See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/265348103

DYNAMIC MODELLING AND CONFIGURATION

STABILIZATION FOR AN X4-FLYER

Article · July 2002

DOI: 10.3182/20020721-6-ES-1901.00848

CITATIONS READS

217 482

4 authors, including:

T. Hamel Robert E. Mahony

University of Nice Sophia Antipolis Australian National University
166 PUBLICATIONS 5,350 CITATIONS 246 PUBLICATIONS 8,063 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Aircraft Flight Control View project

PhD: Real-time Visual Flow Algorithms for Robotic Applications View project

All content following this page was uploaded by Robert E. Mahony on 09 November 2017.

The user has requested enhancement of the downloaded file.

Copyright © 2002 IFAC
15th Triennial World Congress, Barcelona, Spain

DYNAMIC MODELLING AND CONFIGURATION

STABILIZATION FOR AN X4-FLYER.



Tarek Hamel Robert Mahony



Rogelio Lozano
James Ostrowski

CEMIF FRE-CNRS 2494, Univ. d’Evry, 40 rue du

Pelvoux, Evry, France
Dep.



of Eng., Australian Nat. Uni., ACT, 0200 Australia
Heudiasyc, UTC UMR-CNRS 6599, BP 60205



Compiègne, France
GRASP Laboratory, University of Pennsylvania, 3401
Walnut Street, PA 19104-6228, Philadelphia

Abstract: A model for the dynamics of a four rotor vertical take-off and landing (VTOL)
vehicle known as an X4-flyer is proposed. The model incorporates the airframe and
motor dynamics as well as aerodynamic and gyroscopic effects due to the rotors for
quasi-stationary flight conditions. A novel control strategy is proposed for configuration
stabilization of quasi-stationary flight conditions. The approach taken involves separating
the rigid body (airframe) dynamics from the motor dynamics, developing separate control
Lyapunov functions for the coupled systems and then bounding the perturbation error due
to the interaction to obtain strong practical stability of the complete system.

Keywords: Mobile Robots, Unmanned Aerial Vehicle (UAV), control Lyapunov function.

1. INTRODUCTION as well as aerodynamic and gyroscopic effects due to

Recent advances in computer and sensing technology,
and the associated reduction in cost of such systems,
have made physical construction of autonomous mo-
bile robotic systems possible at a reasonable price. Au-
tonomous robotic cars and trucks have been under inves-
tigation for some years and, as well as the development
of considerable body knowledge, there now several suc-
cessful commercial systems. More recently, interest is
growing in more complicated systems such as submarine
vehicles (Egland et al., 1996) and unmanned aerial vehi-
cles (Koo et al., 1998), because of their commercial pos-
sibilities. Fixed-wing unmanned aircraft are being rou-
tinely used for military and meteorological purposes and
have been in service for years. The vehicle considered in
this paper is an autonomous hovering system, capable of
vertical take-off, landing and quasi-stationary (hover and
near hover) flight conditions. In this paper, we propose
a model for the dynamics of a four rotor vertical take-
off and landing (VTOL) vehicle known as an X4-flyer.
The model includes the airframe and motor dynamics
the rotors. The fixed pitch, rigid rotors and four motor
actuation leads to significant differences in the model
proposed to other VTOL systems previously studied in
the literature (Hauser et al., 1992, Koo et al., 1998).
The control design proposed is based on separating the
rigid body (airframe) dynamics from the motor dynam-
ics, developing separate control Lyapunov functions for
the coupled systems and then bounding the perturbation
error due to the interaction to obtain strong practical
stability of the complete system. The control strategy
is novel in two ways. Firstly, the system dynamics are
controlled in two separate dynamic systems correspond-
ing to the rigid body dynamics and the motor dynamics.
The separate system errors are combined into a single
control Lyapunov function via a transient error bounding
argument. This approach avoids the either the necessity
of including a dynamic extension in the controller design
(Koo et al., 1998, Mahony et al., 1999) or the need
to use approximate linearization, saturated control or
dynamic reduction (high gain) controllers (Frazzoli et
al., 2000, Teel 1996, Sepulchre et al., 1997). The closed
loop system is practically stable for trajectory tracking f2
of the center of mass of the robotic vehicle. The second f1
Q2
novel aspect of the control is the use of a quaternion Q1
representation of the rotation error in order to obtain a f3
a
simple, smooth control design that contains only a single E1
f4
singularity in error space corresponding to an error of
Q3
180 degrees in the rotation. This is a considerable ad-
vance over earlier work by the authors and compliments Q4
recent work by Frazzoli et al.(Frazzoli et al., 2000). The
approach taken is based on earlier work by Egland et e1
al.(Egeland et al., 1996). a a
E 3 E 2
e2
e 3
2. THE X4-FLYER MODEL
 Fig. 1. The four rotors hover system with Force and
The X4-flyer is a system consisting of four individual Torque Control.
electrical fans attached to a rigid cross frame. It is an
omnidirectional (vertical take-off and landing) VTOL Let = %> denote the linear velocity expressed in the
vehicle ideally suited to stationary and quasi-stationary inertial frame and ? %@' denote the angular velocity
flight conditions. Control of an X4-flyer is achieved by of the airframe expressed in the body fixed frame. -FLet
EG-
differential control of the thrust generated by each elec- A
denote the mass of the rigid object and B %DC
tric fan. Up down motion is controlled by collectively denote the constant inertia matrix around the centre of
increasing or decreasing the power of all four motors. mass (expressed in the body fixed frame ' ). Newton’s
Sideways motion is achieved by pitching in the desired equations of motion yield the following dynamic model
direction and increasing collective thrust to overcome for the motion of the airframe:
the tendency of the vehicle to side-slip towards the H
ground (cf. Figure 1). The rolling motion is achieved I = (1)
A H AKJML -+N
by increasing, for example, the power of the left rotor =  . O
P (2)
and decreasing that of the right rotor in proportion to H
.QR. sk  ? S"  (3)
preserve total collective thrust. By the same principle, H
differential control of the forward and rear rotors leads to B ? T ?  BU?  N VXW (4)
control of the pitching motion of the vehicle. The ‘yaw’
control mechanism is more subtle. When a rotor turns, it The notation sk  ? " denotes the skew-symmetric matrix
has to overcome air resistance. The reactive force acts such that sk  ? " =  ?  - = for the vector cross-product
 and any vector = %YC . The vector O%' combines
on the rotor in the direction opposite to the rotation
of the rotor. In the X4-flyer both sets of front-rear and the principal non-conservative forces applied to the X4-
left-right rotors turn in opposite direction (cf. Figure 1). flyer airframe including thrusts (generated by the rotors
However as long as all rotors produce the same torque, cf. Figure 1) and drag terms associated with the rotors
more precisely produce the same reactive torque, which downwash on the airframe. The torque V %I' is derived
is mostly a function of speed of rotation and rotor blade from differential thrust associated with pairs of rotors
pitch, the sum of all air resistances is zero and there along with aerodynamic effects and gyroscopic effects.
is no horizontal rotation. If one set of rotors increase Due to the rigid rotor constraint the dynamics of each
their speed, the induced torque will cause the X4-flyer rotor disk around its axis of rotation can be treated
to rotate in the direction of the induced torque. It is im- as a decoupled system in the generalized variable ZK[
portant to note that because of the “  ” arrangement, this denoting angular velocity of a rotor around its axis. The
operation has no effect on translation in  or  direction. torque exerted by each electrical motor is denoted \F[ .
The effect on up/down motion can be compensated by The motor torque is opposed by an aerodynamic drag
]
reducing the pitch or speed of the other diagonal pair. [ . Newton’s equations are
The following dynamic model of an X4-flyer is pre- H ]
B
^ ZK[  \F[ T [ (5)
sented for the simple case where the rotors are fixed
pitch, rigid rotors and thrust control is obtained through where B_^ is the moment of inertia of a rotor around its
control of the torque to the motors. Let  

 axis.
denote a right-hand inertial frame such that   denotes
the vertical direction downwards into the earth. Let the The lift generated by a rotor in free air may be modelled
vector  !#" denote the position of the centre of as `
,
mass of the airframe in the frame  relative to a fixed L -
[ /aTPb Z [
origin $%& . Let '(

*) +, ) +- )  be a (right-hand)
body fixed frame for the airframe. The orientation of the where b*c4$ is a proportionality constant depending on
rigid body is given by a rotation .0/'213 , where the density of air, the cube of the radius of the rotor
.4%&5768:9;" is an orthogonal rotation matrix. blades, the number of blades, the chord length of the
blades, the lift constant (linking angle of attack of the
< blade airfoil to the lift generated), the drag constant
The authors propose the term ‘X4-flyer’ as a simple, highly descrip- (associated with the airframe) and the geometry of the
tive name that will apply to a wide range of four rotor flying robots.
 ,
wake . For quasi-stationary manoeuvres in free, still air 3. CONTROL DESIGN METHODOLOGY
it is a reasonable assumption that the scalar bdc($ is
indeed a constant. In this section a backstepping control design is provided
for the model Eqn’s 7-11 proposed in the previous sec-
The reactive torque (due to rotor drag) generated by a tion.
rotor in free air may be modelled as
] , Let p~€" be the desired position trajectory. The dy-
[ /eRf Z [
namics associated with tracking such a trajectory fully
The constant f depends once again on the factors men- determine two degrees of freedom (pitch and roll) in the
tioned above for rotor thrust and particularly on the pitch attitude of the airframe. The yaw of the airframe must be
angle of the rotor blades. separately assigned. There is no ‘correct’ way in which
The thrust applied to the X4-flyer airframe is this assignment should be made. In this paper we use the
h ` h classical ‘yaw’, ‘pitch’ and ‘roll’ Euler angles ‚ƒ„…†"
g k k ,
 [ Rb Z [ (6) commonly used in aerodynamic applications (Murray
[ji  [ji  et al., 1994). Although these angles are not globally
Recalling the discussion preceding the model Eqn’s 1- defined they provide a suitable local representation for
4, the aerodynamic torque inputs applied to the X4-flyer all quasi-stationary manoeuvrers undertaken by an X4-
structure using the combination flyer. The yaw angle trajectory is specified directly in
 of, the
- produced forces terms of the angle ‡~€" . The relationship between the
and air resistances are \ 4 \  \  \ " -
 ), ,) ) ) Euler angles used and the rotation matrix is
\ Rlb Z , T Z h
), , , ˆU‰
ˆUŠq‹ŒŽ‹_‰„ˆUŠ Uˆ Œ‹_ŠˆUŒŽ‹‰
ˆUŠ N  ‹ ŒŽ‹Š
 T - T
\ Rlb Z Z ˆU‰
‹Š‹ŒŽ‹_‰„‹_Š N Uˆ ŒŽˆUŠˆUŒŽ‹‰
‹Š ‹ ŒŽˆŠ
_ W
)- , , , , .Q T
\ Rf Z , N Z h T Z  T Z - ‹‰ ‹ŒŽˆU‰ ˆUŒŽˆU‰
) T
where l represents the displacement of the rotors with (13)
respect to the centre of mass of the X-4 flyer. The trajectory tracking control problem considered is:
The final torque contribution to the X4-flyer dynam- Find a smooth static state feedback  \   \ , H  \ -  \ h " de-
ics comes from gyroscopic effects. Each rotor may be
pending only on the measurable states ‡ .* ? " , the
thought of as a rigid disk rotating around the axis L - in
angular velocity of each rotor  ZK[ " and arbitrarily many
the body-fixed-frame with angular velocity ZK[ . The axis
derivatives of the smooth trajectory F~#e€"S‡~€"U" such
of rotation of the rotor is itself moving with the angular
that the tracking error e;€"T‘p~€"S €"T’~€"U" is
velocity of the airframe. This leads to the following
asymptotically stable.
gyroscopic torques applied to the airframe
h Define “
m n L -
T ^ ?  " ZI[ W  /eX;e€"‚Tp~€"
) “ “
[ai  o H
, /e w  = T p~" N  (14)
Based on the above discussion the following model is 
proposed: w
H where  is a positive constant. Let 5  be the first storage
 
I = (7) function for the backstepping procedure. It is chosen for
o
JML - LH g the full linear dynamics Eqn’s 7-8 “
 T A . -
= (8) “
o o
H k k, k k,
5   ”  N ” , W (15)
. R. sk  ? S" 
Q (9)
H m
B ? T ?  B? N N \ W (10) Taking the time derivative of 5  and substituting for Eq.
H , ) )
B ^ ZI[  \p[ Tqf Z [ W
_ (11) 8 yields
“ “ “•
o o
The dynamic equations may be thought of in two parts; l  w  k  k, N w  k , k, N g L - N JML -
5 T w  , –T . T˜p— ~"S
A
firstly the rigid body  dynamics
, - of the airframe Eqn’s 7- l
g
10 with inputs   \  \  \ " and secondly the Eq. 11 (16)
) ) )
that links the motor torque inputs \F[ to the rigid body From the point of view of a classicalg backstepping con-
op™
forces and torques via the mapping
g , , trol design the vectorial term  A " . L - is the virtual
T*brT*bDT*bsT*b Z  Z  input to this stage of the backstepping design. To apply
 , ,
\ , $tlb $ T*lb Z , Z , this approach in full generality it is necessary to dynam-
)  , Rx , g
\ - lbu$vTPlbv$ Z - Z - ically extend the thrust input in order that the vectorial
) w w w w , ,
\ T T Z h Z h virtual control assigned (for Eq. 16) can be cascaded
)
h E h (12) through the attitude dynamics (Mahony et al., 1999 and
It is easily verified thatw the matrix xy%zC defined Frazzoli et al., 2000). In the case of the X4-flyer, the
above is full rank for b_ {l|q
c $ .
š
The following shorthand notation for trigonometric function is used:
}
For a detailed discussion of the aerodynamic model of a helicopter ›_œ+Ÿž8 –¡_¢_£j¤;¥§¦‘¨œ* ž©¢«ªŸ¬£j¤;¥§¦‘­œ+Ÿž8®«¯_¬£j¤M¥§°
rotor the reader is referred to any standard text on helicopter modelling
(cf. for example (Prouty 1995)). A condensed discussion is given in
(Mahony and Hamel 2001).
 “
k k kg k k n k
highly coupled nature of the motor dynamics that gen- Note that the error terms , , ³ and .R ³ T enter bi-
erate both torque and thrust mean that this approach is linearly into the last two terms of this expression.
not recommended. Alternatively, the vectorial input can
g The next stage of the control design involves control-
be split into its magnitude , that is linked directly to n
the motor torques via Eqn’s 11 and 12, and its direction ling the attitude dynamics such that the error  .@³ T "
L
. - , that defines two degrees of freedom in the airframe is minimized. Designing controllers to stabilize attitude
g dynamics has been an awkward problem in recent papers
attitude dynamics Eqn’s 9 and 10. Assigning the thrust
immediately and then controlling the attitude dynamics (Koo et al., 1998, Mahony et al., 1999, Frazzoli et al.,
leads to a design approach similar to those proposed 2000). The key problem comes in finding an elegant
for the VTOL (Teel 1996, Sepulchre et al., 1997). Such method of representing the attitude of the system that
control strategies lead to time scale separation between does not suffer from singularities and leads to a sim-
the attitude and linear dynamics of the airframe dynam- ple control design. In this paper we employ a quater-
ics and require significant control response in the thrust nion representation of the rotation in order to obtain a
g globally defined smooth static control for the attitude
input . Once again the coupled nature of the motor
dynamics indicate that this approach is not advised. The dynamics with a single singularity corresponding to an
approach taken in the present paper discards the con- attitude error of 180 degrees. The attitude deviation . ³ w is
cept of exact linearization or classical backstepping and parameterized by a rotation » ³ around the unit vector ³ .
falls back on a control Lyapunov function design for the Using Rodrigues’ formula one has (Murray et al., 1994)
n w o w ,
full dynamics. A backstepping design for the rigid-body .Q
³  N½¼U¾e¿  » " sk  ³ " N 
»
T&ÀÁ ¼  ³ U" " sk  ³ "
³
dynamics of the airframe is undertaken, however, we
do not attempt to directly cancel the effect of the vir- The quaternions describing the deviation . ³ are given by
tual control error in the rigid-body dynamics within the (Egeland et al., 1996):
backstepping control design itself. Rather, these errors » »
 ¼U¾e¿ ³ w Âà ¼ ³
are left as perturbations to the rigid-body dynamics. The ³ /a ” ³  ³ /eÀÁ ”
errors introduced into the control design are linear in an
error criterion that forms the basis of a second control which are subject to the constraint:
k k,  Ã
, o
Lyapunov design for the motor dynamics Eq. 11. N  (22)
³ ³
Applying classical backstepping g one would assign a •
The deviation matrix . ³ is then defined as follows:
op™ , k k, n
virtual vectorial control for  A " . L - “ Â
³ 4 ³ Ã T
.Q N ”   N ” ÂÃ
sk  Â ³ "
g w  w  N w , w , ³ " ³ ³ ³ (23)
L AKJ‚L - A A
 . - "«~±/e T — ~ N
F  " ,  cq$ n
The attitude control objective is achieved when .Ä ³  .
(17)
g It is easy to see from Eqn’s 22-23 that this is equivalent
Here  . L - "«~ denotes the desired vectorial control in- o
to  $ and ³ à  . Indeed, it may be verified that
put. Taking the norm of the right hand side of Eq. 17 •
“ k n kÅ n n ”‡Ç ” k  k
leads to .q
³ T  tr U‚.‘
³ T " ‚.‘
³ T "U"Æ (24)
g k A˜JML w  w  N w , k ³
A A
~± - T — ~ N
p  " , W (18) Based on this result the attitude control objective used is
The desired rotation matrix .²~ to drive  ³ to zero. Deriving    Âà " yields (Murray et al.,
o
L - g L - 1994, pg. 74) •
.²~ /e g  . "«~ (19) o o
~ H H
 Âà n N  Âà Â
³  ”  ³ sk  ³ "–" ? ³  ³ T ” ³ ?³ (25)
is obtained by solving for SM{ƒ
…|" using Eq. 13 and
subjecth to the constraint given by the specification of
~#€" . where ? ³ defines the error angular velocity

Substituting for “ Eqn’s 18-19 one obtains ?³ R.²~# ? T ? ~" (26)

“ “
o
H w  w , n g and ? ~ represents the desired angular velocity. It is
, qT  T , T L -
A w  ‚.q
³ T " ~.²~
defined by in Appendix A.
o
g L
T A w  ³ . - (20) The virtual control ?É
³ È is defined to ensure the following
storage function decreases,
where • o
Ê k k,
g g g  /e ”
³  T ~(.Q
³ R.P. ~ %&576±´9;" W ³

From the above discussion the dynamics of the first Set

” wpË Âà Â
storage function 5  (Eq.
“ 16) can “ be bounded as ? ³ È T ³ ³
l w  k  k, w , k , k, With the above choice one has•
5 ¶µ T T (21) kg k
l Ê H ¶µ wFË Â Ã , k  k , N
kg k “ kg k “ Âà wpË Â Ì N w w , ³ k k
~ k , kk n k N ³ k , k T ³ ³ ³ ³   N " g ³
N ~
A w  .‘³ T A w 
where the error approximation follows from the error in-
troduced in the definition of Í in Eq. 36 and Ì represents
·
This only one of a number of possibilities for fully determining ¸¶¹ . the final error used in the process of backstepping
The key point is that ¸¶¹ is fully defined by the vectorial constraint o
on ¸¶¹Sº š combined with some additional constraint that fixes the yaw
Ì N ÂÃ Â W
/e ” wpËP? ³ ³ ³ (27)
parameter.
   
to simplify the control design a control input lineariza- where Z Ï 4 Z Ï  eW W  Z h Ï " . This leads directly to
tion of equation Eq. 10 is undertaken. Define H w Ù Ó
p
Ó
  m  ~ ³ T ³  (33)
Î N N Ó
/eT BpÏ ?  BU? BFÏ BpÏ \ (28) and guarantees that the error ³ converges exponentially
) )
~ to zero.
where \ is the desired torque input. Since B is full rank
)
then this is certainly a bijective control input transforma-
~
tion between \ and Î . With this choice Eq. 10 becomes Theorem 3.1. Consider the system dynamics defined by
) H Eqn’s 7-11 along with the control inputs proposed in

Î N the body of the paper. Assume that there are constants
?³  ³ .²~ BpÏ \³  (29) kg k k k
Ú Ú
bÛc$ such that ~ c and ³ à cÜb . Let Ý be an
where • upper bound on desired acceleration
~ H k k
Î Î
³ R.²~ T ? ~" N .¶~
sk  ? ~#". ~ ? ³ &Ð ¿Ñ  \ ³  \ T \ W p— ~ µ Ý
) )
w  w ,
The torque error \ ³ acts as a perturbation error in the Let  be two positive control gains (cf. Eqn’s 14 and
control Lyapunov function for the rigid body dynamics 17). Set
o
and will be used as a basic error signal for design of w w  N w , wpÒ , w , wpË , w , ,
 ²Þ w  b  b  T½ß; J N Ýà" "
the control Lyapunov function for the motor dynamics.
The ‘error’ input Î ³ may be arbitrarily assigned via its
and choose the remaining control gains to satisfy
dependence on Î . J ,
wpË ß; N Ýá"
Set c w , w , , 
•  b
o o wpÒ
Î ” wpË wpË Â Ã -  Âà  wpË Â Ã , Ì N   cq$‡
³ /a ³ ³ T ” ³ sk  ³ "„? ³ T ³ ” ³ ³ ?
A ,
” w Ë, Âà  ” wpË#wpÒ Â Ã , Ì wpÙ w , w , wpË N w , wpÒ J
, ” Ç ” wwpÒ
T c   " N½â  N Ý " 
á
³ ³ T ³ (30) Þ ,
N wFË#wpÒ T ” w " W
Define 5 , the storage function for the attitude deviation
o o Then, for any initial condition such that the initial value
k k, kÌ k,
5 ,  ” N ” W (31)
³ of the Lyapunov function
o ,
k ,|ä J N
Taking the derivative of 5 , , substituting for the deriva- ã
´$"7R5  :$;" N 5 , ´$;" N
kÓ
w  w  N
Ý
w ,
” ³ ´$;" 
tive of Ì and taking care to identify all terms that are  "
due to the error approximation made in Eqn’s 36 and 29 (34)
leads to the Lyapunov function is bounded for all time
H ,
wpË Â Ã , k  k , wpÒ Â Ã , k Ì k , ä J N
5 | , µ T ³ ³ T ³
ã
e€" w  w  N
Ý
w ,
kg k k  k  "
w w , ³ k Âà kak Ì k k k B Ï k Ì k{k k
N   N " g  ³ N N \³
³ ³ " ” wpË and is asymtotically stable. The tracking error ;e€"XT
~
(32) F~#e€"U"
is locally exponentially stable.

This completes the control design for the attitude dy- Proof 3.2. Let “ “ •
namics. Observe that all the error terms in this expres- L k k k k k k kÌ k kg k k k
4   ,   ³  \³ "
sion are bilinear in the error variables. Moreover, the ³  %CXå

final two termsk ink the expression depend on the control be a vector of absolute errors of the backstepping errors.
kg k
errors ³ and \ ³ . Recalling Eqn’s 21, 24, 32 and 33, it may be directly ã
verified that the derivative of the Lyapunov function
The control design for the motor actuators are based on •
kg k k k is bounded by
minimizing the control errors ³ and \ ³ for the rigid ãH µ L L
T x
body dynamics. To simplify the notation set
, -  where
Ó g
 \  \ S"  4  \
)  ) ) , - x/e
Ó g w 
~±4 ~„€ \ "Ô~€ \ «" ~€ \ "«~" $ $ $ $ $
) ) )
Ó Ó Ó Ç ” kg k o
, ³ 
and ,T ~ . Furthermore,
, set Z Ä Z   WWW  Z h " , $
w ,
T w 
~
$ T w  $
A ” A
Z v Z   W{WW  Z h " , \pÕ Ö \   W{WW  \ h " and B
× 
Ç ” kg k w
diag  B_^  WWW  B_^ " . Then Eq. 11 can be written in block ~ w Ë Â Ã,
p
$æT A w  ³ $ T ” kg k $
w k ÂÃ ~ k
form , k  k
H ,
B
× Z  \FÕ Tqf Z wFç Â Ã ³ B Ï
$ $ $ ³ T ” kg k T â wpË
,  Ó o w w k à k ~
Ó
Note that ³ x* Z TØx Ï ~" . Taking the derivative of ³ wFÙ
Ó $èT ” A w  T ” kg k T ” kg k $
³ it follows that • ~ ~ k
H k
”   H B Ï wpÙ
Ó Ó Ó
³ x* Z  \FÕ Tqfx Ï "‚T x Ï ~" W $ $ $ T â wpË $
•
Set
   H  wpÙ wpÙ
The quadratic expression T L x L is guaranteed negative
Ó N Ó Ó
\pÕ /eRfx Ï Z Ï x Ï ~ÉT Z Ï ³  cq$ definite if and only if the symmetric matrix x is positive
 definite. This is true if and only if the principal minors Recalling that sk  ? ~• " L - T sk  L - " ? ~ • one has
of x are positive.  ,
L L  W
? ~ a/ T , .²~_Í& ? ~  .¶~Í (37)
The first two principal
w w
minors are positive definite due
This process determines the first two components of
to the choice of   , cé$ . The third principal minor is -
the desired angular velocity ? ~ . To determine ? ~ it is
positive if ” kg k, necessary to recall the kinematic relationship between
wFË ~
c A , w , w k ÂÃ k , W the Euler angles and the angular velocity of a rigid body
 ,
³ (cf. for example (Murray et al., 1994)). One has
Recalling Eq. 18 and applying the bounds in the theorem H
“ ‡H ~ o ‹ Œ ˆ Œ
statement one has $
Uˆ ‰ ˆUŒ ˆU‰ ‹_Œ
g A J N A w  w  N w , k , k ƒH ~  $ T ? ~ W
~ µ  Ýà" N  " ˆU‰
ˆ‰Ø‹_‰ ‹Œî‹‰ ˆUŒ
…~
Using the bound Eq. 34, it follows that - H
kg k ä ” A J Solving for ? ~ in terms of ‡~ using the first row of this
~  N Ýà" W
k ÂÃ k equation leads to
In additionwpË applying the bound it follows that ³ cÜb - ˆU‰ H ,
Œ
choosing superior to the bound given in the theorem ? ~  ˆUŒ  ~ÉT ? ~
statement ensures that the third minor of x is positive ,
definite. Once this choice is made it iswpç clear that the where ? ~ is given by Eq. 37.
fourth minor is also positive definite for cq$ .
The final two minors are of interest since they involve
REFERENCES
the interaction terms associated with the approximations
made in the virtual control inputs during the control Egeland, O., M. Dalsmo and O.J. Sordalen (1996). Feed-
design. Thus, unlike the case for classical backstepping back control of a nonholonomic underwater vehicle
g
designs, the dynamics of ³ and \ ³ interact with all stages with constant desired configuration. International
of
wpÙ
the error dynamics. However, since the control gain Journal of Robotics Research Vol 15, 24–35.
used at this stage is independent of any further Frazzoli, E., M.A. Dahleh and E. Feron (2000). Tra-
calculations it may be chosen arbitrarily and indeed jectory tracking control design for autonomous he-
may even be chosen time varying to avoid robustness licopters using a backstepping algorithm. In: The
problems in the asymptotic limit of the control design. American Control Conference. Illinois, USA.
A more complete discussion of the potential of the Hauser, J., S. Sastry and G. Meyer (1992). Nonlinear
proposed control design is beyond the scope of this control design for slightly non-minimum phase sys-
paper. Straightforward
wpÙ
but tedious calculations show tems : Applications to v/stol aircraft. Automatica
that choosing superior to the bound given in the Vol. 28, No 4, 665–679.
theorem statement ensures the the final two minors of Koo, T.J., F. Hoffmann, H. Shim and S. Sastry (1998).
x are positive definite. The proof follows by applying Control design and implementation of autonomous
Lyapunov’s direct method. helicopter. In: The 37th Conference on Decision
and Control. Florida, USA.
Remark 3.3. k g The
k
theorem k statement
k
includes simplify- Mahony, R. and T. Hamel (2001). Adaptive compen-
ing bounds ~ c Ú and ³ à cêb as well as a bound sation of aerodynamic effects during takeoff and
k k
F— ~ µ Ý . In practice, this covers all situations in which landing manoeuvres for a scale model autonomous
one wishes to apply the control design. It is the authors helicopter. European Journal of Control (EJC) Vol
opinion that these theoretical bounds can be significantly 7,no 1, 43–58.
relaxed, however, such an analysis is beyond the scope Mahony, R., T. Hamel and A. Dzul (1999). Hover con-
of the present paper. trol via approximate lyapunov control for a model
helicopter. In: The Conference on Decision and
Control. Phoenix, Arizona, USA. pp. 533–534.
APPENDIX A Murray, R.M., Z. Li and S.S. Sastry (1994). A Math-
ematical introduction to Robotic Manipulation.
Consider the kinematics of the desired attitude .²~ CRC Press.
H
.¶~±R.²~ sk  ? ~" (35) Prouty, R.W. (1995). Helicopter Performence, Stability
and Control. Krieger Publishing Company, reprint
From Eqn’s 3 and 35, it follows with addition, original edition (1986). USA.
H
.@
³ ë. ³ sk  ? ³ " Sepulchre, R., M. Janković, and P. Kokotović (1997).
Constructive Nonlinear Control. Springer-Verlag.
Deriving the expression of .¶~ L - (Eq.
• 19) one obtains London.
sk  ? ~#" L - . ~ Í& (36) Teel, A. (1996). A nonlinear small gain theorem for
~ L - the analysis of control systems with saturation
Where Í is defined as known part of . The
~ ì S.²~ "
g IEEE Transactions on Automatic Control Vol 41,no
derivative is not exactly known due to the error term ³ 9, 1256–1270.
in Eq. 20. Direct calculation leads to the error bound
kg k
k l L - k µ w  N w , ³
ÍíT S.²~ "  " kg k W
l ~

View publication stats