Proceedings o f the 2003 IEEE loterostional Conference on Robotics &Automation
Taipei, Taiwan, September 14-19, 2003
Globally Adaptive Decentralized Control of Time-Varying
Robot Manipulators
Su-Hau
Hsu' and Li-Chen
Fu','
'Dept. ofElectrical
Engineering'Dept. of
Computer
Science 8
Information Engineering National Taiwan University, Taipei, Taiwan.TEL: 886-2-23622209;
FAX:886-2-23657887
E-mail: lichen@ccms.ntu.edu.tw
Abstract
-
In this paper, we develop a globally adaptive de- centralized control scheme of time-varying robot manipula- tors for trajectory tracking control. Since the proposed adap- tive control law is in a decentralized manner, only low-cost hardware is required for implementation. Furthermore, even if the time-varying parameters of the robot manipulator change arbitrarily fast, both the position and velocity tracking errors of the manipulators will converge to zero after em- ployment of the proposed adaptive control law. For practical implementation, the adaptive law can be combined with a leakage term so that there is no chattering in the motion of the manipulator but at the price a residual set of the tracking errors as mentioned above, of which the size can be however made smaller by use of some larger design parameters. Finally, in order to illustrate the performance of the proposed scheme, simulation r e s u l t s are also provided, which turn out to he quite satisfactory.1. INTRODUCTION
Research in the area of trajectory tracking control of robot manipulators has been widespread. Many industrial tasks of the robot manipulator such as material handling and pan assembly involve such a problem. On the other hand, in many situations, some of the unknown parameters such as the mass of the payload or the mass of the links, may be time varying. For example, robotic pouring and tilling operations belong to this case. A large number of control design exist in the literature that works well for both cases with the both known and unknown constant pa- rameters [Slotine and Li, 7; Spong, 91. However, it is worth noting that, because the time-varying parameters of robot manipulators enhance complexity of the overall systems, the control schemes developed for the time-invariant robot manipulators are not suitable for the time-varying ones [Reed and Ioannou, 61.
During recent years, quite a few research works have
been devoted to the trajectory tracking control of time-varying robot manipulators. In Tao [12], a robust adaptive controller is proposed to guarantee asymptotic tracking in the presence of time-invariant parameters. However, the transient performance is related to the speed of parameter variations and cannot he arbitrarily improved. In Tomei [14], a robust adaptive control law is proposed which ensures the arbitraxy transient performance. In Song and Middleton
[PI,
a robust switching-type control law is proposed for the robot manipulators with time- varying parameters. In this research, the property of Hadarmad product is applied such that the dynamic model can be ex- pressed in a parameterized form and the zero asymptotical tracking error can he obtained. In Su [ll], an adaptive slid- ing mode control is proposed which guarantees asymptotic convergence of the tracking error. In the work by Pagilla et al. [ 5 ] , with hounded time-varying parameters an adaptive control law is designed, of which the effectiveness is dem- onstrated by providing experimental results. In the litera- ture, different control laws have been proposed for the tracking control of time-varying robot manipulators; how- ever, most of them are in centralized manner. Since the decentralized controller structure is adopted by the majority of modern robots in favor of its computation simplicity and low-cost hardware setup [Fu, 2; Liu, 3; Tang et. al., 131,how to best improve the tracking performance of time-varying robot manipulators through decentralized control becomes an interesting research topic.
In this study, a globally adaptive decentralized tra- jectory control law is proposed for time-varying robotic
manipulators.
11. PROBLEM STATEMENT
In this section. the dynamic model of a time-varying robot manipulator is introduced and the difference between
the time-invariant ease and the time-varying one is also described. The present dynamic model is derived under on the assumption that the generalized constraints are inde- pendent of time but the payload mass can he time varying. It is also under on the assumption that the time-varying payload mass has no contribution to producing the general- ized force. The dynamical model of robot manipulators is derived by Enler-Lagrange equations [Spong and Vidyasa- gar, IO]. Let q = (ql,
... ,
q,JT be the set of position vari- ables of n joints of the robot manipulator. The scalar func-tions
K
: R" x R" x [O, -) R andP
: R" x [O, -)+
R denote the kinetic energy and the potential energy, respec- tively, where apparently the kinetic energy K : R" x R" x [0,-)
+
R can he explicitly expressed as defined as followsK
= i q T M ( q , f ) q (1)where M : R" x [0, -)
+
R"" is the inertial matrix of the robot manipulator. It is worth noting that the inertial matrix M and the potential energy P are explicitly dependent on f such that the robot manipulator is called the time-varying one. The Lagrangian L is obtained by taking the differencebetween the kinetic energy K and potential energy P , i.e., L
=
K
- P . After substituting the Lagrangian L into theEuler-Lagrange equations, the overall dynamics of the ro- bot manipulator can he obtained as follows [Tao, 121:
M ( q , t ) i ; + ~ 4 + C ( q , q , f ) q + g ( q , f ) = z (2)
where the (k, j ) element, 1 S k, j S n , of C : R" X R" X [0, -)
+
RnXn 1s:c -
-Ai(!%+&-?
aq; ab ( 3 )i-l
with mu being the ( i . j ) element of
M,
and g(q, I) = W ( q ,r)/aq. This dynamic model has the following properties that
will be used in the controller design [Su, 1 I]:
Property I:The inertial matrix M i s a symmetric and positive
definite matrix,
which satisfiesp , l <
M <
bMI,
VqE R", V I E [0, -), for some constants ~r,, p M > 0.
Property 2: The matrix C satisfies IlCllS pcllqli, Vq
E Rn, V
4
E R", V f E [0, -), for some constant pc > 0Property3: z T ( n ; ( - % - 2 C ) z = 0 , V z ~ R " , V ~ E CO,
-)
It is worth noting that, if the inertial matrix M is not explicitly independent of 1, then Property 3 can be rewritten as:
zT(fi
-2c)z = 0 , VZ E R", which is well-known in the literature ofrobot manipulators.Property 4: The vector g satisfies llgll< pG, Vqe R",
Vf E [O, -), for some constant pG > 0
Compared with the time-invariant robot manipulators, be- sides that the parameters in the dynamic model of the time-variant robot manipulators are Considered as time functions, additional terms related to the rate of change of time-varying parameters are also derived such that Property
3 of time-varying robot manipulators is different from that
of time-invariant robot manipulators.
Now let q d f ) denote the desired position trajectory for tracking, which is generally chosen twice differentiable to guarantee smoothness of the motion. Let TT = [T,,
.. .
, TJ.Iftheithelement,iE{l,
._.
,n},ofthecontrolinputisonlya function ofjoint configuration and velocity of the ith joint, the control input 7 is called a decentralized control. In this study, a decentralized control scheme is to be developed such that q(t)
+
q&) andq(t)-+
q r ( f ) as f+
-.
111. CONTROLLER DESIGN
In this section, one adaptive decentralized control schemes will he developed for the control of a general d i n k rigid time-varying manipulator. Let qdf) denote the desired position trajectory. The following error signals are defined as e = q - qd and s
=
d + Ae where the feedback gain matrix A is a positive definite matrix. Now the dy- namics defined hy the signals e and s can be derived as:d = -Ae+ s (4a) (4b) M(q,t)S = -C(q,
4,
f)s+.r-Nf, 4.4)W,
q,4)
= M ( q , f)(& + A 4 + C(q,4.
f ) ( Q d + where ( 5 ) aM(1,'l ' + g ( 9 ) - 7 4behaves as the perturbation term In the following, without
loss of
generality, several technical
assumptionsare
made to pose the problem in a tractable manner.Assumption I : The feedback gain matrix A is a di- agonal positive definite matrix; that is, A = diag
(XI, ...,
1,)
> O f o r h , > O , i e { l ,...
,n}.Assumpfion 2: The desired position trajectory 461) and its time derivatives q d ( f ) , q d ( t ) are all bounded time-varying signals.
Assumpfion 3: The time-varying parameters of robot
manipulators are hounded and smooth functions in time. Furthermore, there is a positive constant q
>
0 such thatIn this study, we start with investigating the following lemma. In this sequel, we adopts the truncated L., norm
defined as
Ibllr,.,
E maxl,,sup,e~o, &(t)l for all real vec- tor-valued functions x E c"[0, for some T > 0. To assess the tendency of growth, the following lemma will provideus useful facts to develop the control scheme in this study.
Lemma 1: Under Assumption 3, ifthere exists a constant T
> 0 such that 11s exists, then there arepositive constants
a,,
az,
a3,
and% such that f o r t E [0,TJ
:Ile(t)ll ~al//eoll+a2ll~ll,,~ (6)
Il@)lls
a3lleoll+ a 4 1 1 ~ l l T , ~ (7)where eo = e(0) is the initial condition of e, and positive constants
PI,
8,
and p 3 are such thatfor t E [0, r ] :Ilv(t,
4.4)11s
PI+ B ~ / / ~ / l ~ , ~
+ D 3 l l ~ l l : , ~v ( t , q, q ) = v ( t , q , q ) - + + s
(8) where V ( t , q , q ) E [O, -) x R" x R"
+
R" is defined asfol-lows:
(9)
-
Proof: Consider the linear system defined by
e = - h e + s , e(0) = e , . Since the matrix -A is Hunvitz, we
obtain (6) for t E [O,
rl
whereai
andaz
are some positive parameters. Furthermore, according to the definition of the linear systems, we then obtain (7) for t E [O,TJ
where a, and a4 are appropriate positive parameters. Next, according tu the definition of the perturbation term v in ( 5 ) and As-sumption 3, we have (8) for I E 10, U , where
PI,
p2,
andp,
Now, an adaptive decentralized control scheme is presented to solve this tracking control problem. Consider the control lawrT
= [T,,..
., r,]
given as follows:-ti;-+,
- e , ~ , s i - 8 , ~ , s ~ if~ ( - , I s ~ I ~ E ~
1
sgn(sj) -e,_,sj-ei-&
if i,_,Js,l> E~fur i = 1,
. . . ,
n, where the parameter €Ij,
, i E (1,._.
, n }needs to he adjusted on line, and Of-, > 0, i E { 1 ,
. . . ,
n } and j ~ { 2 , 3}, and sgn(-) denotes the signum function [Khalil,41. Furthermore, the auxiliary signal Eir i E { I ,
._.
, n } , sat- isfies:are some chosen positive parameters.
(10)
Ti =
E < = - P E . I ,, E , @ ) > 0 (11)
where p i > 0. It is worth noting that the time-varying signal
E,, i E { 1,
.
. .
, n } . are always positive, which are used as the time-varying boundary layers. In order to account for the parametric uncertainty from the manipulator and the de- sired position trajectories, we choose the adaptive law asfollows
k ,
=Y,&,I
(12)where b j - , ( 0 ) L O , and the adaptive gain " f j - ] > 0, i ~ { l ,
.., , n}. With (lO)-(l2), the closed-loop system becomes differential equations with discontinuous right-hand sides [Filippov, I]. A considerable amount of works has been camed out to deal with situations like this. Here, we base our results for these differential equations on Filippov's concept. Note that the adaptive control law (10)-(12) is
apparently in a decentralized manner, and its performance can be summarized into the following theorem.
Theorem 2 (Adaptive Decentralized Control Scheme): Under Assumption 1-3, consider the error dynamics of the robotic manipulator with the adaptive control law (10)-(12) defined above. Then, a// signal are bounded, and, both, the
position tracking error e(t) and the velocity tracking error
$ 1 ) will converge to zero as t
+
-.
Proof: The proof proceeds in the following two steps. Step 1: Prove the signal s(t) is ultimately bounded. Con- sider the Lyapunov-like function as follows:
V ( t ) = + T M S (13)
Consider IIsIIT,,- < I, for some finite T > 0. Then, after taking the time derivative of (13) along s ( t ) for t E [O,
T J ,
we oh- tain the following result:d=s'r-sV~sT.r+II~II(PI
+ P J ,
+P,l,') (14) where the functionV
is defined as that in (9). Then, the following two different cases will he obtainedCase1:
i 3 j + l < ~ ,
f o r t s[ n , q ,
. ,
+11sIl,(P,
+ P A
+ P J 3
where
ej,
,,,m = min(81 j ,...,
0._,}, jE R
3). From bothCase 1 and 2, we can conclude that for I E [0, 7J:
v s
-~2.m,"l14: (17)when Ilsllz t [""'*~~,":"""]"', which implies that there ex- ists lZ > 0 such that for I E [O,
T J ,
lIs(t))Iz
5 /, < /, if lI is suf- ficiently large. Similar to the previous argument, 1, can hedefined as l2
=E[
""'+~~,~,'""'1
'"
where II is suffi- ciently large, and hence one can show that s(t) is actually ultimately bounded so that there exists l, > 0 such that 1, =Step 2: Prove all signals are bounded and the signals
+
0 as t+
m. Now consider the Lyapunov-like function as fol-lows:
supz.ro,-,
/I
s(t) 112.V(t) = jsTMs+
x[+y::,(e,-l
-e:-$
+ p ; ' E i ] (18)i=,
where 8: I , i E { 1,
. . . ,
n } is the desirable hut unknown parameter. After taking the time derivative of (18) along the solution trajectories of the closed-loop system, the follow- ing two different cases will he obtainedCase I: Bi-IJsi)
<
ci
for t E IO,-1,
where
2, we conclude that for sufficiently large
= m i n i 8; I ,
...
,e.-,).
From both Case 1 and we obtain:c's
-e*,mi.llsll:
(21)for t E [0, -), which implies all signals are bounded and
Ils(r)ll, is L2. Finally, to show the zero convergence of the
tracking error e(t) and e ( t ) , we need to show Ils(t)lIz is uniformly continuous and then apply Barbilat's Lemma [4].
Sufficiently, we investigate houndedness of the signal i ( t ) from (4),
(S),
and (IO), and easily we can verify such a condition. Hence, the zero convergence is thus insured.W
R e m a r k In Theorem 2, we would like to emphasize that here in OUI proposed scheme, the control gain I&-,, i E { I,. . . ,
n}
and j E { 1, 2}, only needs to be chosen positive, and the adaptive decentralized control law (10)-(12) will then successfully drive the tracking errors e(<) and e ( ( ) of the closed-loop system to converge to zero globally. Hence, this control law can he claimed to be a global one.Due to the numerical noise caused by convergence of
&i, i E { I,
... ,
n}. to zero, practical implementation of thescheme should he considered, e.g., using addition of the leakage term [ 2 ] . Now, consider another modified adaptive law as follows:
e i . , = Y j ~ , ( l ~ ! l - I S ~ s . J (22)
where 6,-,(0) t 0 and the adaptive gains y(-, > 0, i E { I ,
. . . ,
n } , and the leakage-term constant IS > 0. Note that the adaptive law (22) is apparently in a decentralized structure, too. The performance of adaptive control law (IO), (1 I ) and (22) can he summarized into the following corollary. Corollary 3 (Adaptive Decentralized Control Scheme with the leakage term): Under Assumption 1-3, considerthe error dynamics of the robotic manipulator with the adaptive control law (IO), (11) and (22). Then, a / / signal are bounded, and, theposition tracking error e(t) will con-
verge to a residue set whose sire can be reduced by use of
larger O Z . ~ ~ , and y1,en, where = minlOz.I,
... ,
OZ-,,},andy,,,i.=min{yl.l,
... ,yn.ll.
Proof: This proof is similar to that in the proof of Theorem
2, and, hence, is neglected here.
W
IV. SIMULATION
RESULTS
In order to demonstrate the performance of the pro- posed adaptive decentralized controller, several numerical results are provided now. A two-link planar robot manipu- late with two revolute joints is considered here [Spong and Vidyasagar, IO; Su, I I]. Let, for i = 1, 2, d, denote the mass of link i, 1, denote the length of link i , lCi denote the distance from the previous joint to the center of mass of link i, and I, denotes the moment of inertia of Link i about an axis com- ing out of the page, passing center of mass of Link i. The inertial matrix Min ( 1 ) has four elements
m,, =dllzl + d 2 ( l : + / ~ ~ + 2 l , / , , c o s y , ) + I ,
+I,
m,, = m , , = d , ( / ~ 2 + l ~ c o s q 2 ) + 1 2 (23)mZ2 = d21:2
+I,
and the potential energy P i s as follows:
where go = 9.8 is the gravitational magnitude. The mat& C in (2) has four elements as follows:
P(q, t ) =
(4L
+ m,l,)g, sin y1 + m,l,,g, s i n k , + q 2 ) (24)(25) where h = -dz
h
lc2 sin y2. If we consier on an unknown time-varying load camed by the robot as part of the second link, then the parameters m2, l,,, and 1, will explicitly he time-dependent. In this case, the time rate of the change of the inertia matrix MbecomesCII = 42h, ci2 =
(4,
+4,)h,where 0, = d2 l:z, 02 = d2 I f , = I2, 04 = d2 l2
IC>. which
satisfies Assumption 3. We assume that the two physical links are thin cylinders with uniform density and identical length. The parameters used in this simulation study are listed as follows: d , = 10, /I = 1, /,I = 0.5, II = 516, 4 = 5 tAd2(t), 12= 1, lcz=0.5 + A h ( / ) , 12= 5d21:/12 where Ad2(()
= 0.5cos(0.21), Al,(t) = 0.2cos(0.2t). For practical consid- eration, the adaptive control scheme with the leakage term is employed. The desired trajectory q d ( t ) = [qdl(t), qm(t)lT is defined by qdl(t) = sin t, qm(t) = sin 1. The initial condi- tions used in the numerical study are ql(0) = --P, q2 (0) = 0,
k,(o)=
m)=
0,m=
q2w=
0, 0,~,(0)=0,.,(0)= 0, ~ ~ ( 0 ) = E2(0) = 1. The control gains, adaptive gain constants, and auxiliary signals are, respectively, defined as hi = h2 =6, 01.2 = 02-2 = 50, 01-3 = 02-3 = I , ? ] = ~2 = 50, o = 0.5, and
pI = p 2 = 0.5. After compute simulations, Figures 1 and 2 illustrate the numerical results of tracking for Links 1 and 2, respectively. From Figures (la), (2a), (Ib) and (2b), it can be concluded is found that the position and velocity track- ing errors approach to a residual set as the time approaches infinity. Figures (IC) and (2c) demonstrate that the updated parameters due to the adaptive law are kept bounded with time. Finally, Figures (Id) and (2d) illustrate that the two control inputs pertaining to this adaptive control scheme also remain bounded.
V. CONCLUSION
In this paper, we develop a globally adaptive decen- tralized control scheme for the time-varying robot manipu- lator performing tasks of trajectoly tracking. Because the proposed adaptive control law is in a decentralized manner, only low-cost hardware is required for implement. Even if the time-varying parameters of robot manipulators change fast, both the position and velocity tracking error of robot manipulators will also converge to zero by use of the pro- posed adaptive control law. For practical implementation mentioned, the adaptive law can be combined with a leak- age term such that there is no chatting in the control to ma- nipulators and the tracking errors of joint positions and velocities of robot manipulator will converge to residual set whose size can be made smaller by use of larger design parameters. Finally, in order to illustrate the performance of the proposed scheme, satisfactory simulation results are
also provided. All the simulation results are satisfying.
REFERENCES
A. F. Filippov, “Differential equation with discon-
tinuous tight hand side,” American Mathematical so- ciety Translations, vol. 42, no. 2. pp. 199-231, 1964. L.-C. Fu, “Robust adaptive decentralized control of robot manipulators,” IEEE Trans. Automat. Contr ,
M. Liu, “Decentralized control of robot manipulator: nonlinear and adaptive approaches,” IEEE Trans. Automat. Conk, Vol. 44, No. 2, pp.357-363, 1999.
H. K. Khalil, Nonlinear systems, Second edition, Prentice-Hall, New Jersey, 1996.
P. R. Pagilla, B. Yu, and K. L. Pau, “Adaptive control of time-varying mechanical systems: analysis and experiments,” IEEE/ASME Trans. Mechatron., Vol. 5 ,
J. S . Reed, and P. A. Ioannou, “Instability analysis and robust adaptive control of robot manipulators,”
IEEE Trans. Automat. Contr, Vol. 5 , No. 3, pp.
381-386, 1989.
J. I. Slotine and W. Li, “on the adaptive control of robot manipulators,” Int. J. Robotics. Res., Vol. 6, pp.49-59, 1987.
Y D. Song, and R. H. Middletion, “Dealing with the time-varing parameter problem,” IEEE Trans. Auto- mat. Confr,Vol. 37,No. I0,pp. 1597-1601.
M. W. Spong, “On the robust control of robot ma- nipulators,” IEEE Trans. Automat. Contr, Vol. 37, No.
11, pp. 1782-1786, 1992. vol. 3 7 , ~ ~ . i,pp.io6-110,1992.
NO. 4, pp. 41n-418,2000.
[ I O ] M. W. Spong, and M. Vidyasagar, Robot Dynamics and Control, John Wiley & Sons, New York, 1989. [ I l l C.-Y. Su, “Adaptive sliding model control of nonlin-
ear robotic systems with time-varying parameters,”
Syst. ConfrolLett., Vol. 23, pp. 35-41, 1994.
[ 121 G. Tau, “On robust adaptive control of robot manipu- lator,” Automatica, Vol. 28, No. 4, pp. 803-807, 1992. [13]
Y.
Tang,M.
Tomizuka, G Guerrero, andG
Monte-mayor, “Decentralized robust control of mechanical systems,” IEEE Trans. Automat. ContK, Vol. 45, No. 4, pp. 771-776, 1992.
[I41 P. Tomei, ”Robust Adaptive Control of Robots with Arbitrary Transient Performance and Disturbance Attenuation,” IEEE Trans. Automat. Contr., Vol. 44,
mm.,*./,
(la) Position tracking error e,
nnr<.-i
(lb) Velocity tracking error e,
I r n ,
, , , , ,
, , , ,
,
m
Y
e4
(IC) Estimated parameter
6,
I r n .Irn .m I) I
.
6 8 10 12 I. 16 I8 m n m / a r / kc(Id) Control input T~
Numerical results of Link 1 Fig. 1
d
T r n , * 4
(2a) Position tracking error e ,
nrner,
(2b) Velocity tracking error e, Irn rn M Y m .a 4 0 2 I 6 8 I Y 1) I. 16 >I( = mrnta=, ~Irn (2c) Estimated parameters
6,
,
(2d) Control input 5,Numerical results of Link 2 Fig. 2