Proceeding of the 2004 American Control’Conference ThA13.

2
Boston. Massachusetts J u n e 30 -July 2,2004

Controller and Observer Design for Lipschitz Nonlinear Systems

Prabhakar R. Pagilla and Yongliang Z h u

Absfmcl- We consider three related problems for stabilization of a [ I I], it was shown that global stabilization of nonlinear systems is
class of Lipschitz nonlinear systems: (1) fuDstite feedback eontroller possible using linear feedback for a class of systems which have
design, (2) observer design, 2nd (3) output feedback controller design. triangular structure and nonlinearities satisfy certain norm bounded
Suffirirot conditions are developed for the design of 80 exponentially
stable linear full-state feedback controller and an exponentially stable growth conditions. A backstepping design procedure for dynamic
nonlinear obsrner. G k r n that the sufficient conditions of the eon- feedback stabilization for a class of triangular Lipschitz nonlinear
troller and observer problem are satisfied, we show that the proposed systems with unknown time-varying parameters was given in [12].
controller with estimated state feedback from the propored observer Output feedback control of nonlinear systems has been extensively
will achieve exponential stabilization. Simulation results on an example
are given to numerically verify the proposed designs. studied in recent literature [131, [14], [IS], [16].
Observer design techniques for Lipschitz nonlinear systems
I. ~NTRODUCTION were considered in [17], [18], [19], [2O], [21]. The observer
design techniques proposed in these papers are based on quadratic
The output feedback control problem for nonlinear systems has Lyapunov functions and thus depend on the existence of a positive
received, and continues to receive, considerable attention in the
definite solution to an algebraic Ricarti equation. In [19]. insights
literature due to its importance in many practical applications
into the complexity of designing observers for Lipscbitz nonlinear
where measurement of all the state variables is not possible. Output
systems were given; it was discussed that in addition to choosing
feedback control design usually involves hvo related problems:
the observer gain in their nonlinear Luenberger-like observer, one
observer design and controller design which uses estimated state
bas to make sure that the eigenvectors of the closed-loop observer
and output as feedback. Unlike linear systems, separation principle system matrix must also be well-conditioned to ensure asymptotic
does not generally hold for nonlinear systems. Therefore, the
stability. The existence of a stable observer for Lipschitz nonlinear
output feedhack control problem for nonlinear systems is much
systems was addressed in [20]; a sufficient condition was given on
more challenging than stabilization using full-state feedback, It
the Lipschitz constant. Some of the results of [20] were recently
is well known that the observer design problem for nonlinear
corrected by [21]. For the nonlinear observer of [20], it was shown
systems by itself is quite challenging. One has to often consider in [21] that two sufficient conditions are required to guarantee that
special classes of nonlinear systems to solve the observer design the observer is exponentially stable.
problem as well as the output feedback control problem. Due to
In this paper, we provide a solution to the output feedback
their practical significance, two special classes of systems that
control problem for Lipschitz nonlinear systems under some
were often considered in the literature are nonlinear systems with
sufficient conditions on the Lipschitz constant. First, we design
a triangular structure and Lipschitz nonlinear systems.
a linear full-state feedback controller and derive a sufficient con-
A systematic approach to the development of observers for
dition under which exponential stabilization is achieved with full-
nonlinear systems was given in [I]; a nonlinear coordinate trans-
state feedback. Second, we propose a Luenberger-like observer,
formation was used to transform the original nonlinear system to
which is shown to be an exponentially stable observer under only
a linear system with the addition of an output injection term. The
one sufficient condition. Given that the sufficient conditions of
nonlinear state transformations were also employed in [2], [3],
the controller and observer problem are satisfied, we show that
141 to obtain linear canonical forms that can be used for observer
the proposed controller with estimated state feedback from the
design. A coniparative study of four techniques that appeared in
proposed observer will achieve exponential stabilization, that is,
the 1980’s for observing the states of nonlinear systems was given
the proposed controller and observer designs satisfy the separation
in [ 5 ] . In 161, a new approach was given for the nonlinear observer
principle.
design problem; a general set of necessary and sufficient conditions
The rest of the paper is organized as follows. In Section II,
was derived using the Lyapunov’s auxiliary theorem.
we give the class of Lipschitz nonlinear systems, the assumptions,
In [7], counterexamples were given to discuss the problem of
the notation used, and some prior results that will be useful for
global asymptotic stabilization by output feedback; a phenomenon
the developments in the paper. The full-state feedback control
called “unboundedness unobservability” was defined; it means that
problem, the observer design problem, and the output feedback
some unmeasured state components may escape in finite time
control problem are considered in Sections 111, IV, and V, respec-
whereas the measurenients remain bounded. Recent research has
tively. Section VI gives an algorithmic procedure for computing
focused on considering a selective class of nonlinear systems by
the controller and observer gains while satisfying the sufficient
placing some structural conditions on the nonlinearities to solve
conditions. An illustrative example is given in Section V11. Section
the output feedback problem. Global stabilization by dynamic
Vlll gives conclusions and some relevant future research.
output fcedback of nonlinear systems which can be transformed
to the output feedback fonn was given in [SI. Output feedback
11. PRELIMINARIES
control of nodincar systems in triangular fonn with nonlinearities
satisfying certain growth conditions was considered in [9], [IO]. In We consider the problem of controller and observer design for
the following class of Lipschitz nonlinear systems:
This work was suppaned by NSF under the CAREER Grant CMS-
9982071. The authors are with the School of Mechanical and Aerospace
Engineering. Oklahoma State University, Stillwater, OK 74078, USA.
i=&+Bit + O(x, U), (la)
Corresponding author e-mail: pagillaeceat .okstate .edu y=cx (Ib)
0-7803-8335-4/04/$17.00 02004 AACC 2379
where x E W“, ii E WP, and y E IRq are.the system state, input, and where the first inequality is a consequence of assumption A2 and
output, respectively. We assume, without loss of generality, that the second inequality is obtaioed by completing squares on the
x = 0 is the equilibrium point of the system. term 2 y l ~ P ~ . x ~For
\ ~ ~any ~ .> 0, V, 5 -qCxTx, if
. r ~qc
-
We use the following notation throughout the paper. IlMIl
denotes the Euclidean n o m of the matrix or vector M.@ denotes
+ +
A, Pc P,A, P,P, + +I = -qJ
7
(9)
the complex conjugate transpose ofthe matrix M. o,,,,,~(M) denotes The choice of the control gain matrix in (9)
the smallest singular value of the matrix M. The square of the
singular values of the matrix M a r e the eigenvalues of the matrix K = BTPc/2 (10)
f l M . det(M) represents the determinant of the matrix M.
results in the following ARE:
Assumprion A I : O ( X , Uis) Lipschitz with respect to the state x,
uniformly in the control U, that is, there exists a constant y such A;$‘, +P,Anc+Pc ( I - B B T / l l B ( ( Z )Pc+ ($ + q c ) /= 0. (I I)
that
Now we consider the problem of the existence of a symmetric
~ ~ @ ( x l , u ) - @ ( x 2 ! ~5Yllxl-*2I1>
)ll X I J 2 EW”,uEWP. positive definite matrix P,, which is the solution to the ARE ( I I).
Assumprion A2: IlQ(x,u)ll 5 yIIxI1, Vu E W. Since A,, is Huwitz, (I-BBT/llBllz) t 0, and (? qc)I > 0, +
Assumplion A3: The pair ( A , B ) is controllable. by Lemma I , the problem reduces to showing that the associated
Assumprion A4: The pair ( C , A ) is observable. Hamiltonian matrix
Defrnirion D I : The number S(M,N) is defined as

(3)
is hyperbolic.
where i = & i I is an identity matrix with appropriate
and Lemma 3; Hc is hyperbolic if and only if
dimension. See [21] for a discussion of the number S.
Lemma 1: [21] Consider the Algebraic Ricani Equation
Proof. The detenninant of‘the matrix ( - i d - I f C $ is
+
A ~ P + P A PRP+ Q = 0. (4)

If R = RT 2 0 , Q = Qr > 0 , A is Hunuitz, and the associated

A
Hamiltonian manix H = IS hyperbolic, i.e., H has
[
:
Q-AT]’ . ’ .
no eigenvalues on thc imaginary MIS, then there exists a unlque
P = P T > 0 to be the solution of (4).
Lemma 2; [ZO], [21] Let y 2 0 and define =(-l)”det ((p+ ~ l ~ ) ( - l + B B ~ / l i B-l ( i~a)I + & ) ( i c o - A ~ J )
=(-l)”det ( c H ( i w ) C ( i o )- (?+ q c ) I )

Then y < S ( A , C ) if and only if H y is hyperbolic. where C ( i w ) =

A io1- ATc
( ~ B T / l l B I I 1
, and the third equality is
obtained by using the formula for determinant of block matrices
111. FULL-STATE FEEDBACK CONTROLLER DESIGN [22. p. 6501 because (p
+qe)l is non-singular. Using the definition
In this section we consider the regulation problem for the system DI, it is seen that, if (13) holds, det( -imI-Hc) # 0 which implies
(la) with full-state linear feedback under the assumptions A2 and that j o , V o E W is not an eigenvalue of Q , which in turn implies
A3. Consider the control input: that H, is hyperbolic. The necessary part of the proof is similar
to that of Lemma 2; we refer the readers to [21] for details.
U = - K X / ~ I -B ~K ~ ~~ ~ (5) nteorem I : For the nonlinear system given by (I), with the
assumptions A2, A3, and with the control input given by (S), the
where Kl is the pre-feedback gain matrix and is chosen such that equilibrium x = 0 is exponentially stable if the condition given by
A
A,, = A - BKI is stable and K is the feedback gain matrix to be (13) is satisfied.
determined later. With this control input, the closed-loop dynamics
of (la) is 1V. OBSERVER DESIGN
Consider the following observer for the system (1):
i = (A. - B K / I I B I I 2 ) x f @ ( x , u ) =AA- , x + O ( x , u ) . (6)
X;= A : + B u + Q ( ~ , u ) + Trg.-c;)
?+E” +L~ g.- c;) (14)
Consider the following Lyapunov function candidate IlClI
A
VJX) =xTP@r, Pc = P,‘ > 0. (7) where eo 2 -?, L I is chosen such that A., = A - L I C is Hunvitz,
The time derivative of V,(X) along the trajectories of (6) is and L is the observer gain matrix.
Let the estimation error Y b x - 2 . The error dynamics is
t.,(x) = XT @Pc +P , A , ) x t 2rTPcO(x,u)
-
< x T (A. 7 P,+P~~,)X+2YllP,*IIllXll

(8) (15)
2380
Consider the following Lyapunov function candidate Since Pc is the solution to the ARE (Il), we have

V,(*? =ZTP,.E (16) &(x) s -llcx~~+ccll~llll~l (26)

The time derivative of V,($ along the trajectory of (15) is +
where CG = IIP3II IIZPBKI/I
Now consider the function
Pn(4=f@?Po +POTa)F+2YTP, (Q(x,u) - @(;,U)). (17)

Choosing
%*? = Cv,(x) + V(.7 (27)
L = P,-'CT/Z: (18) where < > 0 and V,(?) is as given by (16). The time derivative of
using Assumption AI, and simplifying along the same lines as W ( x , q is given by
done in the full-state feedback controller case, we have
*(x,x7 5 -C~=llxll*+ CCcll~llll.11- 11011.11~- (28)
&(?) 5 -q2.?, (19)
Choosing 5 = q,q,/C results in
if P* = P
,' > o satisfies
A~P~+P~,,+P~P,+(~+~,)I-(J+Eo)CTC/IIC
= OI 1 2(20)
*w 5 -C11<llxl12/2 - rl"ll.1I2/2. (29)
Therefore. x and .?exponentially converge to zero. m
for some q. > max(.%,O). Since A,, is Hurwitz and the matrix Remark 2: The number S is realization dependent, that is, its
( f + q o ) I - ( ( ~ + E ~ ) / ( / C I / ~ 0,
) Cby
~C >
lemma I , the above value depends on A , B, C. IfA is unstable to begin with, then any
ARE has a unique Po =Po' > 0 as the solution if the following preliminary control used to stabilize A will affect 6 . Since 6 and y
associated Hamiltonian is hyperbolic: depend on the realization, appropriate coordinate transfonnations
as discussed in [ Z O ] , in some cases, can be used lo increase S and
reduce y.
Remark 3: The bisection algorithm given in [ZI] can be used to
Lemma 4: fIo is hyperbolic if and only if
compute 6 ; it was suggested that 0 and IlAll be used as the initial
J.l....
< 6 (A"., fiC/llCIl)
Proof. Similar to lemma 3.
. (22)
m
guess for the lower and upper bounds, respectively, for S(A,C).
It is possible that the value of S may be greater than IlAIl. The

The following theorem summarizes the results of this section. upper hound must he changed to cmjn
Theorem 2: For the nonlinear system given by I , with the
assumptions AI and A4, if the condition given by (22) is satisfied,
(14) is an exponentially stable observer for the system.
Remark I : Notice that the proposed observer, (14), requires
only one sufficient condition, (22), as opposed to the two sufficient
conditions for the observer given in [21]. The two conditions y. Therefore, if f(y) < 0, then there exists a yi > y such that
are required in [21] because: ( I ) the observer sttucture does not f(y1) < 0. Hence, iff(y) < 0, then there exists an qc > 0 such that
guarantee that the " Q matrix in the ARE (4) is positive definite f ( s ) < 0, that is, (13) holds. Same arguments hold for
and (2) the associated Hamiltonian matrix must be hyperbolic. condition (22) with E, = 0. Hence, instead of checking conditions
The proposed observer, (14), guarantees that the "Q" matrix in given by (13) and (22), one can respectively check the following
the ARE (4) is positive definite. two conditions
V. O U T P U T FEEDBACK CONTROLLER DESIGN Y< S(A;,YB~IIIBII), ~ < ~ ( A , ~ ~ ~ C / I I C I I )(30)
.
Combining the full-state feedback control design of Section
111 and the observer design of Section IV, we design an output Notice that the conditions given by (30) guarantees the existence
feedback controller for the system (I). of qc > 0 and q,, > 0, but not their values. Conditions (13) and
?7zeorern 3: Consider the system (1) with assumptions AI, A2, (22) with specified qc and 1 '. give the rate of convergence of
A3, and A4. If conditions (13) and (22) hold, then the equilibrium controller and observer, respectively.
x = 0 of the system ( I ) is exponentially stable, with Remark 5: The three resulb given by theorems I, 2, and 3 will
be applicable locally or globally depending on whether @ ( x , u ) is
U =- K ~ / ~ ~- KEI F
~ ~ 2 (23) locally or globally Lipschitz.
where .? is the estimate of x generated by (14), K is the gain VI. IMPLEMENTATION PROCEDURE
matrix given by (IO), Further, the observation error, Z = x - X ;
In the following, we give a systematic procedure to compute the
exponentially converges to zero.
Pmof Substituting the output feedback control law given by (23) observer and controller gain matrices with respect to the original
in ( I ) and simplifying we obtain system (I) in the event of the use of the preliminary control and
coordinate transformations.
X = (Am - B B T P , / 2 1 ) B 1 1 2 ) ~ + ~ ( ~ , ~ ~ ) + B B T P c ~ / 2 1 1 5 1 1 2 +(24)
BK~I
A . Observer gain matrir
The time derivative ofthe Lyapunov function candidate V&) given
I ) Pole placement
by (7) along the trajectories of (24) is
Rewrite ( I ) in the following form
pc(x) 5 xT (.:pc+PA,< +P ~ - B B ~ / P~, ~ $1) +
B ~ ~ ~ ) + +Lly +O(x,U),
X = ( A - LlC)x 5u (3 la)
+ x ~ P , B B ~ P ~+Yb
/ ~T
~ B~
~ ~c
~ ~ ~ l ( 2~3 . y=cx ( 3 ~
2301
 where L I is chosen such that ( A - L I C ) is stable. for the existence of the solution P, to the ARE
2) Similarity transformation
Let x = Tax', (31) bccomes
A”P, +P,A’+P, (I- B ~ E ~ ~ / P=I +~ (YF+
~ ~ ’ q).p = 0. (44)

+
( A - L~C)T,X’ q;I (BU L,Y) T;I@(T~, U)
2 = r;l + + If (43) is satisfied, the control gain is chosen to he K’ =
BtTP,/2 where P, = P: > 0 is the solution of (44).
=
Aa’*.’+~’~+r,-’~,y+rI~(~,r’,~),
(324 The gain matrix used in the full-state feedback controller or output
A feedback controller is
y = CTg’ = C‘x‘ ( 3 2 ~
2
where To E Itnx”is a nonsingular matrix. The new Lipschitz K = K 1 +K’TL1/\lT;’Bll . , . (45)
gain ./ is obtained from the following inequality V1I. AN ILLUSTRATIVE EXAMPLE: A FLEXIBLE LINK ROBOT
IlTy1@(T&i,1d) - T 1 @ ( T o x z , u ) l l 5 ./IIXI -x2ll (33) In this section, we consider the observer and controller design
for a flexible link robot [IS], [ZO], [21], [23]. The dynamics of the
for all x l , q E P”and U E PP. robot is described by the following state space representation:
The observer for (32) is given hy
i =Ax +bu + @ ( x , u ) , (46a)
2 = A’? + B ’ +~ T;] L ~ +
Y T , - ~ @ ( T , U~ ),
y=cx (46b)

’r],
+ ( r 2 + a ’ b - c~)i11c112, (344
where
v^= CT. (34b)
em 0 1 0 0
After choosing 2 -y12 and q. > niax(%,O), check the x=[b],A=[ -48.6 -1.25 ,” : ] , E = [
48.6 0
condition
U1 19.5 -19.5 0
i + 8. < s2 @ f > ~ C / l l C ~ I l ) (35) r n 1

for the existence of the solution Po to the ARE

A’TP,+PoA’+PoPot (~~’+~.)l-(~~’+%)d~d/I1~11‘=0.
(36)
if (35) is satisfied. the observer gain is chosen to he L’ = and 9, is the angular position of the motor; U, is the angular
P;ICfT/2 where Po is the solution of the above ARE. velocity of the motor; 91 is the angular position of the link; and
Notice that if one defines .?= TZ
as the estimate ofx, the system
01 is the angular velocity of the link.
Observer design: Since A is not stable, we design a pre-
(34) can be rewritten in terms of 2 by the following equations.
liminary gain L I such that (A - L l C ) is stahie with poles at
5=Aa+Bu+o(a,u) + L b - c ; ) , (374 -9.3275,-8.9203,-9.6711 and -4.7722. The gain L I is

where
v^= G
L = L l +(y/2+&o)ToL‘ijlCTa112.
B. Conrroller gain matrix
(37b)
(38)
L1 = [ 9.3275
-48.7804
-0.0524
19.4066
1.OOOO
22.1136
3.1994
-0.9032
] ’

1) Pole placement The Lipschitz constant of @(x,u) with respect to x is y = 3.33.

Rewrite (1) in the following form Using the similarity transformation x = T&, transform the sys-
X = (A - B K I ) x + B ( u + K l x ) +@(x,u) (39)
tem (46) with To = diag( [ 1 i 1 IO 1).
The new Lipschitz
constant is y l = 0.333. Choose constants &o =0.1111 and q. =
where Kl is chosen such that (A - BKI) is stable. 0.121 1, and check the condition given by (35). It is computed that
2) Similarity transformation S(A‘, f l C ‘ / l l d l \ ) =0.8389, so (35) is satisfied. Solving
Let x = Td’, (39) becomes the ARE (36) results in
2 = T;I(A - E K ~)T& + T,-~B(U+K~m’)
9A’x’ +B‘( U +K , T s ‘ ) + TF1@(Tg‘, U )
+ T;~@(T~x’,u)

where T, E RnX“is a nonsingular matrix. The new Lipschitz

(40) Po= [ 18.6546
-0.0234
0.0396
0.0012
-0.0234
5.9522
-12.5731
1.9503
0.0396
-12.5731
30.8320
-8.8656
0.0012
1.9503
-8.8656
9.7302 I ’

gain ./ is obtained from the following inequality r 0.0268 0.0003 1

I l T F 1 @ ( T ~ i-2 ~~ ‘)l @ ( ~ c ~ z , ~ ) l5l Y/ I / X I
for all X I x2 E W“ and U E WP. Choosing U = -KIT& -
-x2II (41)
L, = -1p - l C ~ =
2 ”
- I1 0.0003
0.0001
0.0000
1.1392
0.5405
0.2641 1
I
Kfx’/+3’ll* results in the following closed-loop system Thephserver for the flexible link robot (46) is in the form of (37)
if= (A’ - B‘K’/ 11B’ II’)~’+ T; I o(T ~ X ‘ ,U). (42) with
9.3334 1.0001

[
Choose qc > 0 and check the condition -48.7804 22.3665
-0.0524 3.3194
L= 19.4066 - 0 . 3 1 6 7 ]
2382
where (38) is used. mamx which was obtained in the design of the full-state linear
The simulation results of the observer, (46), are shown in feedback controller can be used with the estimated state, where
Figures I and 2. In the simulation, the initial value of x , x(O), the estimates are obtained from the proposed observer.
is chosen to be [ 1 I 1 1 1‘;the initial value of x7 n’(O), Systems with Lipschitz nonlinearity are common in many prac-
is chosen to be [ 0 0 0 0 1’.The system is assumed to tical applications. Many nonlinear systems satisfy the Lipschitz
property at least locally by representing them by a linear pan plus
be under no control, that is, U = 0. Fig. 1 shows the motor
angular position, motor a n p l a r velocity, and their estimates. Fig. 2 a Lipschitz nonlinearity around their equilibrium points. Hence,
shows the link angular position, the link angular velocity and their the class of systems considered in this paper cover a fairly large
estimates. From both the figures, one can see that the estimates number of systems in practice.
converge to the true states. There are some challenging problems that need to be addressed
Output feedback control: As done in the observer design case, in the future. It is clear that the number S is realization dependent.
we first use a preliminary control to make ( A - B K l ) stable with So, a natural question to ask is which realization gives the
poles at -5.8989, -5.6390. -4.9245 and -8.9109. The gain K I maximum value for S and further, how does one transform the
is found to be system given in any arbitraly form to this particular realization.
Moreover, it is also not clear as to how one can. in general. find
KI = [ 7.8092 1.1168 -4.3436 1.12 1. transformations that increase 6 and decrease y simultaneously.
Then, a similarity transformation, x = T<x’, is used to reduce the It is also emphasized here that the conditions for both full-
Lipschitz gain with T, = diag( [ 1 1 I I O 1).
The new Lips- state feedback and output feedback stabilization are sufficient
conditions; how to satisfy these two sufficient conditions is a
chitz constant is y/ = 0.333. Choose constants qc = 3.7947(10-4),
and check the condition iven by (43). It is computed that challenging problem which needs to be investigated in the future.
G B f T / I / B ’ I I j =0.3552, so (43) is satisfied. Solv-
REFERENCES
ing the ARE (44) results in
[ I ] A. I. Krener and A. Isidori, “Linearization by output injection and

1
13.4725 1.4496 -8.8421 17.387
nonlinear observers:’ Systems & Conlrol Lellerr, vol. 3, pp. 47-52,
1.4496 0.18736 -0.99393 1.8462 1983.
-8.8421 -0.99393 6.1806 -11.1047 ’ [2] D. Bestle and M. Zeifz, “Canonical form observer design for
17.387 1.8462 -11.1047 26.2607 non-linear time-variable systems:’ Inrernalionol Journal of Conlml,
vol. 38, no. 2. pp. 419431, 1983.
which in turn results in [3] A. I. Krener and W. Respondek, “Nonlinear observers with lineariz-
I able emor dynamics:‘ S/AM Journal on Conlrol and Oplimirafion,
K’ = -B’TP,
2 = [ 15.6553 2.0235 -10.7345 19.9385 1. vol. 23, no. 2, pp. 197-216, 1985.
141 G. Ciccarela, M. D. Mora, and A. Germani, “A Luenberger-like
The control input for the flexible link robot (46) is II = -K? with observer for nonlinear systems:’ lnlernotionol Joumal of Conlrol,
vol. 57, 1993.
K = [ 7.8428 1.1212 -4.3666 1.1243 ] [ 5 ] B. L. Walcatt, M. J. Corless, and S. H. Zak, “Comparative SNdy
of non-linear state obsenation techniques:’ Internorional Journal sf
where (45) is used. Control, vol. 45, no. 6, pp. 2109-2132, 1987.
The simulation results for regulating the states of the flexible [GI N. ~ ~ ~and~c. aavaris. t z i onli~ linear observer design using
robot (46) to zero are shown in Figures 3 and 4. In this simulation, Lyapunov’s auxiliary theorem:’ Sysremr & Conhol Lelrers, vol. 34,
the initial values of x and .?are chosen to be the same as those in pp. 241-247, 1998.
the silnulation for the observer in the previous simulation, ~ i 3 ~ , F. Mazcnc, L. PralY, and W. P. Dayawansa, “Global stabilization by
L7l
output feedback: examples and counterexamples:’ Sysiem & Control
shows the motor angular position, motor angular velocity, and their Lelters, vol. 23, pp. 119-125, 1994.
estimates. Fig. 4 shows the link angular position, the link angular .~R. Marino and P. Tomei, “Dynamic output feedback linearization and
181
velocity and their estimates. Comparing Figures 3 and 4 with global stabil~ation,“S,,~l~,~ & ContmiLetlers, vol. 17, pp. 115-121,
Figures 1 and 2, it is clearly seen that, under the output feedback 1991.
191 J. P. Gauthier, H. Aamouri, and S.Othman, “A simple observer for
control, four states of the robot (#,“,o,,,#land 01) converge to
nonlinear systems applications to bioreactors,” IEEE Tronsaclionr on
zero rapidly; whereas, without control, the states converge to zero Aulo,nrrIic Conool, vol. 37, pp. 875-880, June 1992.
very slowly. Also, the convergencc of the estimated states to their . . C . Oian
1101 . and w. Lin..“Outout . feedback control o f a class of nonlinear
true values is observed. systems: A nonseparation principle paradigm:’ IEEE Trm~oc~ions on
Automalie Control. vol. 47, pp. 171&-1715, October 2002.
vlI1. CONCLUSIONS [ I l l 1. Tsinias. “A theorem on global stabilization of nonlinear systems
by hoar feedback,”Svsrrm & Cunirol Leners. vol. 17, pp. 357-362,
In this paper, we considered the full-state feedback control 1991.
problem, the observer design problem, and the output feedback [I21 1. Tsinias, “Backslepping design for time-varying nonlinear systems
control problem for a class of Lipschitz nonlinear systems. We with uilknown parameters:’ Sysrcms & C o m d Letters. vol. 39.
. .
urooosed a linear full-state feedback controller and a nonlinear pp. 21g-227, 2000.
observer and gave sumcient conditions under which exponential [I31 A. Tcol and L. Praly, ”Tools for scmiglobal stabilization by partial
state and output feedback:’ SIAM Jouraal on Conlml and Oplimizo-
stability is achieved. Generally. for nonlinear systems, stabilization .
lion .I.. __,
.I,., _I n I A A - I ..””,
77 mrr. IYY .,,_.
140<
by state feedback plus observability does not imply stabilization [14] L. Praly and M. Arcak, “On certlintyequivalence design of non-
by output feedback, that is, separation principle usually does not linear observer-based conuollcrs:’ in Pmcerdings of d e 41st IEEE
hold for nonlinear systems. Huwevcr, for the class of nonlinear Conference on Decision and Conlrol, (!AS Vegas, Nevada. USA),
pp. 1485-1490, December 2002.
systems considercd in this paper, by using the proposed full-state
115) A. N. ~ t a and ~ ~H.i K. a a 1 i l , “A separation principle for tile
linear feedback controller and the proposed nonlinear observer, stabilization of a class of nonlinear systems,” IEEE Tranroc~ionvon
we show that the separation principle holds; that is, the same gain Auromotic Contrul, vol. 44, pp. 1672-1687, 1999.

2383
[I61 A. N. Atarsi and H. K. Khalil, "A separation principle for the Control
of a class of nonlinear systems," IEEE Tronsacrions on Aulomolic
Conrrol, vol. 46, pp. 742-746, 2001. ,
[I71 F. E. Thau, "Observing the state of non-linear dynamic systems,"
Inlernolional Journal of Control, vol. 17, no. 3, pp. 4 7 1 4 7 9 , 1973.
[IS] 1. R. Raghavan and J. K. Hedriek, "Observer design for a class of
nonlinear systems:' lnlcrnolional Journal of Conlml. vol. 1, pp. 17 I-
185, 1994.
[I91 R. Rajamani, "Observers' for Lipschitz nonlinear systems," IEEE
Tronroctionr on Auromalic Contml,vol. 43, no.3, pp. 3 9 7 4 0 1 , 1998.
[20] R. Rajamani and Y.M. Cho, "Existence and design of observer for
nonlinear systems: relation to distance Lo unobservability:' Interno-
rionul Journal of Canrml. vol. 69, no. 5, pp. 717-731, 1998.
[21] C. Aboky, G. Sallet, and I. C. Wvalda, "Observers for Lipschitz
non-linear systems:' Intrrnarionol Journal of Confrul, MI. 75, no. 3,
pp. 20&2.212,2002.
(221 T. hilath, Lineor Syslemr. Englewaod Cliffs, N.J.: Prentice-Hall,
Inc., 1980.
[23] M. Spong, "Modeling and control of elastic joint robots:' ASME
Journal of Dynomic Systems, Memurrmeni. and Conlml, vol. 109,
pp. 310-319, 1987.
[24] R. Eking, "Beween controllable and unconuollablc," Sysrem &
Conlrol Leriers, vol. 4, pp. 263-264, 1984.
Fig. 2. Lek angular position 81, link angular vclocity 01,and their
[25] D.L. Boley and W. S . Lu, "Measuring haw far a controllable system estimates 81 and 61 are shown. The control is U = 0.
is from an uncontrollable one:' IEEE Transodium on Auromolic
Conrrol, vol. 31, no. 3, pp. 249-251, 1986. Moxp*MnWIUar"mL
[26) M. Wick and R. A. DeCarlo, "Computing the distance to an
uncontrollable system:' IEEE Tronroclionr on Aulomorie Conlml.
vol. 36, no. 1, pp. 39-49, 1991.
[27] R. Byers, "A bisection method for measuring the distance of a stable
matrix to the unstable maeices," SIAM Journal of Scienrflc and
SIolislicol Compirling, vol. 9, pp. 875-881, 1988.

-I' I
0 0.5 1 1.5 2 2.5 3 31 4

Fig. 3 . MoJoor angular position e,, motor angular velocity w, and their
estimates 9, and & are shown. The control is U = -K2

uni po.lb" B"6 im .

U
"
*

___
Fig. 1. Motor angular position e,,,, motor angular velocity o,,,and
estimates e,, and & are shown.T h e control is U = 0.
, their -0.2
0 0.6 I
~

7.5 I 2.5 1 3.5 .

Fig. 4. Lek angular position er, link angular velocity 01. and their
estimates 61 and 6,are shown. The conuol is U = -K2

2304