You are on page 1of 7

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

net/publication/4049299

A continuous asymptotic tracking control strategy for uncertain multi-input


nonlinear systems

Conference Paper · February 2003


DOI: 10.1109/ISIC.2003.1253913 · Source: IEEE Xplore

CITATIONS READS
44 344

4 authors, including:

Marcio de Queiroz Jiaju Chen


Louisiana State University Guiyang Medical University
169 PUBLICATIONS   5,038 CITATIONS    163 PUBLICATIONS   2,844 CITATIONS   

SEE PROFILE SEE PROFILE

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

Multi-Agent System View project

MD simulation of biotin carboxylase View project

All content following this page was uploaded by Jiaju Chen on 20 November 2015.

The user has requested enhancement of the downloaded file.


Proceedings of the 2003 IEEE
International Symposium on Intelligent Control
Houston, Texas • October 5-8, 2003

A Continuous Asymptotic Tracking Control Strategy for Uncertain


Multi-Input Nonlinear Systems
B. Xian, D.M. Dawson, M.S. de Q u e i r o z t, a n d J. C h e n
Department of Electrical and Computer Engineering tDepartment of Mechanical Engineering
Clemson University Louisiana State University
Clemson, SC 29634-0915 Baton Rouge, LA 70803-6413
[xbin, ddawson, jianc]@ces.clemson.edu dequeiroz~me.lsu.edu

Abstract 0 is a constant unknown parameter), adaptive control


[10, 18] is often considered to be the method of choice. If
In this paper, we present a novel continuous control ~(t) can be upper bounded by a norm-based inequality,
mechanism that compensates for uncertainty in a class sliding mode or variable structure controllers [21] (i.e.,
of multi-input nonlinear systems. The control strategy discontinuous control strategies) can be developed. If
is based on limited assumptions on the structure of the ~(t) is periodic and the period is known, learning con-
system nonlinearities. A Lyapunov-based stability argu- trol schemes can be designed [13]. The approach of [13]
ment is employed to prove semi-global asymptotic track- is also applicable to the case where ~(t) can be repre-
ing. The control mechanism has the interesting feature of sented as an integral of a predefined kernel function mul-
"learning" the unknown system dynamics. For the sake of tiplied by a unknown influence function. While this brief
clarity, the proposed control design is initially presented discussion is not intended to be a comprehensive review
for a first-order, single-input case. Using this result as of all the various methods of control, we also note that
a stepping stone, the design is then extended to higher- there has been much progress in the areas of adaptive
order, multi-input systems. control for time-varying systems [6], robust control [15]
(i.e., the continuous counterpart of sliding mode control),
control of exosystems [20] (i.e., ~(t) is the solution of a
1 Introduction linear time-invariant system with unknown coefficients),
adaptive control for classes of nonlinearly parameterized
The control of uncertain nonlinear dynamic systems is systems [1, 5, 12], and neural network-based control [11].
a topic that continues to challenge control theoreticians. In addition to the problem of modeling uncertainty, the
This topic is also of practical importance since many real- obstacles associated with dealing with high-order, multi-
world systems exhibit nonlinear dynamic behavior. To input nonlinear systems represent additional control de-
make matters even more difficult, in many practical cases, sign challenges. Since this paper addresses the control
the mathematical model is poorly known or uncertain. To of a class of uncertain, multi-input nonlinear systems,
cope with the uncertainty issue, an abundance of design a brief review of related work in this area is also war-
tools have emerged during the last two decades that fa- ranted. For example, recently in [9], a general proce-
cilitate the systematic construction of controllers for var- dure was presented for designing robust switching adap-
ious classes of uncertain nonlinear systems. As one might tive controllers for a broad class of multi-input nonlin-
expect, the choice of a specific control design method is ear systems which includes feedback linearizable systems,
strongly influenced by the type of uncertainty associated parametric-pure feedback systems, and systems with a
with the system model. To illustrate this point (and facil- linear-in-the-parameters control Lyapunov function. A
itate the paper motivation), consider the following scalar more restrictive class of multi-input systems than [9] was
system considered in [22] in the design of an adaptive robust con-
~b = - w + ~(t) - ~(t) (1) troller. Adaptive controllers were proposed for a class of
feedback linearizable systems in [2, 3, 19]. For a more in
where w(t) is the state variable, and qb(t) represents a
depth review of control approaches for multi-input non-
feedforward control term injected to compensate for the
linear systems, the reader is referred to the references in
unknown nonlinear function ~(t). The main issue here
the aforementioned papers.
is: How does one design ~(t) such that w(t) is regu-
lated to zero7 If ~(t) can be linearly parameterized (i.e., In this paper, we present a full-state feedback tracking
~p(t) = Y(t)O where the nonlinearity Y(t) is known, and controller for a class of uncertain, multi-input nonlinear

0-7803-7891-1/03/$17.00 © 2003 IEEE 52


systems. Specifically, we examine the following class of Let Xd(t) E ~ be a given reference trajectory that is
higher-order, multi-input systems continuously differentiable up to its third derivative such
that
dn - l z ) dixd(t)
d~z E£~, i = 0,1,2,3. (7)
H z,$,..., dtn_ 1 ,t91(t),t + dt ~
dtr~
To quantify the control objective, we define the tracking
(
G z, ~,..., dt~_l,~2(t),t ) u. (2) error e(t) E ~ as follows

e---Xd--X. (8)
where z(t) is the output vector, u(t) is the control input
vector, O/(t), i = 1,2 denote unknown, time-varyingpara- Our objective is to obtain asymptotic tracking with a
meter vectors, H (.) is an uncertain nonlinear vector func- continuous control law using (7) and norm-based, in-
tion, and G (.) is an uncertain nonlinear matrix function. Oim(xd) oif(xd)
The proposed controller produces semi-global asymptotic equality bounds on the functions OXid and Ox~ '
tracking under the assumption that g (.) and G(.) are i = 0,1,2.
second-order differentiable, and G (.) is a positive-definite
symmetric matrix. To achieve this result, we utilize a R e m a r k 1 For simplicity of presentation, we have as-
new continuous control mechanism which is inspired by sumed re(x) and f(x) do not depend explicitly on time
a simple example provided in [16]. The control approach or on unknown time-varying parameters. However, it
compensates for the uncertainty associated with H (-)and should be emphasized that the proposed control approach
G (.) with limited assumptions with regard to structure can compensate for these phenomena provided the time-
of the nonlinearities. varying effects satisfy second-order differentiability condi-
The paper is organized as follows. For the sake of clar- tions similar to those given in (6). That is, the functions
ity, we first present the proposed control design for the re(x) and f (x) could be easily replaced by m(x, Ol (t), t)
first-order, single-input, scalar version of (2), i.e., and f (x, 02(t),t) where Oi(t), i = 1, 2 denote unknown
dz time-varying parameter vectors and other time-varying
d--[ = H(z, 01 (t), t) + G(z, O2(t), t)u. (3) disturbance that may appear nonlinearly in the model;
hence, the system given by (~) represents the same class
We then illustrate how the control approach can be ap- of systems introduced in (3).
plied to the system given by (2).
2.1 Control Law
2 First-Order Single-Input Sys- Based on the subsequent stability analysis, we propose
tems the following control law I to achieve the stated control
objective
To explain the control technique, we first examine a first-
order, single-input nonlinear system having the general (ks + 1)e(t) - (ks + 1)e(to)+
form
rn(x)~c + f (x) = u (4) fit( [(ks + 1 ) a e ( T ) + ~sgn(e(7))] dT (9)

where x(t) E IR is the system state, u(t) E IR is the con-


where ks, a, fl E IR are positive control gains, to is the ini-
trol input, and re(x), f(x) E 1R are uncertain nonlinear
tial time, and sgn(.) denotes the standard signum func-
functions. It is assumed that m(x) and f(x) satisfy the
tion. The control law of (9) ensures asymptotic tracking
following assumptions:
provided the control gains ks and 3 are chosen sufficiently
A s s u m p t i o n S l . The function m(x) is positive and large relative to the norm of the initial tracking error
bounded as follows and a reference trajectory-based bound, respectively. The
proof of this result is presented in the following two sub-
< < (5) sections. In particular, we first develop the closed-loop
where m E ~ denotes a positive constant, and error system under the proposed control and then ana-
~ ( x ) c R denotes a positive nondecreasing function. lyze its stability using a Lyapunov-based argument.

A s s u m p t i o n $2. The functions re(x) and f(x) are


2.2 Error System Development
second-order differentiable with respect to x(t) such
that To facilitate the following analysis, we define the filtered
tracking error [21] r(t) e ~ as follows
re(x), ore(x) o m(x) E£o¢ if x(t) E£.oo
Ox ' Ox 2 (6) •
r - - e -}- Ole (lo)
Of(x) f (x)
f(x), Ox' Ox 2 e£.~ if x(t) E g ~ . 1 T h e s e c o n d t e r m in (9) is u s e d to e n s u r e t h a t u(to) --O.

53
where a was introduced in (9). After differentiating (10) 2.3 Stability A n a l y s i s
and then multiplying both sides of the resulting equation
Before presenting the main result of this section, we state
by re(x), we have
the following two lemmas which will be invoked later.
m(x)i ~ = m(x)(~d -6 c~) -6 ~h(x)2 -6 ](x) - it (11)
L e m m a 1 Let the auxiliary function L(t) E l~ be defined
where the time derivative of (4) was used. After some as follows
simple manipulations, (11) can be rewritten as follows L ~- ~ (N~ - # ~ ( ~ ) ) . (22)

.~(x)~ = -~l~h(x)r - e - it + N ( x , ~ , t ) (12) If the control gain fl is selected to satisfy the following
sufficient condition
where

N(x,~, t) A m(x)(i~d + e~e) + m(x) ( x + l r ) + e + ](x).


fl > INn(t)l + - 11 I
O/
Nd(t) , (23)

(13) then
After taking the time derivative of (9), we obtain f t l L(~') d~" <- ~b (24)
/t = (ks + 1)r + flsgn(e) (14)
where the positive constant ~b E I~ is defined as
upon use of (10). After substituting (14) into (12), we
obtain the following closed loop system G ~ # I~(t0)l- e(to)Nd(to). (2~)
m(x)÷ = --~17h(x)r - e - (ks + 1)r - flsgn(e) -6 N(x, 2, t) Proof. See Appendix A.
(15) L e m m a 2 Let the region D in the state space be defined
Before analyzing the stability of the closed-loop sys- as follows D A {~ E I~'n I [l~[I < ~} where ~ E I~ is some
tem, we perform the following manipulation on (15). Let positive constant and let V(t, ~)" I~+ x D ~ ~+ be a con-
Nd(t) A N(Xd, iCd,t); hence, from (13), we can show that tinuously differentiable function such that
a~(x~) ~ + Sf(xd) ~ (16)
Nd = m ( x d ) ~ + ax~ Ox~ Wl (~~(t<_' ~)V(t' - W<-
( ~W2(~)
) (26)
and
a~n(xd) ~ + Vt >_ to and V~ E D where WI(~), W2(~) E IR are contin-
uous positive definite functions, W(~) E ~ is a uniformly
continuous positive semi-definite function, and to denotes
O2f (Xd) the initial time. Provided (26) is satisfied and ~(to) E S,
Om (Xd) J:di~d -6
2 OXd ax~ ~ + we have
Of(Xd) .. lim W(~) = 0 (27)
OXd Xd.
(17) t- +c<)

where the region in the state space denoted by S is defined


We note that Nd(t), 7Vd(t) E £oo due to (6) and (7). Now, as follows
after adding and subtracting Nd(t) to the right-hand side
of (15), we have S&{~ED[W2(~)_<6} where
6 < min WI(~)
II~ll=s
(28)
m(x)÷ = - l ~ h ( x ) r - e - (ks + 1)r - flsgn(e) +
2 with 6 E ~ being some positive constant.
Nd + N(x, ic, t) (18)
Proof. Direct application of Theorem 4.4 in [7].
where
A N- Nd. (19) We now state the stability result for the proposed con-
troller with the following Theorem.
R e m a r k 2 It is not difficult to show by the Mean Value
theorem that IV (.) defined in (19) can be upper bounded T h e o r e m 1 The control law of (9)ensures that all system
as [2S] signals are bounded under closed-loop operation and that

lim e(t), ~(t) = 0 (29)


t---~oo
where z(t) E 1~2 is defined as
provided that the control gain fl is adjusted according to
Z--- [~ ~]~ (21) (23), and the control gain k8 is adjusted as follows
and the bounding function p(llzll) e ~ is nondecreasing
in IizI[. The inequality given by (20) will be utilized in the 1
subsequent stability analysis. k~ > ~ ~--q--,0

54
where p(.) was defined in (20), At, A2(x), and A3 e lI~ are We now apply Lemma 2 to (36) and (40). Specifically,
defined as follows we see from (36) that lower and upper bounds for (34)
are given as follows
A1 ~ ½ min{1,m} A2(x) A max {½~(x), 1} (31)
A3 ~ min {1, a } ,
W~(y) - ~ iryll ~ W~(y) = ~(x)rlyrl ~ (41)
and 7o C ~ is defined as follows
while the upper bound for the derivative of (34) is given
by
~ e2(t0)+ [2d(t0)+ m-l(x(to))f(x(to))
(32)
~o = +ae(t0)2] ~ +/~ le(to)l - e(to)Nd(to) "
W(y) - ~ Ilzll ~ . (42)
Following Lemma 2, we utilize the right-most inequality
P r o o f . Let the auxiliary function P(t) C ]~ be defined as in (40) to define the region T~ as
follows
T~ ~ { y E IE3 }i}y]l< p - l ( 2 ~V~-~3k~)}. (43)
P(t) ~- ~ b -
f2 L(T)d7 (33)
From (34) and (40), we know V(t,y(t)) E £ ~ ; hence,
where (b and L(t) were defined in Lemma 1. It is easy to e(t), r(t) e £ ~ . From (10), we then know ~(t) e £ ~ .
see that the use of Lemma 1 ensures P(t) >~ O. Continuing Using (7) and (8), we can conclude that x(t), ~(t) e £ ~ .
with the proof, we define the function V(t, y)" I~+ × l~3~ We now know from (6) that re(x), f(x) E £ ~ . Finally,
l~+ as follows we can use (4) to show that u(t) E £ ~ .
Using the above boundedness statements, it is clear
V a le2 lm(x)r2 (34) from (18), (21), and (42) that l~V(y(t)) e £.00, which is a
= ~ +2 +P
sufficient condition for W(y(t)) being uniformly continu-
where y(t) E I~3 is defined as ous. If we define the region S as follows

y = [z T v/-fi]T (35) s~ { yez~lw~(~)<al ( p-~(2~) , (44)

and z(t) was defined in (21). Note that (34) can be we can now invoke Lemma 2 to state that lim 7 IIz(t)II 2 -
t---*(x)
bounded as
0 Vy(to) C S. From (21) we then know that lim r(t) = 0
' t----~oo
~ Ilyll = ~ v ~ ~(x)llyll = (36) Vy(to) E S. Finally, the structure of (10) allows us to
conclude that [17] lim e(t) ~(t) = 0 Vy(to) E S.
t---*(:x~
where (31) has been utilized. Note that the region of attraction in (44) can be made
After taking the time derivative of (34) and substitut- arbitrarily large to include any initial conditions by in-
ing from (10) and (18), we obtain creasing the control gain ks (i.e., a semi-global type of
stability result). Specifically, we can use the right-most
= - a e 2 - r 2 + riV - ksr 2 + r (Nd --/3sgn(e)) -- L
equality in (41) and the definition given by (44) to calcu-
= - a e 2 - r 2 + r.N - ksr 2 (37) late the region of attraction as follows

upon use of (22). After applying (20) and (21), we can w~(y(to)l < ~ (p-~(2 ~57~k~))~
upper bound the right-hand side of (37) as follows
A1 -1(24A31%) (45)
< -~ Ilzll ~ + [1~1p(llzll)Ilzll- k, I~1~]. (38) Jly(t0)l] < A2(x(to)) p

After completing the squares on the bracketed term in which can be rearranged as
(38), we obtain
( I ks> 1 P 2 ( ~ A2(x(t°)) ) (46)
p~(llzll) ilzlj~ (39) ~ ~ Iry(t0)rt .
~- A3- 4k~
By using (10) and (35), we can write an explicit expres-
From (39), we can state that sion for ]]y(to)]l as follows
< -'Yiizii 2 for
]]y(t0)lJ = v/e2(t0) + (~(t0) + ae(t0)) 2 + P(to). (47)
k~ > ~ 1p 2 (llzll) (or Ilzll<p- 1(2 ~X/-A~3ks))(40)
From (4), (8), and the fact that u ( t o ) = 0, we have that
where 7 E I~ is some positive constant. ~(to) = 5cd(to) + m-l(x(to))f(x(to)); (48)

BB
hence, (47) can be rewritten as Let xd(t) E ]RTM be the reference trajectory that is con-
tinuously differentiable up to its (n + 2)th derivative such

~ e2(to)+ [k2]d(to)+rn-l(x(to))f(x(to))+ that x(~)(t) E £ ~ f°r i = O , l , . . ,n+2" (54)


Ily(t°)ll = de(to) 2 +/3 le(t0)l - e(to)Nd(to)
(49) To quantify the control objective, we define the tracking
where (3a) and (25) were used. The condition given by error el (t) E IRTM as follows
(30) and (32)directly follows from (46) and (49). I
e l - - Xd -- x . (55)
R e m a r k 3 The result of (29) has an interesting implica-
tion on the "steady-state" behavior of the proposed control As before, our objective is to obtain asymptotic track-
law. Namely, after applying (29) to (~), we have ing with a continuous control law with full-state feed-
back (i.e., x (i), i = 0, ..., n - 1 are assumed measur-
u(t) = m(xd)iCd + f (xd) as t ~ co. (50) able) and only using (54) and norm-based, inequal-
(n--i)
ity bounds on the functions M(i)(xd,:kd, ...,x d ) and
That is, the control tends to behave like an exact-model- f(i)(Xd, ±d, ..., X(dn--1)), i = 0, 1,2.
knowledge, nonlinear feedforward controller as time ap-
proaches infinity. This indicates that the proposed control
law has the ability to learn the unknown system model. 3.1 Open-Loop Error System Develop-
ment

3 Extension to Higher-Order We begin by defining the following auxiliary error signals


Multi-Input Systems ~(t) E ]RTM, i = 2, 3, ..., n
A
e2 = e l -t- e l
In this section, we discuss the extension of the proposed ix o
e3 = e2 zr-e2 - t - e l
control design to higher-order, multi-input nonlinear sys- ix •
e4 = e3 -t- e3 -t- e2
tems. Specifically, we consider a system of the form
: (56)
ei = e i - 1 -Jr-e i - 1 + e i - 2
M ( x , ±, ..., X(n--1))X (n) ~- f(x, ±, ...,x (~-1)) = u (51)
A •
where x(i)(t) E ]Rm, i = 0 , . . . , n - 1 are the system e n = C a - 1 -t- e n - 1 ~ e n - 2

states, (-)(~)(t) denotes the ith derivative with respect


where el (t) was defined in (55).
to time, u(t) E IRm is the control input vector, and
M(x,x,...,x (n-l)) E ~rnxrn and f ( x , x , . . . , X (n-l)) E
]Rm are uncertain nonlinear functions which satisfy the 3.2 Control Law
following assumptions: 2
Based on subsequent stability analysis, we design the con-
A s s u m p t i o n M 1 . The matrix M(.) is symmetric and trol input u ( t ) a s follows
positive-definite, and is bounded by
u(t) (Ks + I.~)en(t) - (Ks + Im)en(t0)+
m ]]t~][2 ~< ~TM(.)~ < m(.)]l~[I 2 V~ E N "~ (52)
ftl [(Ks + Im)hen(~')+
where rn E IR denotes a positive constant, and r~gn(e~(~))l d~ (57)
~ ( x ) E R denotes a positive, nondecreasing func-
tion. where Ks, F and A E ~mxm a r e positive-definite, diag-
onal, control gain matrices, I m E ~m×rn represents the
A s s u m p t i o n M 2 . The functions M(.) and f(.) are m x m identity matrix, and sgn(.) is defined as follows
second-order differentiable such that

M(.), 1VI(.), l~I(.) E / 2 ~ if x, ~k, ..., x (n-l) E / 2 ~ sgn(~) ~ [ sgn(~l) sgn(~2) ... sgn(~m) ]T
= . (bs)
f (.) , t (.) ,]~ (.) E E ~ if x, x, ..., x (n-l) E £cxD.

(53) (Note that u(t0) = 0). After taking the time derivative
of (57), we have
2The symbol "." is used from now on to indicate the function's
dependency on x , ± , ...,x ( n - l ) . Note t h a t the comments made in
Remark 1 are also applicable to M ( - ) and f (.).
fl = (Ks + Im)(~,~ + A e n ) + Fsgn(en(~-)). (59)

56
3.3 Stability Analysis [11] F.L. Lewis, S. Jagannathan, and A. Yesildirak, Neural
Network Control of Robot Manipulators and Nonlinear
The main result of this section is delineated by the fol- Systems, Taylor & Francis, 1999.
lowing theorem.
[12] W. Lin and C. Qian, "Adaptive Control of Nonlinearly
T h e o r e m 2 The control law of (5"/) ensures that all sys- Parameterized Systems: A Nonsmooth Feedback Frame-
tem signals are bounded under closed-loop operation and work," IEEE Trans. Automatic Control, Vol. 47, No. 5,
that pp. 757-774, May 2002.
lim ei(t) = 0 , i=0,1,...,n, (60) [13] W. Messner, R. Horowitz, W.W. Kao, and M. Boals, "A
t---~ (x)

New Adaptive Learning Rule," IEEE Trans. Automatic


provided that the elements of the control gain matrices F
Control, Vol. 36, No. 2, pp. 188-197, Feb. 1991.
and Ks are selected sufficiently large. 3
[14] R. Ortega and A. Astolfi, "Nonlinear PI Control of Un-
P r o o f . Please refer to [24] for details. certain Systems: An Alternative to Parameter Adapta-
tion," Proc. IEEE Conf. Decision and Control, pp. 1749-
1754, Orlando, FL, Dec. 2001.
References
[15] Z. Qu, Robust Control of Nonlinear Uncertain Systems,
[1] A. Annaswanny, F. Skantze, and A. Loh, "Adaptive Con- Wiley, 1998.
trol of Continuous-Time Systems with Convex/Concave
[16] Z. Qu and J.X. Xu, "Model-Based Learning Controls and
Parametrizations," Automatica, Vol. 34, No. 1, pp. 33-
Their Comparisons Using Lyapunov Direct Method,"
49, Jan. 1998.
Asian J. Control, Vol. 4, No.l, pp. 99-110, Mar. 2002.
[2] G. Cambion and G. Bastin, "Indirect Adaptive State [17] M.S. de Queiroz, D.M. Dawson, S.P. Nagarkatti, and F.
Feedback Control of Linearly Parameterized Nonlinear Zhang, Lyapunov-Based Control of Mechanical Systems,
Systems," Int. J. Adaptive Control and Signal Process- Birkh~iuser, 1999.
ing, Vol. 4, pp. 345-358, Sept. 1990.
[18] S. Sastry and M. Bodson, Adaptive Control: Stability,
[3] I. Kanellakopoulos, P.V. Kokotovic, and R. Marino, "An Convergence, and Robustness, Prentice Hall, 1989.
Extended Direct Scheme for Robust Adaptive Nonlinear
Control," Automatica, Vol. 27, No. 3, pp. 247-255, Mar. [19] S.S. Sastry and A. Isidori, "Adaptive Control of Lineariz-
1991. able Systems," IEEE Trans. Automatic Control, Vol. 34,
No. 4, pp. 405-412, Apr. 1989.
[4] R. Ellis and D. Gulick, Calculus with Analytic Geometry,
[20] A. Serrani, A. Isidori, and L. Marconi, "Semiglobal
Saunders College Publishing, 1994.
Nonlinear Output Regulation with Adaptive Internal
[5] A. Fradkov, R. Ortega, and G. Bastin, "On Semi- Model," IEEE Trans. Automatic Control, Vol. 46, No.
Adaptive Control of Convexly Parametrized Systems," 8, pp. 1178-1194, Aug. 2001.
Int. J. Adaptive Control and Signal Processing, Vol. 15,
[21] J.J. Slotine and W. Li, Applied Nonlinear Control, Pren-
No. 5, July 2001.
tice Hall, 1991.
[6] P.A. Ioannou and J. Sun, Stable and Robust Adaptive [22] B. Yao and M. Tomizuka, "Adaptive Robust Control of a
Control, Prentice Hall, 1996. Class of Multivariable Nonlinear Systems," IFAC World
Congress, Vol. F, pp. 335-340, San Francisco, CA, 1996.
[7] H. Khalil, Nonlinear Systems, Prentice Hall, 1996.
[23] M.S. de Queiroz, J. Hu, D. Dawson, T. Burg, and S.
[8] T. Koshy, Fibonacci and Lucas Numbers with Applica- Donepudi, "Adaptive Position/Force Control of Robot
tions, John Wiley & Sons, 2001. Manipulators without Velocity Measurements: Theory
and Experimentation", IEEE Transactions on Systems,
[9] E.B. Kosmatopoulos and P.A. Ioannou, "Robust Switch-
Man, and Cybernetics, Vol. 27, No. 5, Oct., 1997, pp.
ing Adaptive Control of Multi-Input Nonlinear Systems,"
796-809.
IEEE Trans. Automatic Control, Vol. 47, No. 4, pp. 610-
624, Apr. 2002. [24] B. Xian, D.M. Dawson, M.S. de Queiroz, and J. Chen,
"A Continuous Asymptotic Tracking Control Strategy
[10] M. Krstic, I. Kanellakopoulos, and P. Kokotovic, Nonlin- for Uncertain Multi-Input Nonlinear Systems," Techni-
ear and Adaptive Control Design, John Wiley ~ Sons, cal Report ME-MS1-03, Department of Mechanical En-
1995. gineering, Louisiana State University, Feb. 2003.
3The sufficient conditions for the control gain matrices are ex-
plicitly delineated in [24]. Appendices are available upon request.

57

View publication stats

You might also like