IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 39, NO.

12, DECEMBER 1994

Y. Bar-Shalom and K. Birmiwal, “Variable dimension filter for maneu- variables to reach the sliding mode, and the asymptotic convergence
vering target tracking,” IEEE Trans. Aero. Electr. Sys.. vol. AES-18, pp. of error dynamics can then be obtained on the linear sliding mode
621-629, 1982.
[ I]-[6]. To get fast error convergence on the sliding mode, however,
P. L. Bogler, “Tracking a maneuvering target using input estimation,”
IEEE Trans. Aero. Electr. Sys., vol. AES-23, pp. 298-310, 1987. the sliding mode parameters must be chosen such that the poles
S. S. Lim and M. Farooq, “Maneuvering target tracking using jump pro- of the sliding mode dynamics are far from the origin on the left-
cesses,” in Proc. 30th IEEE Con$ Decis. Contr., vol. 2, pp. 2049-2054, half of the s-plane. This will, in turn, increase the gain of the
1991. controller. Considering the saturation property of control input signals
G. C. Goodwin and K. S. Sin, Adaptive Filtering, Prediction and
Control. Englewood Cliffs, NJ: Prentice-Hall, 1984. in practical robot control, a sliding mode controller with high gain is
C. F. N. Cowan and P. M. Grant, Eds., Adaptive Filters. Englewood undesirable. On the other hand, a linear sliding mode technique can
Cliffs, NJ: Prentice-Hall, 1985. guarantee only the asymptotic error convergence on the sliding mode,
A. B. Kurzhanskii, “The identification problem-theory of guaranteed and therefore error dynamics can not converge to zero in a finite time.
estimates,’’ Autom. Rem. Contr., vol. 52, pp. 447465, 1991.
In this paper, a robust MIMO terminal sliding mode control
A. R. Rao and Y.-F. Huang, “Recent developments in optimal bounding
ellipsoidal parameter estimation,’’ Math. Comp. Simul., vol. 32, pp. scheme is developed for n -link rigid robotic manipulators based
515-526, 1990. on [11]-[13]. It is shown that an MIMO terminal switching plane
S. A. Kassam and H. V. Poor, “Robust techniques for signal processing: variable vector is first defined, and the relationship between the
A survey,” in Proc. IEEE, vol. 73, pp. 433481, 1985. terminal switching plane variable vector and system error dynamics
R. J. Barton and H. V. Poor, “An RKHS approach to robust L2
estimation and signal detection,” IEEE Trans. Inform. Theory, vol. IT-36, is investigated. By using the MIMO terminal sliding mode techniique
pp. 485-501, 1990. and a few structural properties of rigid robotic manipulators, a robust
A. Albert, Regression and the Moore-Penrose Pseudoinverse. New controller can then be designed. Unlike conventional linear sliding
York: Academic Press, 1972. mode control schemes, the switching plane variable vector in this
D. G. Luenberger, Optimizarion by Vector Space Methods. New York:
paper has a nonlinear term of the velocity error. By suitably designing
Wiley& Sons, 1969.
R. Sh. Liptser and A. N. Shiryayev, Sratistics ofRandom Processes, vol. the controller, the switching plane variables can reach the termlinal
2. New York: Springer-Verlag. 1977. sliding model in a finite time, and the output tracking error can lthen
A. V. Borisov and A. R. Pankov, “Optimal filtering in uncertain- converge to zero in a finite time on the terminal sliding mode. It is
stochastic systems,” Tech. Rep. 91 -27, Industrial Operations & Eng. also shown that this scheme is more practical because the gain 0 1 the
Dept., Univ. Michigan, 1991.
terminal sliding mode controller can be significantly reduced with
respect to the high gain of linear sliding mode controllers in practical
situations where the sampling interval is non zero.
Similar to the linear sliding model technique, strong robustness
with respect to large uncertain dynamics can be obtained by using the
A Robust MIMO Terminal Sliding Mode proposed control scheme. Also, the controller design is simple iri the
Control Scheme for Rigid Robotic Manipulators sense that only a few uncertain bounds based on structural properties
of robotic manipulators are used in the controller parameter designs.
Man Zhihong, A. P. Paplinski, and H. R. Wu The paper is organized as follows: In Section 11, an n-link rigid
robotic manipulator model and i t s a few useful structural properties
are formulated, and an MIMO terminal nonlinear sliding mode is
Abstruct- In this paper, a robust multi-input/multi-output(MIMO) defined to describe desired error dynamics. In Section 111, a robust
terminal sliding mode control technique is developed for d i n k rigid
robotic manipulators.It is shown that an MIMO terminal switching plane MIMO terminal sliding mode control scheme is developed for rigid
variable vector is first defined, and the relationship between the terminal robotic manipulators, the stability of error dynamics and robustness
switching plane variable vector and system error dynamics is established. with respect to uncertain dynamics are discussed in detail, and an
By using the MIMO terminal sliding mode technique and a few structural advantage of the proposed scheme in its practical applications is also
properties of rigid robotic manipulators, a robust controller can then be
designed so that the output tracking error can converge to zero in a finite remarked. In Section IV, a simulation for a two-link rigid robotic
time, and strong robustness with respect to large uncertain dynamics can manipulators is performed in support of the proposed control scheme.
be guaranteed. It is also shown that the high gain of the terminal sliding Section V gives conclusions.
mode controllers can be significantly reduced with respect to the one of
the linear sliding mode controller where the sampling interval is nonzero.
11. PROBLEMFORMULATION
The dynamics of an 11-joint robotic manipulator can be described
I. INTRODUCTION
by the following second-order nonlinear vector differential equation
Sliding mode control is one of the most important approaches
to handling systems with large uncertainties, nonlinearities, and M ( q ) i + F ( q . 4)(i + G ( q )= u ( t ) (2.1)
bounded external disturbances. Generally, in most of sliding mode where q ( t ) is the U x 1 vector of joint angular positions, ~ ( tis )the
control schemes for multi-inputlmulti-output (MIMO) systems, an n x 1 vector of applied joint torques, M ( q ) is the 71 x n symmetric
MIMO linear sliding mode is first designed to describe the desired positive-definite inertia matrix, F ( q . 4)4i s the n x 1 vector of Coriolis
system error dynamics, a robust controller drives the switching plane and centrifugal torques, and G ( q )is the 71 x 1 vector of gravitational
torques. Further, we assume that vectors ( I , q , and 4 are measurable
Manuscript received August 24, 1993; revised March 4, 1994.
M. Zhihong is with the Department of Computer and Communication (see Fig. 1).
Engineering, Edith Cowan University, WA 6027, Australia. Defining .r = (q’ . ;1 )’ , expression (2.1 ) can be written as
A. P. Paplinski and H. R. Wu are with the Department of Robotics and
Digital Technology, Monash University, Clayton, VIC 3 168, Australia.
IEEE Log Number 9405659.

0018-9286/94$04.00 0 1994 IEEE

Authorized licensed use limited to: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on December 05,2024 at 01:37:39 UTC from IEEE Xplore. Restrictions apply.
 EEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 39, NO. 12, DECEMBER 1994 2465

where
Ad= [< - E ~ , * * * , E P -en,
, O,-**,OlT. (2.13)
Remark 2.2: Using expressions (2.9), (2.12), and (2.13), the
Fig. 1. An n-link rigid robotic manipulator model.
MIMO terminal switching plane variable vector in expression (2.7)
can be written into the form
s = Cd
= C(e + AE)
= Ce + C1(F - E) (2.14)
Fig. 2. A desired reference model. where
z = [ E : . .. < I T . (2.15)
The reference model for the plant to follow can be represented as It will be seen later that it is convenient to use expression (2.15) of
switching plane variable vector S in controller design and stability
analysis.
Remark 2.3: The ith element of S in expression (2.7) can be
written into the following form
where P = diag(P,), Q = diag(Q,), and BI = diag(b,) (1 5 i 5
n) are constant matrices which are chosen such that the reference s, = c,,< + i*. (2.16)
model (2.3) is stable. Vectors T , q,, and qm are assumed to be
measurable (see Fig. 2). Similar to the conventional sliding mode control technique, if the
Defining E = q - q,-, e = [ e T , and using expressions (2.2) controller is designed such that s. (i = 1 , . ,n) converge to zero,
and (2.3), we obtain the error differential equation as then we say that the switching plane variables s, (i = l , . . . , n )
reach the terminal sliding mode
1 = A,e + Bh(q, q, U , r ) (2.4) C,*E~+i,=O (i=l,*..,n). (2.17)
where B = [0, TIT and It has been shown in Zak [12], [13] that E, = 0 is the terminal
attractor of the system (2.17). Let the initial value of E , at time t = 0
h(q, i , U , T ) = M(q)-lfJ + hl(% i , r ) (2.5) be E,(O) and parameter p be chosen as shown in Remark 2.1, then
the relaxation time t , for a solution of system (2.17) is given as

Expression (2.18) also means that, on the terminal sliding mode in

To design robust control system with the error convergence in a (2.17), the output tracking error converges to zero in a finite time.
finite time, we define the following MIMO terminal switching plane The details on the terminal attractor and its applications can be found
variable vector in [12] and [13].
Remark 2.4: For the simplified analysis, matrix CZ in (2.8) is
S=CE (2.7) chosen as an unity matrix. Matrix CZ,however, can be chosen as
a different diagonal matrix for the improvement of the convergence
where
of the error dynamics on the terminal sliding mode.
1 For further analysis, the followinguncertain bounds for n-link rigid
c =[C, C2] = (' ... c"" ... (2.8)
robotic manipulators in expression (2.1) are assumed to be known.
1) Arnin(M(q1-l) 2 a1 (2.19)
1
2) llMw-'ll 2 a2 (2.20)
E ..E; il ..
=[E:. (2.9) + +
3) llF(q.,ai G(q>ll< b l bZllQll+ b311i112 (2.21)
Remark 2.1: In (2.9), p = p l / p 2 , where positive integers p l and where a l , az, bl, b 2 . and b3 are positive numbers.
Remark 2.5: According to the characteristics of industrial robotic
p z are selected such that
manipulators, the above assumptions are reasonable. Assumption
p2 = ( 2 m + I), m = 1, 2 , . . . (2.10) A.2.1) has been used in Leung et al. [20], and Assumptions A.2.2)
and A.2.3) have been used ih Singh [8]. Also, an estimation method
for the uncertain bound parameters in Assumptions A.2.1kA.2.3) has
pz > P l . (2.1 1) been developed in Grimm [9] by neglecting some dynamics.
It is easy to shown that the selections of p1 and p z in (2.10) and The objective of this paper is to design a robust tracking controller
(2.1 1) can guarantee 0 < p < 1 and the tracking error E, can then using the MIMO terminal sliding mode technique based on uncertain
converge to zero on the terminal sliding mode in a finite time for all bounds in assumptions A.2.1kA.2.3) instead of the upper and the
bounded initial conditions. lower bounds of all unknown system parameters so that, for any
Vector E in expression (2.7) can also be written into the following bounded uncertainties in parameten of robotic systems, the error
'form dynamics can be driven into the terminal sliding mode in a finite
time. On the ferminal sliding mode, the error dynamics can then
d=e+Ad (2.12) converge to zero in a finite time.

Authorized licensed use limited to: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on December 05,2024 at 01:37:39 UTC from IEEE Xplore. Restrictions apply.
 2466 IEEE TRANSACTIONS ON AUTOMATIC CONTKOL. VOL. 39, NO 12, DhCEMBER 1994

111. CONTROLLER
DESIGN On the terminal sliding mode
Theorem 3.1: Consider the error dynamics in expression (2.4) with Ci. = 0 (3. I 1)
assumptions in (2.19)-(2.21). If the control input vector is designed
such that the error dynamics satisfy (2.17), then the output tracking error
converges to zero in a finite time.
(3.1) Remark 3.1: It can be see that, unlike the conventional linear
sliding mode control schemes in [I]-[6], the output tracking error
can converge to zero in a finite time by using the proposed control
scheme due to the fact that the output tracking error can be driven into
the terminal sliding mode in a finite time, and the error dynamics can
then converge to zero in a finite time on the terminal sliding mode.
Remark 3.2: The proposed terminal sliding mode control scheme
E, = diag (pty-'. . . . .ye:-' )i (3.3) has strong robustness with respect to large parameter uncertainties
because only five control parameters are adjusted in the controller,
and and the adjustable parameters depend only on the uncertain bounds
P =PI/PZ and PZ > P I 2 ( P Z + 1112 in expressions (2.19)-(2.21).
Remark 3.3: On the terminal sliding mode in expression (2.15),
(see Remark 3.3) (3.4) the signal vector t, in expression (3.3) can be written as
then the output tracking error vector converges to zero in a finite time. e,. = d i a g ( p ~ ~ - ' : . . . p t ~ - ' ) i
Proof? Consider the following Lyapunov function
1
I- = -srs.
2 (3.5)

Differentiating I' with respect to time, we have

(3.12)
= 51 3
i-
Expression (3.4) shows that, although mathematically the positive
= SI (C;. + ClC, - Cli) number p in (2.17) or (2.18) satisfies (2.10) and (2.1 1) to guarantee
= +~
s'[cA,,,~ ~ / i ( i qu . . r ) + ~ l c-, C,~I the terminal convergence of variable c , , the number p must satisfy
= s ~ [ - P Q-, ~ 1+ ,, ~ ( q ) - ' u- Blr + ~ 1 6 , (3.4) in this control scheme in the sense that the signal vector 6,.
in (3.12) or (3.3) must be bounded as the output tracking error t l
- J ~ ( Y ) - ' ( Fi)Y+
( Y . G(Y))]. (3.6)
converge to zero on the terminal sliding mode.
Let Remark 3.4: In [19], a linear sliding mode control scheme using
the same assumptions on uncertain bounds in A.2.1)-A.2.3) was
(3.7) developed. In [19, (7.16)], the term IIClill, on the terminal sliding
mode, can be expressed as
and considering Assumption A.2.1), we have kl ( t ) 2 1.

L1 21
112
Using control law (3.l), expression (3.6) can then be written as IlClill = C((P,F,) (3.13)

and the definition of the sliding mode parameter matrix CI can be

found in (7.7) and (7.8) of [19]. 119, (7.7), (7.8)l
Similarly, the term IIC1t,.IJin expression (3.2), on the terminal
sliding mode, can be expressed as
r
(3.14)

(3.8) When the sampling interval is nonzero, the output tracking errors will
persist around the origin of the error space after the trajectories reach
their vicinity, and therefore the ideal error convergence in both [I91
and this paper cannot be obtained. If the tracking errors in these two
control schemes are required to reach the vicinity of the origin at the
same time, however, the linear sliding mode parameters c,1 in [19]
must be chosen to satisfy the following relationship

> (bi + b z l l ~ l+l h11411') - llF4 + GI1 > 0. (3.9) C,I >> c,, (3.15)

Then and then the following inequality can often be satisfied

< - k l a 2 h ( t ) l l s J I < 0 llsll # 0.
i- (3.10) IlClill >> IIClfrll. (3.16)

Considering the fact that b ( t ) is greater than a positive number by Therefore, by comparing expression (7.16) of [I91 with (3.2) of this
suitably selecting the bound parameters bl , b l , and b 3 , expression section, we can find that the control gain has been significantly
(3.10) means that switching plane variable vector S converges to reduced by using this scheme. This feature can also be seen from
zero in a finite time. the simulations in following section and [19].

Authorized licensed use limited to: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on December 05,2024 at 01:37:39 UTC from IEEE Xplore. Restrictions apply.
 IEEE TRANSACTIONS ON AUMMAnC CONTROL. VOL. 39, NO. 12, DECEMBER 1994 2467

141 7

i
1
E 04t/ i

oz
1

-021 I
0 1 2 3 4 5 - 7 s

Time t (s) Time t (6)

(b)

.-laO~ '
1 2 3 4 5 3 7
I
8
U
-,+ . ,
1 2 7 4 5 6 7 8
"e 1 b)
mm I (SI

(C)

Fig. 3. (a) The output trackings of joint 1 and joint 2, (b) The output tracking errors of joint 1 and joint 2, (c) The control inputs of joint 1 and joint 2.

Remurk3.5: We have proved that the output tracking error can dynamic equations are given as [2]
converge to zero in a finite time by the use of the proposed control
scheme. The control law (3.1) is discontinuous across the sliding
mode surfaces S = 0, however, which may excite undesired high
frequency dynamics [4]. To eliminate the effects of the chamrings,
we use the following boundary layer control law in place of the
discontinuous control law in expression (3.1)

(3.17) where
where 6 > 0.
By using the above boundary layer control law, we can guarantee
the attractiveness of the boundary layer. For the region inside the
boundaries, the ultimate boundedness of the error dynamics can be
guaranteed to within any neighborhood of the boundary layer [3], [4].

IV. SIMULATION EXAMPLE

To illustrate the control schemes proposed in this paper, a simula-
tion example for a two-link robotic manipulator is studied. The full

Authorized licensed use limited to: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on December 05,2024 at 01:37:39 UTC from IEEE Xplore. Restrictions apply.
 2468 IEEE TRANSACTIONS ON AUTOMATIC CONTROL VOL 39 NO 12 DECEMBER 1994

II.--, -_- - - ~ p ~ _ p _

I
-- - . ~

'.

'\
'--.
.-_

Time I (s) Time I (I)

(C)
Fig. 4. (a) The output trackings of joint 1 and joint 2, (b) The output tracking errors of joint 1 and joint 2, (c) The control inputs of joint I and
joint 2 using boundary layer controller.

The parameter values are Since we are interested in trajectory tracking and hope that the
r l =1 m, r2 = 0.8 m transient response is determined entirely by the sliding motion, we
consider a situation characterized by the same initial values of both
--3
J 1 -: kg.m, ,J2 = 5 kg.m
the reference model state and the plant state. In this simulation, we
1111 =0.5 kg, ~ t 1 2= 1.5 kg. pick up the initial values of . r ( f ) and . r r n ( f to
) be
A reference model for the manipulator to follow is given by
s(0)= .r,(O) = [0.2 2 0 017.
The parameters of the uncertain bounds in (2.17)-(2.19) are chosen as
where
(11 = 0.1. (12 =2
b l = 2. bp = 1. b:j = 2.

0 o 1 0 1 0 01 Terminal sliding mode is prescribed as

--In>=
-4 0 -0
- 1 B",, =
O h
F, + il =0
('
0 -4 0 -51
Oh
€2 + =o.
i2

Fig. 3 shows the output trackings, tracking errors, and input torques
and r ( t ) = [5 51' fort > 0. by the use of the control law (3.1). It can be seen that the effects of

Authorized licensed use limited to: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on December 05,2024 at 01:37:39 UTC from IEEE Xplore. Restrictions apply.
 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 39, NO. 12, DECEMBER 1994 2469

system uncertaintiesare eliminated, and good tracking performance is S . N. Singh, “Adaptive model following control of nonlinear robotic
achieved. To eliminate chatterings, we implement the boundary layer systems,” IEEE Trans, Automat. Contr., vol. AC-30, pp. 1099-1100,
control law (3.17). Here we take 6 = 0.01. Good system performance 1985.
W.M. Grimm, “Robot non-linearity bounds evaluation techniques for
is shown in Fig. 4. As can be seen from these figures, the chattering robust control,” Int. J. Adnptive Contr. Sig. Process., vol. 4, pp. 501-522,
is eliminated. 1990.
-, “Robustness analysis and synthesisof model-based robot control,”
VDI-Fortschrittberichte, Reihe 8, No. 202, VDI Duessedorf, 1990.
V. CONCLUSIONS S . T. Venkataraman and S. Gulati, “Control of nonlinear systems
In this paper, a robust control scheme for rigid robotic manipulators using terminal sliding modes,” in P m . Amer. Contr. Con$, 1992, pp.
89 1-893.
using the MIMO terminal sliding mode technique has been proposed. M. Zak, ‘Terminalattractorsfor addressablememory in neural network,”
The main contributions of this paper are that an MIMO terminal Physics Lett. A, vol. 133, no. 1 0 , pp. 18-22, 1988.
sliding mode is defined, and a robust terminal sliding mode control -, “Terminal attractors in neural networks,” Neural Net., vol. 2, pp.
scheme for n-link rigid robotic manipulators is developed with the 259-274, 1989.
result that the output tracking error can converge to zero in a finite Man Zhihong and M. Palaniswami, “A robust tracking control for rigid
robotic manipulators,” IEEE Trans. Automat. Contr., 1993, to appear.
time. In addition, the robot control systems using the proposed scheme -, “A variable structure model reference adaptive control for non-
have a strong robustness property not only because on the sliding linear robotic manipulators,” Int. J. Adaptive Contr. Sig. Process., 1993,
mode, the error dynamics is insensitive to uncertain dynamics, but to appear.
also because only three uncertain bounds based on the structure -, “Decentralized three segment nonlinear sliding mode control
for robotic manipulators,” in P m . IEEE In?. Workshop on Emerging
p r o w e s of rigid robotic manipulators are used in controller design. Technologies and Factory Automation, Melbourne, 1992, pp. 607-612.
It has also been remarked that this scheme is more practical in -, “A robust adaptive tracking control scheme for robotic manipu-
the sense that the gain of the terminal sliding mode controller lators with uncertain dynamics,” in P m . IEEE Int. Con$ Systems Eng.,
can be significantly reduced with respect to the ones of linear Japan, 1992, pp. 127-130.
sliding mode control schemes developed in [ 191, where the sampling -, “A robust tracking controller for robotic manipulators,” in
Proc. IEEE Region IO Int. Con$-TENCON ’92, Melbourne, 1992,
interval is nonzero. A few problems for the practical implementation pp. 953-957.
of this scheme, however, have been noted. Like all other control M. Zhihong, “Robust variable variable structurecontrol for rigid robotic
techniques, the ideal error convergencecannot be obtained in practical manipulators,” Ph.D. dissertation,Dept. Elec. and Electronic Eng., Univ.
control systems where sampling interval is nonzero. To implement Melbourne, 1993.
T. P. bung, Q.J. Zhou, and C. Y.Su, “An adaptive variable structure
this scheme, some nonlinear electronic hardware to deal with the model following control design for robotic manipulators,” IEEE Trans.
nonlinear function 6; needs to be built. The advantage that the Automat. Contr., vol. 36, pp. 347-352, 1991.
proposed terminal sliding mode controller has smaller control gain
with respect to linear sliding mode controller, however, cannot be
counteracted by the above factors. In this paper, the terminal sliding
mode technique is used to control only the second order robotic
systems. The research on design of high-order terminal sliding mode
control systems, however, is under author’s investigations based on A Common Lyapunov Function for Stable
[12] and [13]. LTI Systems with Commuting A-Matrices

ACKNOWLEDGMENT
Kumpati S. Narendra and Jeyendran Balakrishnan

The authors would like to thank the anonymous reviewers for their
Abstmct--The paper demonstrates that a common quadratic Lyapunov
valuable comments and suggestions. function exists for all Iinearsystemsoft h e f o r m i = A p , i = 1,2,.. . ,N,
where the matrices Ai are asymptotically stable and commute pairwise.
REFERFiNCES This in turn assures the exponential stability of a switching system
i ( t )=A ( t ) z ( t )where A@)switches between the above constant matrices
K-K.D. Young, “Design of variable structure model following control Ai.
system,” IEEE Trans. Automat. Contr., vol. AC-23, pp. 1079-1085,
1978.
- ,“A variable structure. model following control design for robotics I. INTRODUCTION
applications,”IEEE Trans. Automat. Contr., vol. 4, pp. 556-561, 1988. In recent years, the scope of control theory is being enlarged to
J-J. E. Slotine, “Sliding controller design for nonlinear systems,” Int. J.
Contr., vol. 40,pp. 421434, 1984. include intelligent control systems. One of the main features of such
J-J. E. Slotine and S . S . Sastry, ‘Tracking control of nonlinear system intelligent control systems is the systematic application of the idea
using sliding surface with application to robotic manipulators,” Int. T. of switching between different controllers [l], [2]. One of the first
Contr., vol. 38, pp. 46-92. 1983. questions to be resolved in this context is that of the stability of the
K. S . Yeung and Y. P. Chen, “A new controller design for manipulators
using the theory of variable structure. systems,” IEEE Trans. Automat.
overall system.
Contr., vol. AC-33, pp. 200-206, 1988. Many of the stability problems that arise in intelligent control
L. C. Fu and L. T. Liao, “Globally stable robust tracking of nonlinear systems can be addressed by consideringthe followingbasic problem:
systems using variable structure control and with an application to
a robotic manipulator,” IEEE T m . Automat. Contr., vol. 35, pp. Manuscript received November IO, 1993; revised March 18, 1994. This
1345-1350, 1990. work supported by National Science Foundation Grant ECS-9203928.
C. M. Dorling and A. S . I. Zinober, ‘”ho approaches to hyperplane The authors are with the Center for Systems Science, Department of
design in multivariable variable structurecontrol systems,” Inr. J. Contr., Eleceical Engineering, Yale University, New Haven, CT 06520-8267 USA.
vol. 44, pp. 65-82, 1986. IEEE Log Number 9405660.

0018-9286/94$04.000 1994 IEEE

Authorized licensed use limited to: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on December 05,2024 at 01:37:39 UTC from IEEE Xplore. Restrictions apply.