
figure 2.
For a two link manipulator,the equations of motion are
For design of the H-infinity controller, the steps in section 4.0
71 =M11 91 + M12Q+C1 +G1 will be fdlowed. The first step in the process is lo develop
9 = = 2 1 9 1 +M2292+C2+G2 the appropriate optimization strudure. As mentioned before
where the mixed performance / sensitivity format is utilized.
M i l = 3.82 +2.12 COS(^^)
M i 2 = M21 = 0.71 + 1.06 Cm(q2) Slep 2 requires determining appropriate weighting functions
M z = 0.71
on the closed loop sensitivity function, S and the closed loop
transfer fundion, T. In this design, desirable closed loop
frequency response was chosen as having a c r m v e r of
N1 =-2.12sin (q2) ( q t & + & / 2 )
25-30 radsec, integral action at low frequency and sharp
N2 = - 2.12 sin (92) $1 / 2
rolldf ( 4 0 dbldecade) above the crossover. This produced
weighting functions for bdh S and T as
G1 = 81.82 COS(q1)+ 24.06 COS(q1+ 92)
G2 = 24.06 coS(q1 + 92) s+30
ws = .001/30 s+.oo1
These etpations correspond to a PUMA 560 robot arm and
are taken from [5] for comparison purposes. Figure 5 shows 8+30
d picture of the manipulator. WT = 8OOMO * 9+800
The feedback linearization uncertainty matrix is given by With the reduced order controller, a simulation of the
= ma& = m12 "13
Ah
,'
m14
Ah'24 1 uncertain system to Ihe commandedinput
q l = 90 [ 3 exp(-V0.3) - 4 exp(-V0.4) 1 deg
s + 30
Ah'f = -
s+5m
151
-
H-infMy was shawn as a good framework for desigring
robot controllers. The ability to expliciUy model uncertainties
resuRing from feedback linearization make the procebre
especiatly usdul. Certainly some experimentation and
1.
2.
d-H
2
Craig, John J.,
aod,
. .& w e A.,
R
-
- Mechanff
Addson-Wesley Publishing Company, 1989.
--. . . .
.. . .. . . ....
I 0-a
10-1 10' 10'
Frequency
100
- Rad/Sec 103
k0
20
0
-eo
-100
-12ot
I__---_.-
' '''"b' ' " ""'
.
'
, .
'
.
".'''''' "
Figwe 7 Plot of S and 7 for flan1 and Inner Loop (hshedklne - S, Solid Line - T)
152
.- . ..
.. . ..
n
>
Frequency - Ra3/Sec ..
n -:0
f
-51
-151
'
. . . . . ....
. .
.....
. . . ..... . . . . .
........
-20r
-2s
-30
lo-' 10" 100 10' 10' 10'
Freauency -
Raa/Sec
figure 9 Plol of Closed Loop Syslem, T, and WT'' (Dashed Llne - WT-' , Solid Line - T)
Time (sec)
153