A Set of Decentralized PID Controllers For An N - Link Robot Manipulator
A Set of Decentralized PID Controllers For An N - Link Robot Manipulator
A Set of Decentralized PID Controllers For An N - Link Robot Manipulator
For correspondence
405
406
torques are not isolated from each other due to the coupling caused by nonlinear inertial terms.
Among the works reported in this area are Narendra & Oleng (2002), Tang et al (2000), Tarokh
(1996), Tang & Guerrero (1998), Wang & Wend (1999), Liu (1999) and Seraji (1989). Narendra
& Oleng (2002) have shown that in strictly decentralized adaptive control systems, it is theoretically possible to track desired outputs with zero error. Tang et al (2000) and Tarokh (1996) have
presented an asymptotically stable decentralized adaptive control scheme to enable accurate trajectory tracking. Tang & Guerrero (1998) obtained an extremely simple controller, consisting of
a linear state-feedback with an additional signal designed to compensate for the coupling among
the joints, parameter uncertainty and bounded disturbances. In Wang & Wend (1999) the dynamics of the subsystems are divided into two parts: a nominal system and uncertainties. Based on
the nominal system and the bounds on the uncertainties, the Riccati equation approach is used to
control the motion of robot manipulators. A set of nonlinear decentralized tracking error for each
subsystem is formulated so that the passivity property of robot dynamics can be used in the controller design (Liu 1999). For manipulator tracking tasks, decentralized approaches are not that
straight forward since the overall system cannot be decomposed into subsystems whose states
and control inputs are not totally decoupled from one another because of the inherent coupling
such as moment of inertia and Coriolis force. Several attempts have been made to control independently each robot joint attached to actuator by processing local measurements available from
that joint. As a result, how to improve the tracking performance of robots through decentralized
control is still an interesting topic in control literature.
In this paper, a class of decentralized stabilizing controllers is designed for an n-link robot
manipulator using Kharitonovs theorem and boundary stability condition for an interaction free
subsystem. The control objective is to achieve accurate tracking of desired joint trajectories. The
significant results of Siljak & Stipanovic (2000) demonstrate how the Linear Matrix Inequalities
(LMIs) formulation can be used to quadratically stabilize nonlinear interconnected system via
decentralized linear constant feedback laws. Motivated by the work of Siljak & Stipanovic, we
tried to establish based on LMI approach, how the designed set of decentralized controllers for
each subsystems can be utilized to stabilize the interconnected nonlinear systems.
The paper is organized as follows. In section 2, problem formulation and a set of decentralized
PID controller for interaction free subsystems have been developed using the basic principles
of Kharitonovs theorem and stability boundary condition. The stability analysis of composite
system (with interaction terms) using the designed decentralized controller has been studied in
section 3 based on LMI formulation. Simulation results are presented in section 4 to demonstrate
the effectiveness of the proposed control. Concluding remarks are given in section 5.
407
margins (equal to zero) are maintained along the stability boundary in a controller parameter
plane or parameter space.
(1)
is an n 1 Coriolis
where M() is an n n symmetric positive definite inertia matrix, V (, )
and centrifugal vector, G() is an n 1 gravity vector of the manipulator, is the n 1 vector
representing joint angular positions, and is the n 1 vector of applied joint torques. For
simplicity, it is denoted that
= V (, )
+ G () .
N (, )
(2)
due to unknown load on the manipulator and
There are uncertainties in M() and N (, )
unmodelled frictions. The following bounds are assumed on the uncertainties (Qu & Dawson
1996).
(i) There exist positive definite matrices Mu () and Ml () such that
Mu () M () Ml () > 0.
and a nonnegative function n max (, )
such that
(ii) There exist Nu (, )
N (, )
n max (, ).
Nu (, )
The state variables to be x1 = , x2 = and the control to be
.
u = Mu ()1 ( Nu (, )).
Then
x1 = = x2
= M()1 Mu ()u + M()1 (Nu (, )
N (, ))
x2 = = M()1 ( N (, )
1
1
(3)
= M(x1 ) Mu (x1 )u + M(x1 ) (Nu (x1 , x2 ) N (x1 , x2 )).
The state space representation of (3) is given by
x = Ax + B(u + f (x)u) + Bh(x),
(4)
where f (x) = M(x1 )1 Mu (x1 ) I , and h(x) = M(x1 )1 (Nu (x1 , x2 ) N (x1 , x2 ))
0 I
0
x1
A=
,B =
and x =
.
0 0
I
x2
Bf (x) is the uncertainty in the input matrix and in order to make its effect maximum f (x) is
taken as
f (x) = Ml (x1 )1 Mu (x1 ) I 0,
(5)
408
where Ml and Mu are the lower and upper bounds of inertia matrix that is found out as explained
in Lin & Brandt (1998). In order to make the system in the input decoupled form equation (5) is
written as
f (x) = daig(max ( f (x)),
(6)
(7)
(8)
(9)
where Q is a positive definite matrix. For the development of the decentralized control scheme,
it is convenient to view each joint as a subsystem of the entire manipulator system and state
variables are rearranged in (4) and are rewritten as
xi = Asi xi + Bni u i + Bi h i (x) ;
yi = Csi xi ,
i = 1, 2, . . . , n,
(10)
where
Bni = Bi + Bi max ( f (x)), f (x) = Ml (z)1 Mu (z) I 0,
u = Mu (z)1 ( Nu (x))
i
xi1
h(x) = M(z)1 (Nu (x) N (x)) , xi =
=
and z = [x11 x21 ........xn1 ]T
i
x
i2
0 1
0
Asi =
, Bi =
, Csi = 1 0 .
0 0
1
If the variation of f (x) is considered, its minimum value is zero. So the input matrix Bni in
equation (10) varies from Bi to Bni . Hence, the transfer function of the ith joint with nonlinear
interaction terms h i (x) = 0, i 1,2,. . . n becomes
G i (s) =
1
,
ai s 2
(11)
409
where ai ai ai+ is the interval parameter associated with the jth joint having ai =
bi and bni are the elements of Bi and Bni , respectively, i.e.,
1/bni , ai+ = 1/bi . The parameters
0
0
and Bni =
.
Bi =
bi
bni
Since from equation (11) the plant is an interval plant, a parameter plane method based on the
gain-phase margin tester method and Kharitonovs theorem has been adopted in designing a set
of stabilizing PID controllers.
ei (t)dt,
(12)
where ei (t) = di (t) i (t) is the position tracking error of ith joint, K pi , K ii and K di are
respectively the proportional, integral and derivative gains of the ith joint controller, di denotes
the ith joint desired constant position and i be the actual joint position.
The control problem is to provide a complete solution to the constant gain stabilizing control parameters K pi , K ii , K di such that the position error ei (t) reduces to zero with time, i.e.,
lim ei (t) = 0.
t
Figure 2 shows the s-domain representation of the ith subsystem given in equation (10) with
h i = 0 (wi = Bi h i = 0). The open loop transfer function can be written as
G oi (s) = G i (s)Ci (s).
(13)
G i ( j)Ci ( j) = i e ji ,
(14)
For s = j, we have
wi = B i h i
di
ei
+
-
ith PID
controller
ui
Bn i
.
+ + xi
xi
Cs
+
As
410
E i(s)
di (s)
Ci (s)
ui
Gi ( s ) = 1 / ai s 2
i(s)
1 j(180+i )
e
G i ( j)Ci ( j) = 0,
i
(15)
1 + Ai e ji G i ( j)Ci ( j) = 0,
1
where Ai = 1i = |G i ( j)C
and i = 180 + i .
i ( j)|
It is noted that Ai is the gain margin of the ith subsystem when i = 0 and i is the phase
margin when Ai = 1. More specifically, one can easily determine the gain margin and phase
margin of the system by adopting the gain-phase margin tester Ai e ji , which can be represented
by an additional block in cascade with G i (s)Ci (s) and shown in figure 3.
The controller Ci (s) is designed to simultaneously stabilize the ith subsystem (9) with
h i (x) = 0. Let
(16)
(17)
This is a third order interval polynomial, and it is sufficient to check the stability of one
Kharitonov polynomial P + ( j) (Meressi et al 1993), which is given below.
a + s 3 + (K pi s + K ii + s 2 K di ) (Ai cos i j Ai sin i ) = 0.
(18)
The objective here is to find all possible sets of K pi , K ii and K di that make the characteristic
equation (18) to be stable. Equating the real and imaginary part of (18) to zero we get two
expressions as
3 a + K pi Ai cos (i ) + Ai sin (i ) (K ii 2 K di ) = 0,
Ei (s)
di
Ai e j i
Ci (s)
ui
Gi (s)
(19)
411
K pi Ai sin (i ) + Ai cos (i ) (K ii 2 K di ) = 0.
(20)
It can be noted that the number of controller parameters are more than the number of equations and it is necessary to assign one of the controller parameter (say K di ) and the remaining
controller parameters are solved from (19) and (20) with conditions of marginal stability
i.e., Ai = 1 and i = 0. To have the stable region in the parametric plane, one generally finds the stability boundary first, and then determines the stable region by the sign of
J j = Re/ K pi Im/ K ii Re/ K ii Im/ K pi , where Re and Im are real and imaginary
parts of equation (18). If the sign of J j is positive (negative) facing the direction in which is
increasing, the left (right) side of the stability boundary is the stable region (Siljak 1969). Thus,
one can obtain the range of controller parameters (K pi , K ii ) for a fixed value of K di from the
boundary of the stability region for the system.
The controller is designed based on the linear system, i.e., with h i = 0. A stability analysis for
the composite nonlinear system (10) based on Lyapunov method is investigated by transforming
quadratic terms into an equivalent LMI framework (Boyd et al 1994).
(21)
where wi = Bi h i , and it is required that nonlinear term wi satisfies the quadratic constraints.
wiT (t, x)wi (t, x) i2 x T WiT Wi x, i = 1, 2, . . . , n,
(22)
where i > 0 are interconnection parameters and Wi are constant matrices of appropriate
dimensions. The input to the system (21) with the PID controller is
t
u i (t) = K pi ei (t) + K di ei (t) + K ii
ei (t)dt.
(23)
(yi ) dt,
0
t
yi dt, so
xai = K ii yi = K ii Csi xi .
(24)
To study the stability of the interconnected system substitute for u i in (21) and augment it with
(24) to have the following form.
E ni xwi = Ani xwi + wni (t, xw ) ,
(25)
412
where
I + Bni K di Csi 0
Asi Bni K pi Csi Bni
, Ani =
,
0
1
K ii Csi
0
xi
wi (t, x)
.
, and xwi =
wni (t, xw ) =
0
xai
E ni =
(26)
where Anew = diag {An1 An2 , . . . , Ann } and E new = diag {E n1 , E n2 , . . . , E nn } are matrices
T
T , . . . , w T T is a function of
, wn2
of appropriate dimensions and the nonlinear term wn = wn1
nn
T
T , . . . , x T T . Since the set of stabilizing PID controllers is obtained, the matrices
xw = xw1 , xw2
wn
E new and Anew are, in turn, of interval form. In (26), the nonlinear function is constrained as
n
T
T
2 T
wn (t, xw )wn (t, xw ) xw
i Wai Wai xw ,
(27)
i=1
r2
1 T
1
1 T
T
P2 + P2T Arnew
P1 P2T E new
+ Arnew
P3 P2T Wa1
Arnew
r2 T
r2
P T Ar 1 + P E r 2 T P
E new
P3 P3T E new
P3T
0
1
2
3 new
new
P2
P3
I
0
Wa1
0
0 1 I
..
..
..
..
.
.
.
.
Wan
T
Wan
...
...
...
..
...
< 0,
0
..
.
n I
(28)
where r1 , r2 , . . . , 2k , i = 1/i2 , k is the size of the matrices Anew and E new , i = 1,2,. . . ,n.
Proof:
The constraint (27) is equivalent to the quadratic inequality
N
2
T
xw
W W 0
0.
xwT wnT (x) i=1 i ai ai
wn (x)
0
I
(29)
413
For the descriptor system (26), we introduce an augmented system (Lin et al 2005) to get the
T
following equation for the augmented vector z(t) = xwT (t)xwT (t) ,
I 0
0 0
xw
xw
=
0
Anew
I
E new
xw
xw
+
0
wn (x)
.
(30)
I 0
0 0
, A=
0
Anew
I
E new
,z =
xw
xw
and w (x) =
0
wn (x)
.
Let us choose a Lyapunov function candidate (Cao & Lin 2004) for the descriptor system (30) as
V = z T F Pz,
(31)
P1 0
is nonsingular with P1 = P1T > 0, and FP = (FP)T due to special
P2 P3
structures of F and P. We compute
where P =
T
V = z T A P + P T A z + w T (x) Pz + z T P T w (x) .
In order that the descriptor system (30) is stable, it is required that
P1 > 0,
T
z T A P + P T A z + w T (x) Pz + z T P T w (x) < 0.
(32)
(33)
+ + wnT (x) P2 xw + wnT (x) P3 xw + xwT P2T wn (x) + xwT P3T wn (x) < 0.
These inequalities can be rewritten as,
P1 > 0,
T
T P + PT A
T P + P PT E
Anew
Anew
2
3
1
2 new
2 new P2
xw
T P
T P PT E
xwT xwT wnT (x) P3T Anew + P1 E new
E new
P3T xw < 0.
2
3
3 new
wn (x)
P
P
0
2
(34)
414
N
T
T
T
T P + P PT E
i2 WaiT Wai Anew
3
1
2 new P2
Anew P2 + P2 Anew +
i=1
T P
T P PT E
P3T Anew + P1 E new
E new
P3T
2
3
3 new
P2
P3
inequalities (29)
< 0,
(35)
where P1 > 0 and a number > 0. By repeatedly applying the Schur complement formula
(Boyd et al 1994) to equation (35) with = 1, the above equation can be rewritten as
T WT
T P + PT A
T P + P PT E
T
Anew
Anew
2
3
1
2 new
2 new P2
a1 Wan
T
P Anew + P1 E T P2 E T P3 P T E new
P3T
0 ...
0
new
new
3
3
P
P
I
0
.
.
.
0
2
3
< 0, (36)
0
0 1 I . . .
0
Wa1
.
.
.
.
.
.
..
..
..
..
..
..
Wan
0
0
0 . . . N I
where i = 1/i2 .
The matrices Anew and E new of (26) are interval matrices (obtained from (25)). As discussed
by Mansour (1988), Jiang (1987) and Garofalo et al (1993), a sufficient condition for the stability
robustness of interval matrices, i.e., matrices having the elements varying within given bounds,
requires that the derivative of Lyapunov function be negative definite when evaluated at the
so-called
matrices. The corner matrices of an n n interval matrix A are defined as
corner
2
r
r
A = ai j , r = 1, 2, . . . , 2n with airj = ali j or aui j , i, j = 1, 2, . . ., n, where ali j and aui j are
minimum and maximum values, respectively of ijth element of interval matrix. Hence equation
(36) should be satisfied for all the corner matrices of Anew and E new for composite system (30)
to be asymptotically stable. The matrix Wai is assumed such that constraint (27) is satisfied
and the bounding parameter i is to be maximized. Hence (36) can be reformulated as an LMI
optimization problem as stated in (28). In other words, system (21) is robustly stabilizable by
the set of designed decoupled stabilizing PID controllers provided the LMI (28) has a feasible
solution for all corner matrices. This completes the proof.
= (a3 sin 2 ) 2 +221 2 ,
V (, )
(a3 sin 2 ) 1
415
m2
l2
l1
m1
G() =
a4 cos 1 + a5 cos (1 + 2 )
a5 cos (1 + 2 )
.
(37)
,
=
h(x) =
2 + 1.72x 2 + 3.44x x
h 2 (x)
5.12x12
12 22
22
x11
x21
= 1 , x2 =
= 2 .
(38)
and x1 =
x12
x22
1
2
The stabilizing set of gains for links 1 and 2 for system (38), obtained by solving equations (19),
(20) assuming values for K di from 1 to 100 and varying from 0.01 to 25 Hz are shown in
figure 5. The shaded region in figure 5 is the stabilizing controller parameter space of joints 1
and 2. The set of controller gains for joints 1 and 2 are taken as
K p1 10.1 500 , K i1 10.1 500 , K d1 10 100 ,
(39)
K p2 10.1 500 , K i2 10.1 500 , K d2 10 100 .
Knowing the ranges of controller gains (31) of the joints 1 and 2, genetic algorithm based
optimization technique (Goldberg 1989) is used to maximize the fitness function J f given by
1
,
(40)
Jf =
1+ J
n
t
ei2 dt and ei (t) = di (t) i (t).
where J =
0 i=1
416
The optimal controller parameters are obtained for fixed as well as time-varying desired
positions. The genetic operations used are arithmetic crossover, uniform mutation and ranking
selection. The population size of 50 is taken and GA is run for 25 generations.
= 150.32, K d1
= 86.84,
K p1 = 176.83, K i1
(41)
K p1 e1 (t) +
K i1
t
e1 (t)dt + K d1
e1 (t)
t
u 2 (t) = K p2 e2 (t) + K i2
t
(42)
e2 (t)dt + K d2
e2 (t)
t
Figures 6 and 7 show the desired position and actual position of joints 1 and 2 with the designed
control laws (42). The position errors of each joint are also plotted in figures 8 and 9, respectively.
417
= 43.29, K d1
= 56.31,
K p1 = 129.11, K i1
= 20.4, K d2
= 17.04.
K p2 = 77.99, K i2
(43)
418
The control laws with optimal controller parameters for both joints are obtained as
u 1(t) = K p1 e1 (t) +
K i1
t
e1 (t)dt +
0
u 2(t) = K p2 e2 (t) + K i2
t
0
K d1
e1 (t)
t
= 129.11e1 (t) + 43.29 e1 (t)dt + 56.31e1 (t),
t
e2 (t)dt + K d2
e2 (t) = 77.99e2 (t) + 20.4 e2 (t)dt + 17.04e2 (t).
0
(44)
Figures 10 and 11 show the desired and actual positions of joints 1 and 2 with the designed
optimal control laws (44). The position errors of each joint are plotted in figures 12 and 13.
419
Figures 613 reveal the effectiveness of the proposed decentralized PID control scheme and
further it ensures tracking errors converge to zero asymptotically.
4.3 Stability analysis of two-link robot manipulator
The stability analysis of the two-link robot manipulator (38) with the designed set of controllers
(39) was studied by solving the LMI optimization problem (28) for all the corner matrices of
Anew and E new . The designed ranges of Anew and E new are calculated using equations (25)(26)
with the controller gains (39) and are given by
Anew = diag {An1 , An2 } ,
E new = diag {E n1 , E n2 } ,
(45)
420
where
0
1
1 0
An1 = 2900 58.6 0 5.8 ,
E n1 = 58 580
0
500 10.1
0 0
0
1
1 0
An2 = 2900 58.6 0 5.8 and E n2 = 58 580
0
500 10.1
0 0
0 0
1 0 ,
0 1
0 0
1 0 .
0 1
421
As given in equation (27), the term wn (t, xw ) can be bounded by a quadratic inequality and is
constrained as
2
T
2
2
wn1
(t, xw )wn1 (t, xw ) = 3.36x12 x22 + 1.72x12
+ 1.68x22
2
2
T
xwT 12 Wa1
17.07x12 x22 8.54 x12
+ x22
Wa1 xw ,
2
T
2
2
wn2
(t, xw )wn2 (t, xw ) = 3.44x12 x22 5.12x12
1.72x22
2
2
T
xwT 22 Wa2
29.44x12 x22 14.72 x12
+ x22
Wa2 xw (46)
(since x12 and x22 are much less than unity and the terms associated with the power of x12 and
x22 equal to three or more than three are neglected), where 1 , 2 > 0 and Wa1 and Wa2 are
found out as
0 0 0 0 0 0
0 0 0 0 0 0
0 2.92 0 0 0 0
0 3.87 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
Wa1 =
(47)
, Wa2 = 0 0 0 0 0 0 .
0 0 0 0 0 0
0 0 0 0 2.92 0
0 0 0 0 3.87 0
0 0 0 0 0 0
0 0 0 0 0 0
Two elements each of An1 , An2 and E new are of interval form, i.e., four corner matrices for
each of An1 , An2 and E new are obtained. Table 1a shows the corner matrices of An1 , An2 and
E new . Thus sixteen corner matrices are possible for Anew with the eight corner matrices of An1 ,
An2 . These corner matrices of Anew are given in table 1b. This sixteen combinations of Anew
and four corner matrices of E new are considered, thus there are sixty-four combinations for
which optimization problem (28) is solved using LMI control toolbox (Gahinet et al 1995) with
Wa1 , Wa2 taken as equation (47). As the number of links increases the number of LMIs to be
solved increases exponentially (23n where n is the number of links of the manipulator). Table
1c shows the values of 1 , 2 obtained by solving the LMI problem (28) with the designed
range of controller parameters given by (39). It is seen that a feasible solution exists for all the
corner matrices. Hence, it is concluded that the set of decentralized PID controllers based on
Table 1a. Corner matrices of An1 ,
A1n1 = An1 (2, 1), An1 (3, 1)
A2n1 = An1 (2, 1), An1 (3, 1)
A3n1 = An1 (2, 1), An1 (3, 1)
A4n1 = An1 (2, 1), An1 (3, 1)
An2 (3, 1)
An2 (3, 1)
An2 (3, 1)
An2 (3, 1)
1
E new
= E n1 (2, 1),
2
= E n1 (2, 1),
E new
3
E new
= E n1 (2, 1),
4
= E n1 (2, 1),
E new
E n2 (2, 1)
E n2 (2, 1)
E n2 (2, 1)
E n2 (2, 1)
where Ani (2, 1) and Ani (2, 1) denote the lower and upper limits of (2, 1)th element of matrix Ani .
A9new = A1n1 A3n2
3
2
A10
new = An1 An2
3 3
A11
new = An1 An2
4 3
A12
new = An1 An2
1 4
A13
new = An1 An2
2 4
A14
new = An1 An2
3 4
A15
new = An1 An2
4 4
A16
new = An1 An2
422
A1new
A2new
A3new
A4new
A5new
A6new
A7new
A8new
A9new
A10
new
A11
new
A12
new
A13
new
A14
new
A15
new
A16
new
2
E new
3
E new
4
E new
0.1585
0.1584
0.1566
0.0323
0.1584
0.2184
0.1566
0.0323
0.1566
0.1566
0.1565
0.0323
0.0323
0.0323
0.0323
0.0323
0.1374
0.1376
0.1360
0.0281
0.1376
0.1896
0.1360
0.0281
0.1360
0.1360
0.1361
0.0281
0.0281
0.0281
0.0281
0.0280
0.1584
0.1583
0.1584
0.1584
0.2184
0.2186
0.2185
0.2043
0.1566
0.1565
0.1567
0.1566
0.0323
0.0323
0.0323
0.0323
0.1376
0.1376
0.1376
0.1376
0.1898
0.1898
0.1898
0.1773
0.1361
0.1360
0.1361
0.1361
0.0281
0.0281
0.0281
0.0281
0.1584
0.2184
0.1566
0.0323
0.1583
0.2186
0.1566
0.0323
0.1584
0.2185
0.1567
0.0323
0.1584
0.2043
0.1566
0.0323
0.1376
0.1897
0.1361
0.0281
0.1376
0.1898
0.1360
0.0281
0.1376
0.1898
0.1361
0.0281
0.1376
0.1772
0.1361
0.0281
0.2226
0.2227
0.2227
0.2044
0.2227
0.2229
0.2227
0.2042
0.2227
0.2227
0.2226
0.2042
0.2044
0.2042
0.2042
0.2041
0.1934
0.1935
0.1935
0.1772
0.1935
0.1936
0.1935
0.1774
0.1935
0.1935
0.1934
0.1774
0.1772
0.1774
0.1774
0.1772
Kharitonovs theorem and stability boundary equation stabilizes the two-link manipulator system
(38) with the numerical values of local controller parameters (39). The finite numerical values
of i , i = 1,2 indicate that the decentralized robust stability analysis of interconnected nonlinear
system with its maximum nonlinear perturbations.
5. Conclusion
A class of stabilizing decentralized PID controllers was designed for each link of a two-link
manipulator using parameter plane method and Kharitonovs theorem. A Kharitonov region was
obtained graphically such that a PID controller with coefficients selected from this region stabilizes the whole uncertain nonlinear system. Even though the design of PID controllers was
done based on linear system, the linear controller stabilizes the nonlinear system, which is
proved by solving LMI optimization problem and thereby obtaining the bounding parameter of
the interconnection terms. From the simulation results shown, it is concluded that the proposed
controllers closely tracks the constant as well as time-varying desired positions.
Acknowledgements
The authors would like to thank the anonymous referees for their helpful comments and
suggestions.
References
Boyd S, Feron E, Ghaoui L E and Balakrishnan V 1994 Linear matrix inequalities in system and control
theory (Philadelphia: Siam)
423
Cao Y Y and Lin Z 2004 A descriptor system approach to robust stability analysis and controller synthesis.
IEEE Trans. Automatic Control 49(11): 20812084
Gahinet P, Nemirovski A, Laub A J and Chilali M 1995 LMI Control toolbox for use with Matlab (Natik,
MA: The Math works Inc.)
Garofalo F, Celentano G and Glielmo L 1993 Stability robustness of interval matrices via Lyapunov
quadratic forms. IEEE Trans. Automatic Control 38(2): 281284
Goldberg D E 1989 Genetic algorithms in search, optimization, and machine learning (Boston, MA, USA:
Addison Wesley)
Huang Y J and Wang Y J 2000 Robust PID tuning strategy for uncertain plants based on the Kharitonovs
Theorem. ISA Trans. 39(4): 419431
Jiang C L 1987 Sufficient condition for the asymptotic stability of interval matrices. Int. J. Control 46(5):
18031810
Lii G H, Chang C H and Han K W 1993 Analysis of Robust Control Systems using stability equations. J.
Control Syst. Technol. 1: 8389
Lin F and Brandt R D 1998 An optimal control approach to robust control of robot manipulators. IEEE
Trans. Robotics Automation 14(1): 6977
Lin C, Wang Q G and Lee T H 2005 Robust normalization and stabilization of uncertain descriptor systems
with norm-bounded perturbations. IEEE Trans. Automatic Control 50(4): 515520
Liu M 1999 Decentralized control of robot manipulators: nonlinear and adaptive approaches. IEEE Trans.
Automatic Control 44(2): 357363
Mansour M 1988 Sufficient condition for the asymptotic stability of interval matrices. Int. J. Control 47(6):
19731974
Meressi T, Chen D and Paden B 1993 Application of Kharitonovs Theorem to Mechanical Systems. IEEE
Trans. Automatic Control 38(3): 488491
Narendra K S and Oleng N O 2002 Exact output tracking in decentralized adaptive control systems. IEEE
Trans. Automatic Control 47(2): 390395
Qu Z and Dawson D M 1996 Robust tracking control of robot manipulators (New York: IEEE press)
Seraji H 1989 Decentralized adaptive control of manipulators: Theory, simulation, and experimentation.
IEEE Trans. Robotics Automation 5(2): 183201
Siljak D D 1969 Nonlinear systems: The parameter analysis and design (N.Y: John Wiley & Sons)
Siljak D D and Stipanovic D M 2000 Robust stabilization of nonlinear systems: The LMI approach. Math.
Prob. Eng. 6: 461493
Tang Y and Guerrero G 1998 Decentralized robust control of robot manipulators. Proc. of 1998 American
Control Conference, vol. 2, pp 922926
Tang Y, Tomizuka M, Guerrero G and Montemayor G 2000 Decentralized robust control of mechanical
systems. IEEE Trans. Automatic Control 45(4): 771776
Tarokh M 1996 Decentralized adaptive tracking control of robot manipulators. J. Robotic Syst. 13(12):
803816
Wang J-Q and Wend H D 1999 Robust decentralized control of robot manipulators. Int. J. Syst. Sci. 30(3):
323330
Yakubovich V A 1977 The S-procedure in nonlinear control theory, English translation in Vestnik Leningrad
Univ. Math. 4: 7393