1987 Harashima - Tracking Control of Robot Manipulators Using Sliding Mode.pdf
1987 Harashima - Tracking Control of Robot Manipulators Using Sliding Mode.pdf
1987 Harashima - Tracking Control of Robot Manipulators Using Sliding Mode.pdf
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. PE-2, NO. 2, APRIL <strong>1987</strong> 169<br />
<strong>Tracking</strong> <strong>Control</strong> <strong>of</strong> <strong>Robot</strong> <strong>Manipulators</strong> <strong>Using</strong><br />
<strong>Sliding</strong> <strong>Mode</strong><br />
FUMIO HARASHIMA, SENIOR MEMBER, IEEE,<br />
JIAN-XIN XU, AND HIDEKI HASHIMOTO, MEMBER, IEEE<br />
Abstract-A new methodology using the theory <strong>of</strong> variable structure computing inaccuracies will be caused if the system consystems<br />
is developed for accurate tracking control <strong>of</strong> robotic manipu- tains unknown parameter variations. An adaptive method<br />
lators. The proposed method, based on a sliding mode controller, pro- tajectory contro maniators in taskoreted<br />
duces a discontinuity control input so that the nonlinear interactions<br />
and influence caused by unknown parameter variations can be de- ordinates is also proposed by [6]. This method requires<br />
pressed whereas the resulting control law is simple and easy to apply neither knowledge <strong>of</strong> the parameters <strong>of</strong> the system nor<br />
to on-line computer control. The digital simulation results <strong>of</strong> a three- complicated calculations. The shortcoming <strong>of</strong> the method<br />
degree-<strong>of</strong>-freedom manipulator show the validity <strong>of</strong> accurate tracking is that the whole design procedure is implemented under<br />
capability and robust performance <strong>of</strong> the system with the developed the assumption that the velocity <strong>of</strong> the desired trajectory<br />
control scheme.<br />
should be sufficiently small.<br />
To realize on-line manipulator control with accurate<br />
INTRODUCTION tracking capability while in high-speed motion, the de-<br />
IN THE ADVANCED control <strong>of</strong> robotic manipulators, signers should consider the following two points which<br />
it is important for manipulators to track arbitrary trajec-<br />
are the key problems in choosing a control law:<br />
tories in a wide range <strong>of</strong> work space. However, if high 1) avoidance <strong>of</strong> complex computations such as those<br />
speed and accuracy in tracking motion are required, the for the nonlinear compensation;<br />
control <strong>of</strong> manipulators is difficult to realize using con- 2) insensitivity <strong>of</strong> the manipulator control to the unventional<br />
methods such as proportional integral derivation known parameter variations or modeling ambigui-<br />
(PID). This is because a multijoint manipulator is a highly<br />
ties.<br />
nonlinear system that includes nonlinear interactions between<br />
each arm. Moreover, the interactions vary over a To overcome these difficulties, the authors developed a<br />
wide range corresponding to the positions and motions <strong>of</strong> sliding mode controller based on the theory <strong>of</strong> variable<br />
the manipulator.<br />
structure systems (VSS). The main feature <strong>of</strong> sliding mode<br />
In general, the dynamics <strong>of</strong> each joint can be described control [7] is to let a sliding mode occur on a prescribed<br />
by a coupled second-order differential equation with Cor- switching surface. While in sliding mode the system is<br />
iolis and centrifugal forces represented by coupling terms. only govemed by the sliding equation; thus the system<br />
As to such complicated dynamics, quite a number <strong>of</strong> pa- will be forced to "slide" along the desired phase trajecpers<br />
have been presented on the subject <strong>of</strong> manipulator tories or near the vicinity <strong>of</strong> the switching surface. The<br />
control which are based on various control concepts. Lin- system is then robust and insensitive to those interactions<br />
ear feedback control using linearized models as the basis as well as unmodeled dynamics [8], [9].<br />
for decoupling and control has been proposed [2], [3]. The control scheme proposed in this paper is mainly<br />
This method, however, is only valid in case some condi- directed to the tracking problem in high-speed motion. By<br />
tions are satisfied. For example, the rate-linearized con- choosing a lot <strong>of</strong> switching gains, the system first shows<br />
trol method will not perform well while manipulators are the capability to suppress the nonlinear interactions existin<br />
high-speed motion. A nonlinear control system based ing in the manipulator's joints and next shows the capaon<br />
complete decoupling [4] is effective in high-speed mo- bility to suppress the influences resulting from unknown<br />
tion, but the nonlinear compensations are complex and parameter changes.<br />
costly to compute. The Newton-Euler method [5], using<br />
the relationships <strong>of</strong> moving coordinate systems, solves the DYNAMIC MODEL OF ROBOTIC MANIPULATORS<br />
on-line computational problem <strong>of</strong> mechanical manipula- The dynamic model <strong>of</strong> industrial robots or manipulators<br />
tors, although such a scheme still suffers from the fact that is derived from Lagrange's equations [10], which have<br />
the form<br />
Manuscript received October 3, 1985; revised July 15, 1986. This work<br />
was originally presented at the Power Electronics Specialists Conference, d (t3L AL N<br />
Gaithersberg, MD, June 18-21, 1984, under the title, "'Arbitrary Trajec- dt -AJ J - 0 =iF 1, * , (1I)<br />
n<br />
tory <strong>Tracking</strong> Characteristics <strong>of</strong> <strong>Sliding</strong> <strong>Mode</strong> <strong>Control</strong>led Servo System.''"<br />
The authors are with the Institute <strong>of</strong> Industrial Science, University <strong>of</strong> wee0= ~...6 [ an* . T 6] Tar<br />
Tokyo, 7-22-1 Roppongi, Minato-ku, Tokyo 106, Japan. theregeneralze coordinates andveoiy respectively,on and<br />
IEEE Log Number 8612106.tegnrlzdcodntsan eoiy epciey n<br />
0885-8993/87/0400-0169$01.00 ©C <strong>1987</strong> IEEE
170 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. PE-2, NO. 2, APRIL <strong>1987</strong><br />
T = [Tl, , Tn I" is the generalized external force. If the control inputs i1H and u; are selected such that si<br />
The Lagrangian L is expressed as and si always have the opposite signs in the neighborhood<br />
L = K(, O)-P(H) (2) <strong>of</strong> si = 0, that is,<br />
><br />
where P(6) denotes the potential energy <strong>of</strong> the system , 0 lim i < , (9)<br />
and K(6, 0) is the kinetic energy which is written as<br />
then sliding mode will occur on the switching surface si<br />
K = IOTR(6)b (3) = 0. Therefore, the state trajectories <strong>of</strong> the system will<br />
slide and remain on the surface, which implies that the<br />
where the inertia matrix R (6) is symmetric and positive motion <strong>of</strong> the system (6) is dominated by the sliding equadefinite<br />
for all 0.<br />
tion (7). Consequently, the original system, though it may<br />
Substituting (2) and (3) into (1), the equation <strong>of</strong> motion be higher order or nonlinear, behaves like a lower order<br />
for robotic manipulators can be written as and linear system defined by the sliding equation si = 0.<br />
(fn M( \ To<br />
I<br />
guarantee the existence condition (7) <strong>of</strong> the sliding<br />
R(H)H + L 0.<br />
0 -<br />
mode, one need only know the bounds <strong>of</strong> system param-<br />
e _R(6)&<br />
Ki=1 36j / 2 a6 eters for determining the control inputs (8). This clearly<br />
a means that the system is insensitive with respect to pa-<br />
+ - P(O) = T. (4) rameter uncertainties and other disturbances.<br />
Note that the second and third terms <strong>of</strong> the left side <strong>of</strong> the TRACKING CONTROL USING SLIDING MODE<br />
equation represent the Coriolis and centrifugal forces, and<br />
the last term denotes the gravitational forces. Equation (4) Two kinds <strong>of</strong> motions, the point-to-point (PTP) motion<br />
can take a more brief form and the tracking motion, should be considered. Usually<br />
the PTP motion may be regarded as a special case <strong>of</strong> the<br />
R(O)H + M(0) t + g(6) = T (5) tracking motion. Therefore, only the tracking problem is<br />
where<br />
considered here. The tracking capability <strong>of</strong> advanced ma-<br />
=}[06 0 0 022<br />
T nipulators is indispensable for the smooth motion required<br />
g = [,3P/0l, * * *aP/aOn]<br />
in various industrial tasks, such as in an assembly line<br />
T process. In tracking motion the values <strong>of</strong> desired position<br />
Od(t), desired velocity Ad(t), and the desired acceleration<br />
M = {mi,(6)}nx<br />
<strong>of</strong> the manipulator on the entire path are provided by a<br />
supervisory computer.<br />
1 = ln(n + 1). In this paper a new approach is developed to tracking<br />
motion control <strong>of</strong> manipulators.- Each arm is directly con-<br />
SLIDING MODE CONTROL trolled in the manipulator's coordinates (joint space).<br />
Moreover, the proposed method can be easily imple-<br />
<strong>Sliding</strong> mode control based on the theory <strong>of</strong> variable mented in a task-oriented space<br />
structure systems has a number <strong>of</strong> attractive features [7]. When manipulator motion is in low speed the super-<br />
Therefore, it is widely applied to solve various control visory computer is able to neglect those quadratic terms<br />
problems [11]-[ 15]. Consider the following system: 6,6A <strong>of</strong><br />
i<br />
. Then (5) would become<br />
H =f(O, t) + B(H, t)u (6) R(60)0 + g(0) = T. (10)<br />
where 6 e Rn x I, fe Rn x', B E Rn xm, and u E Rmxl. To The desired input torque is computed from<br />
obtain the desired dynamic performance <strong>of</strong> the system, a<br />
set <strong>of</strong> sliding mode equations is previously selected as T = R(6) {I -Kd -Kpe} + g(6) (11)<br />
si(0, t) = 0, i = 1, * , m (7) where Kd and Kp are constant gain matrices. Equations<br />
which defines the switching surfaces. Then the control in- (10) and ( 1) give<br />
puts with discontinuities have been decided according to R(6) { e + Kde + Kp e } = 0 (12)<br />
the switching logic as where e = 6 -<br />
(u+(6, t), if s460, t) < 0 Since the inertia matrix R(6) is nonsingular except in<br />
u1(6, t) = _.(8) some special cases, we may choose Kd and K,p such that<br />
yu, (6, t), If si(6, t) > 0 the characteristic roots <strong>of</strong> (12) have negative real parts,<br />
where u1 (6, t) is the ith component <strong>of</strong> u. The preceding and the error e approaches zero asymptotically.<br />
system with discontinuity control is termed a variable Notice that the effectiveness <strong>of</strong> the preceding control<br />
structure system since the feedback structure <strong>of</strong> the sys- algorithm is grounded on the premise that the manipulator<br />
tem is switching alternatively according to the state <strong>of</strong> the is in low-speed motion, whereas in most cases the manipsystem.<br />
ulator is required to track preplanned trajectories in high
HARASHIMA et al.: TRACKING CONTROL OF ROBOT MANIPULATORS 171<br />
speed, so the influences <strong>of</strong> Coriolis and centrifugal terms<br />
Gravity<br />
M(6) t should also be taken into account. G(8) Compensation<br />
<strong>Control</strong> Scheme In High-Speed Motion<br />
To derive the control laws in high-speed motion, the ld<br />
authors proposed a considerably simpler method based on X<br />
the theory <strong>of</strong> VSS instead <strong>of</strong> a complete nonlinear com-<br />
Feedback<br />
pensation which would need much more computing time.<br />
Compensation<br />
Note that the system state variable 6 and the correspond- ri /<br />
ing differential H can be directly measured. The control Mo,<br />
r<br />
inputs are selected as follows:<br />
T = g(O) + R(){G6d -Ce + rF} (13) S=o<br />
Switching<br />
where C = diag [c, , ca], and r = h,'j}l1. In Surfaces<br />
this case (5) could become<br />
+d'd<br />
R(O){e + Ce} = r(6){F - R1(0) M(H)}H (14) Fig. 1. Block diagram<strong>of</strong> sliding mode control system.<br />
or<br />
posed to contain some unknown parameter variations,<br />
e + Ce = {r - M'(o)}t (15) e.g., the payload change. Taking into account the influ-<br />
X I = R-1(0() M (0). ences caused by the unknown variations <strong>of</strong> system param-<br />
where M'(0) = { m!,(0) }<br />
Define the sliding surfaces as follows:<br />
eters, the equation <strong>of</strong> motion for the robotic manipulators<br />
can be written as<br />
si = ei + ci i = 1, ** n. (16) {R(6, 4) + AR(, A4))} + {M(6, 4)<br />
The existence condition <strong>of</strong> sliding mode<br />
51S, < 0, 0, 0d (17) + AM(0, A))} + {g(0, 4)) + Ag(0, A4)} T<br />
guarantees that the system states continue on si = 0, so (22)<br />
that ei satisfies the equation<br />
where A4) indicates the unknown variations <strong>of</strong> system paei<br />
= -ciei. (18) rameters 4), and AR, AM, and Ag represent the changes<br />
Now that the switching surfaces have been given, one caused by A4) with respect to the R, M, and g, respecneed<br />
only decide the switching logics and the correspond- tively.<br />
ing control gains to guarantee the asymptotic stability <strong>of</strong> To determine the control law, rewrite (22) in the form<br />
the system (15). Since the matrix M'(0) only depends on RO + g = T - ARO - (M + AM) - Ag. (23)<br />
the state variable 0, the upper and lower bounds <strong>of</strong> each<br />
component <strong>of</strong> m!-(6) can be calculated in advance. As- According to the philosophy <strong>of</strong> VSS theory described<br />
sumingXX<br />
previously, the following control law is selected:<br />
m,j < m1-(0) _ me v (19) T= g + R{0d-C + AO +rF + d}. (24)<br />
then by considering (15) and (17) the next inequality could Substitution <strong>of</strong> (24) into (23) results in<br />
be obtained<br />
R(e + Ce) = R{(A - AR*)0 + (r -m*)<br />
E(,yjj - m!) jsi < O, i = 1, * ,n (20) + (d - Ag*)} =0 (25)<br />
and the control gains <strong>of</strong> Ir be determined as<br />
QY:t > fii, if (jsi < 0 or<br />
= ~~~~~~~~(21)<br />
Qry < iM if (sSi > 0. + Ce = (A-AR*)6 + (r-M*)t + (d -g*)<br />
This inequality guarantees that the states <strong>of</strong> system (5)(<br />
move towards and then slide along the given trajectories. (6<br />
Fig. 1 shows the block diagram <strong>of</strong> the sliding mode con- where<br />
trol system.<br />
<strong>Control</strong> Scheme in the System with Unknown Parameter<br />
AR* = {ri(6, A4)}<br />
= R1AR<br />
Variations M* {m1 (6, A))}nx = R'l(Az\M + M )<br />
Following the preceding design procedure, a case willT<br />
be considered in which manipulator dynamics are sup- Ag* = [A gj (O, A4)), ***, Ag* ( , A4 ))] = R-1Ai\g .<br />
nXm<br />
I
172 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. PE-2, NO. 2, APRIL <strong>1987</strong><br />
Under the assumption that a sliding mode starts at and gardless <strong>of</strong> the parameter changes; that is, the robust percontinues<br />
on the intersection <strong>of</strong> switching surfaces si = formance <strong>of</strong> the system will be achieved.<br />
. .<br />
0, (i = 1, *, n), the value <strong>of</strong> si in this case can be<br />
regarded as approximating zero, therefore the following<br />
SIMULATION AND RESULTS<br />
relation is satisfied:<br />
In this section the design guideline described in former<br />
..=- 6d- Ce (27)<br />
sections will be used to design a sliding mode controller<br />
for a three-degree-<strong>of</strong>-freedom manipulator with direct-<br />
Substituting (27) into (26) yields drive arms as shown in Fig. 2.<br />
s=(A - AR* ) (Od - Ce )<br />
inThe parameters <strong>of</strong> the manipulator dynamics are listed<br />
in Table I, where m, s, and L represent the total mass, the<br />
+ (F -M**) + (d - Ag*). (28) position <strong>of</strong> the center <strong>of</strong> mass, and the length <strong>of</strong> each arm,<br />
respectively. Ix, I., and I, are the inertia <strong>of</strong> each arm about<br />
...<br />
Let sji < 0, (i = 1, , n ), then the center <strong>of</strong> mass in its own coordinates. The inertia man<br />
I trix R(0) and matrix M(0) <strong>of</strong> the manipulator dynamics<br />
Z (XAij - AriJ ) (Hdj - §jej)si + >Z (rj - miJ ){jsi (5) have the forms<br />
+(d Ag)S < 0 i = 1, , n. (29) F= r22(0) r23(0)<br />
1<br />
~~~~~~R<br />
(0) = 220) 230<br />
The goal here is then to find a set <strong>of</strong> control gains A, 10 r23(0) r33(0)<br />
r, and d such that the inequality (29) is always satisfied.<br />
One can see that these control gains are easy to determine<br />
if<br />
Arz<br />
M(0)<br />
_ Avr,*' A(01\))_ < Aig 0 m12(0) m13(0) 0 0 0 1<br />
My < m, (0, A4t) < mii m21(0) 0 0 0 m25(0) m26(0)I<br />
Ag_* < lAg*(0, A4) < Akg (30) Lm31(0) 0 0 m34(0) 0 0]<br />
where all the upper and lower bounds are constants.<br />
where<br />
Note that the maximum and minimum values <strong>of</strong> Ar (0, = 0. 6 + 1.35 sin2 02 + 0. 1 sin2 (02 + 03)<br />
A4)), mJ(O, A4)), and Ag*(O, A4)) can be obtained by + 0.4 sin 02 sin (02 + 03))<br />
previous calculation <strong>of</strong> every component <strong>of</strong><br />
R-1(0, 4)) AR(0, A4) r22 = 20.45 + 0.4 cos 03<br />
R1(0, 4){M(0, 4)) + AM(0, A4)} r23 = 0.2 + 0.2 cos 03<br />
R'l(0, 4)) Ag(0, A4))r = 0.2<br />
for all 0. Thus to determine control gains in (29) it is only M12 = 1.35 sin 202 + 0.1 sin 2 (02 + 03)<br />
necessary to know the bounds <strong>of</strong> AA+ on the system param- + 0.4 sin 02 sin (02 + 03)<br />
eters.<br />
Finally, from (29) and (30) the switching gains have M13 = 0.4 sin 02 COS (02 + 03) + 0.1 sin 2(02 + 03)<br />
the forms =<br />
{k± > AyiJ, if si(Od;- Cj) < 0 2 = -0.4 sin 03<br />
l j > if Si(6dj - cjej) > 0 M226 = -0.2 sin 03<br />
in31<br />
= -l/<br />
ZJ > mii, if Sit < 0 in34 = -2/<br />
&y1( < Ain, if > and the gravity terms are<br />
dj> Ag, if s 0. (31<br />
= -39.2 sin 02 -4.9 sin (02 + 03)<br />
As is indicated by the above design procedure, all switch- 5= -4.9 sin (02 + 03)<br />
ing gains are determined by the variation bounds <strong>of</strong> the To examine the validity <strong>of</strong> the proposed method, the<br />
system parameters. So it is clear that the precise tracking whole system is simulated on a digital computer. The decapability<br />
<strong>of</strong> such manipulators will be guaranteed re- sired trajectory in the three-dimensional work space is il-
HARASHIMA et al.: TRACKING CONTROL OF ROBOT MANIPULATORS 173<br />
z<br />
y<br />
-Q<br />
2<br />
O 02<br />
0.2<br />
7~~~~~~~~~~~~~~~P<br />
0.5<br />
Fig. 2. Configuration <strong>of</strong> manipulator with three degrees <strong>of</strong> freedom. Fig. 3. Desired trajectory in work space (meter).<br />
x<br />
TABLE I<br />
MANIPULATOR PARAMETERS<br />
elements <strong>of</strong> (34) are decided. In this case the r is<br />
Link m I IY L S = { L<br />
Number (kg) (kg M2) (kg iM2) (kg i2) (m) (m)<br />
1 13 - - 0.35 0.5 - 20 17.5 6.456 60 0 0<br />
2 10 0.3 0.3 0.15 0.4 0.2 213 0 6 1.168 1.309 0.888<br />
3 5 0.15 0.15 0.10 0.3 0.1 2.174 2.384 1.542<br />
lustrated in Fig. 3. The end <strong>of</strong> the manipulator starts from<br />
{ )ij}<br />
position P, goes along the straight line, and stops at po- Similarly, the control gains A and d can be determined<br />
sition Q.<br />
in the same way as that <strong>of</strong> F<br />
The following experiment contains the unknown parameter<br />
variation A4, which mainly results from a pay- 4.386 0 0<br />
load change. The payload located at the end <strong>of</strong> the ma- 0 0.965 -0.677<br />
nipulator is supposed to vary from 0 to 10 kg. The<br />
switching surfaces are chosen as 0 1.079 0.280j<br />
s = + Ce (32) -4.619 0 0<br />
where C = diag (20, 20, 20) and e = 0 - A- = 0 -0.835 -0.677<br />
The control gains r, A, and d are selected according to L 0 -1.481 0.9471<br />
(30) and (31). First, note that all elements <strong>of</strong> matrix<br />
M*(O, A4) can be obtained from equation =-[0 34.85 79.79]<br />
M*(O, A4) = R'(0))[M(O) + AM(0, A4)] d= -d+.<br />
O Mi2 Mi3 0 0 0 Next, the system's performance will be evaluated in the<br />
= o0 0 m ~m following two cases.<br />
|m 1 O m24 m25 * m26<br />
Case 1: The first motion pattern <strong>of</strong> the manipulator's<br />
Lm* 0 0 Mi34 M3 M end is shown in Fig. 4, where the maximum acceleration<br />
(33 is about 2.7 m/s2 and the maximum velocity is about 1.3<br />
min/s. Fig. 5 shows the tracking error <strong>of</strong> the manipulator's<br />
Then the control gain matrix F can be determined in terms end in work space corresponding to motion pattern one.<br />
<strong>of</strong> (33), i.e.,<br />
Fig. 5(a) is the case without payload variation, and in Fig.<br />
0 'Y12 OY13 0 0 0 5(b) a 4-kg unpredicted payload variation exists. The il-<br />
~~261. ~<br />
lustrated results confirm the fact that the proposed control<br />
r= Y2i 0 0 OY24 'Y25 Y2* (3) algorithm is insensitive to uncertain parameter variations.<br />
_'Y31 ° O34 735 PY36_ Case 2: To illustrate the robustness <strong>of</strong> the developed<br />
sliding mode controller in high-speed motion, the second<br />
Since the upper and lower bounds his, rne <strong>of</strong> inJ (O, motion pattern about the manipulator's end is selected,<br />
z4) can be calculated previously, by ensuring (31) all the which is shown in Fig. 6. Fig. 6(a) is the acceleration
174 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. PE-2, NO. 2, APRIL <strong>1987</strong><br />
(a) . (a)<br />
r2: (n/s) t~~~~.<br />
ir (m/s)<br />
1<br />
(sec) ~~~~~~0. 3 06 .9(sec)<br />
WS)~~~~~~~~~~~~~~~~~~~~~~~~~~~b<br />
0.5 1.0 1. (sec) 0<br />
(b)<br />
Fig. 4. Motion pattern I <strong>of</strong> manipulator's end. (a) Acceleration. (b) Ve- Fig. 6. Motion pattern two <strong>of</strong> manipulator's end (higher speed motion).<br />
locity. (a) Acceleration. (b) Velocity.<br />
0~ ~ ~~~~~~O<br />
(me ter )<br />
(N *m)<br />
0.01 30<br />
(a)<br />
(a)<br />
-01(X 0.5 1.0 1.5 (sec)<br />
0~ ~~ ~~~~~~~~~~~3<br />
(meter)<br />
0.01 30<br />
|\ ' _ @~~~~~~~~~~<br />
Torque due to gravity<br />
0 .2<br />
Q Total torque<br />
0 0.45 0.9 (sec)<br />
® Torque due to coupling inertia<br />
(i Torque due to effective inertia<br />
(d3 Torque due to centrifugal and Coriolis forces<br />
-0.01 '.. .<br />
0 0.5 1.0 1.5 (sec) Fig. 7. Torque components <strong>of</strong> link'3. Payload: 4 kg. (Motion pattern two.)<br />
(b)<br />
Fig. 5. <strong>Tracking</strong> error <strong>of</strong> manipulator's end in work space (motion pattern byteculniera)ndhefth(usdycntfone).<br />
(a) Payload variation: 0 kg. (b) Payload variation: 4 kg. gal and Coriolis forces) cannot be neglected compared<br />
with other terms. In this case the tracking responses <strong>of</strong><br />
pattern with a maximum value <strong>of</strong> 7.5 m/s2, and Fig. the three joints to the desired trajectories and the corre-<br />
6(b) is the velocity pattern with a maximum value <strong>of</strong> 2.1 sponding input torques resulting from sliding mode conrn/s.<br />
The torque components <strong>of</strong> link 3 with a 4-kg pay- trol are shown in Figs. 8, 9, and 10, respectively.<br />
load variation in motion pattern two are shown in Fig. 7. As shown in these figures, the manipulator shows per-<br />
Due to the direct-drive manipulator, the influences <strong>of</strong> fect tracking capability as there are almost no tracking<br />
nonlinear coupling terms such as the third term (caused errors visible between the desired trajectories and the sys-
HARASHIMA et al.: TRACKING CONTROL OF ROBOT MANIPULATORS 175<br />
(rad)<br />
3 t<br />
(rad)<br />
3 03d<br />
2 2 \<br />
0D3<br />
1 / 1 1<br />
t (sec)<br />
0 3 0.6 0.9 0.3 0.6<br />
0.3 0.6<br />
0.9<br />
0 .9<br />
t (sec)<br />
(a)<br />
(N m) (N m)<br />
(a)<br />
600 300<br />
300- ttsec) 150- t(sec)<br />
0 .3<br />
- 30<br />
-600 -300<br />
0.6 0.9 -150 0.3 0.6 0 .9<br />
(b)<br />
(b)<br />
Fig. 8. <strong>Tracking</strong> response <strong>of</strong> first joint to desired trajectory. Payload vari- Fig. 10. <strong>Tracking</strong> response <strong>of</strong> third joint to desired trajectory. Payload<br />
ation: 4 kg. (a) <strong>Tracking</strong> motion. (b) Input torque.<br />
variations: 4 kg. (a) <strong>Tracking</strong> motion. (b) Input torque.<br />
(rad)<br />
servo sampling period is equal to 5 ms. The sensor reso-<br />
3 lution is assumed to be infinite.<br />
2 2 902d<br />
CONCLUSION<br />
The algorithm presented in this paper is for the accurate<br />
02 t (sec) tracking control <strong>of</strong> robotic manipulators. By using the im-<br />
0 0.3 proved sliding mode control law, the nonlinear dynamic<br />
0 . 3<br />
0.6 0 . 9 interactions <strong>of</strong> the manipulator joints are effectively sup-<br />
-1 pressed, and the system is insensitive to the physical parameter<br />
variations. The simulation results demonstrate the<br />
(N.m)<br />
(a) perfect tracking property <strong>of</strong> a robotic manipulator with<br />
three degrees <strong>of</strong> freedom. The nonlinear coupled manipulator<br />
dynamics show well-behaved linear uncoupled<br />
300 characteristics in high-speed motion. In addition, the de-<br />
150 t (sec) sign <strong>of</strong> the sliding mode controller needs less a priori in-<br />
-150 0.43 0.609PI formation about the parameter changes owing to the fact<br />
-300 that control gains need only to satisfy some inequality<br />
300<br />
f conditions associated with parameter bounds. Since the<br />
(b)<br />
variations <strong>of</strong> the control gains are determined according<br />
Fig. 9. <strong>Tracking</strong> response <strong>of</strong> second joint to desired trajectory. Payload to simple switching logic, computation for completing<br />
variation: 4 kg. (a) <strong>Tracking</strong> motion. (b) Input torque.<br />
control inputs can be reduced to the degree that real-time<br />
tracking control is realizable. As for measuring practical<br />
signals, the sensor resolution is always finite and gives<br />
tem's, i.e., the developed control system isin every way rise to the problem <strong>of</strong> affecting the actual control chatterrobust<br />
in high-speed motion even with the existence <strong>of</strong> ing frequency. Therefore, in future research we will take<br />
unknown parametric variations,<br />
Due to the actuator dynamics a simulation has been performed<br />
by introducing first-order filters between each<br />
controller output and the corresponding manipulator input<br />
into account the limitations <strong>of</strong> sensor resolution.<br />
REFERENCES<br />
toqe Th rnfrfnto fec itri hrce- ( 1] V. I. Utkin, "Variable structure system with sliding modes," lEEE<br />
ized by 1 /(1 + ris). T1 = 20mis, T2 = 15 ins, and 73 = Trans. Automat. Contr., vol. AC-22, pp. 212-222, Apr. 1977.<br />
10 ms with respect to joints 1, 2, and 3, respectively. The [2] D. E. Whitney, "4Resolved rate control <strong>of</strong> manipulators and human
176 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. PE-2, NO. 2, APRIL <strong>1987</strong><br />
prostheses," IEEE . rans. Man-Mach. Syst., vol. MMS-10, pp. 47- Fumio <strong>Harashima</strong> (M'75-SM'81) was born in<br />
53, 1969. Tokyo, Japan, on February 3, 1940. He received<br />
13] M. E. Kahn and B. Roth, "The near minimum-time control <strong>of</strong> open- the B.E., M.E., and Dr.Eng. degrees from the<br />
loop articated kinematic chains," Trans. ASME, J. DSMC, vol. 93, University <strong>of</strong> Tokyo, Tokyo, Japan, in 1962,<br />
pp. 164-172, 1971. 1964, and 1967, respectively.<br />
[41 E. Freund, "Fast nonlinear control with arbitrary pole-placement for From 1967 to 1980, he was an Associate Proindustrial<br />
robots and manipulators," <strong>Robot</strong>ics Res., vol. 1, no. 1, pp.<br />
fessor at the Institute <strong>of</strong> Industrial Science, Uni-<br />
65-78, 1982. versity <strong>of</strong> Tokyo. He has been a Pr<strong>of</strong>essor at the<br />
[5] J. Y. S. Luh and M. W. Walker, "On-line computational scheme for above Institute since 1980.<br />
mechanical manipulators," Trans. ASME, J. DSMC, vol. 102, pp. Dr. <strong>Harashima</strong> received the Best Paper Award<br />
69-76, June 1980. in 1978 from the Society <strong>of</strong> Instrument and Con-<br />
[6] M. Takegaki and S. Arimoto, "An adaptive trajectory control <strong>of</strong> ma- trol Engineers (SICE) <strong>of</strong> Japan. In 1983 he received the Best Paper Award<br />
nipulators," nt. J. Contr., vol. 34, no. 2, pp. 219-230, 1981. from the lEE <strong>of</strong> Japan. In 1984 he served as General Chairman <strong>of</strong> IECON-<br />
[7] V. I. Utkin, "<strong>Sliding</strong> <strong>Mode</strong>s and Their Application in Variable Struc- 1984 and received the Anthony J. Hornfeck Service Award from the IEEE<br />
ture System," Moscow, Mir. 1978. Industrial Electronics Society. He is presently the President <strong>of</strong> the IEEE<br />
[8] K. D. Young, "<strong>Control</strong>ler design for a manipulator using theory <strong>of</strong> Industrial Electronics Society.<br />
variable structure system," IEEE Trans. Syst., Man, Cybern., vol.<br />
SMC-8, pp. 101-109, 1978.<br />
[9] J. J. Slotine and S. S. Sastry, "<strong>Tracking</strong> control <strong>of</strong> nonlinear systems Jian-Xin Xu was born in Peking, China on Ocusing<br />
sliding surfaces with application on robot manipulators," Int. tober 1, 1957. He received the B.E.Eng. degree<br />
J. Contr., vol. 38, no. 2, pp. 465-492, 1983. from the Zhejiang University, Hangzhou, China,<br />
[10] P. Paul, <strong>Robot</strong> <strong>Manipulators</strong>: Mathematics, Programming, And Con- in 1982 and the M.S. degree from the University<br />
trol. Cambridge, MA: MIT.<br />
o oy,Tko aa,i 96 ei urnl<br />
[11] K. D. Young and H. G. Kwatny, "Variable structure servomecha- a graduate student in the doctora1 p rogram at the<br />
nism design and applications to overspeed protection control," Au- a University <strong>of</strong> Tokyo. His main interests art in<br />
tomatica, vol. 18, no. 4, pp. 385-400, 1982. control theory, robotics, artificial intelligence, and<br />
[12] S. Lin and S. Tsai, "A microprocessor-based incremental servo sys- cotoFuzzy system theoty.<br />
tem with variable structure," IEEE Trans. Ind. Electron., vol. IE- Mr. Xu is a member <strong>of</strong> the Society <strong>of</strong> Instru-<br />
31, pp. 313-316, 1984. ment and <strong>Control</strong> Engineers <strong>of</strong> Japan, the <strong>Robot</strong>-<br />
[13] A. Sabanovic and D. B. Izozimov, "Application <strong>of</strong> sliding modes to ics Society <strong>of</strong> Japan and the Society <strong>of</strong> Artificial Intelligence <strong>of</strong> Japan.<br />
induction motor control," IEEE Trans. Ind. Appi., vol. IA-17, pp.<br />
41-49, 1981.<br />
[14] 0. Kaynak, F. <strong>Harashima</strong>, and H. Hashimoto, "Variable structure<br />
system theory applied to sub-time optimal position control with an<br />
Hideki Hashimoto (S'84-M'87) was born in Toinvariant<br />
trajectory," Trans. Inst. Elec. Eng. Japan, vol. 104, no. kyo, Japan, on August 15, 1957. He received the<br />
3/4, Mar./Apr. 1984. B.E., M.E., and Dr.Eng. degrees from the Uni-<br />
[15] F. <strong>Harashima</strong>, H. Hashimoto, and S. Kondo, "MOSFET converter- versity <strong>of</strong> Tokyo, Tokyo, Japan, in 1981, 1984,<br />
fed position servo system with sliding mode control," IEEE Trans.<br />
and <strong>1987</strong>, respectively.<br />
Ind. Electron., vol. IE-32, pp. 238-244, 1985. He is currently a Lecturer at the University <strong>of</strong><br />
[16] F. <strong>Harashima</strong>, H. Hashimoto, and K. Maruyama, "<strong>Sliding</strong> mode con- Tokyo, working on control engineering, robotics,<br />
trol <strong>of</strong> manipulator with time-varying switching surfaces, " Trans. Soc.<br />
and artificial intelligence.<br />
Instrum. Contr. Eng., vol. 22, no. 3, pp. 335-342, Mar. 1986. Dr. Hashimoto is a member <strong>of</strong> IEE <strong>of</strong> Japan,<br />
[17] -, "Practical robust control <strong>of</strong> robot arm using variable structure the Society <strong>of</strong> Instrument and <strong>Control</strong> Engineers<br />
system," Proc. 1986 IEEE Int. Conf. <strong>Robot</strong>ics and Automation, vol. <strong>of</strong> Japan, the <strong>Robot</strong>ics Society <strong>of</strong> Japan, and the<br />
1, pp. 532-539, 1986. Society <strong>of</strong> Artificial Intelligence <strong>of</strong> Japan.