Вы находитесь на странице: 1из 6

Gain Scheduled Control for Robot Manipulator's Contact Tasks on Flexible Environments

Jianqing Wul), Zhiwei Luo2)and Koji Ita'),*)


1). Department of Information and Computer Sciences, Toyohashi University of Technology 1-1 Tempaku-cho, Toyohashi 441, Japan tel: +81-532-47-0111 (ex. 562); fax: +81-532-48-3422; e-mail: wu@system.tutics.tut.ac.jp tel: +81-532-47-0111 (ex. 501); fax: +8 1-532-48-3422; e-mail: koji@system.tutics.tut.ac.jp 2). Bio-Mimetic Control Research Center, The Institute of Physical and Chemical Research(R1KEN) 8-3 1 Rokuban 3-chome, Atsuta-ku, Nagoya 4.56, Japan tel: +8 1-52-654-9 137; fax: +8 1-52-654-91 38; e-mail: luo@nagoya.bmc.riken.go.jp

Abstract: This paper studies on a robot manipulator's


contact task on its flexible environments. Because of its jlexibility, the environtnent dynamics will influence the robot's control system, and since it is a distributed parurneter system, the object dynamics seen f r o m the robot's endeffecter will change when the robot moves on its different locations. In this paper, the object's distributed parameter dynamics is approximated into a linear pnrnmeter-vnving system (LPV). The robot's control space is decomposed into a position control sub-space and ci force control sub-space. Optimal state feedback is designed f o r the position control loop, and gain scheduled control is applied f o r the force control loop. The effectiveness of this approach is shown bv computer simulations.

1 Introduction This paper studies on how to control a robot manipulator to perform the mechanical contact tasks on its flexible environments, for example, to maintain a solar battery plate of an artificial satellite as shown in Fig.1.

robot's end-effecter will change when the robot moves on its different locations [ 1],[3]. The problem becomes further complicated such that it is difficult to decompose the robot's control space [2],[4],[5]. Although the position control subloop is not influenced by the contact force, the inverse is not true. In our early studies, we first determined the robot endeffecter's position response, and then treaded the force loop as a linear time-varying system(LTV). We then applied simple adaptive control algorithm for this force loop [4],[5]. Simple adaptive control [8] constructs the system's control input from the input, the state of a reference model and the output of a error system. It dose not require the knowledge about the value and variations of the system parameters. It guarantees the system's Lyapunov stability. However, as well as many other adaptive control algorithms, simple adaptive control does not guarantee the system's transient response. In this paper, we further apply the recent gain scheduled control theory [SI,[ 113 in designing the force control loop. To do this, we first analyze the relation between the robot's contact position and the flexible object's dynamic parameters, we then design the robot's position and force control inputs. We find that, the position loop's control design influences gain scheduled control of the force loop. We show the effective-ness of this control approach using computer simulations.

U
2 Dynamic Model of Flexible Object and R o b o t
Fig. 1 Maintaining the battery plate of an r artificial satellite by a robot a m Because of its flexibility, the object dynamics will influence robot's control system, and since it is a distributed parameter system, the object dynamics as seen from the In this section, without loss of generality, we simplify the contact task shown in Fig. 1 as the contact task on a flexible cantilever, a freeend beam as show in Fig.2(a). For brevity, we assume the following two conditions. 1) the flexible object is very stiff and only make

0- 7803- 3219- 9/96/$5.00 O 1996 IEEE


512

AMC'96-MIE

a small deflection when the robot imposing a contact force on it. 2) there is no friction on the objects surface.

where 1 is the length of the beam, subscript i is the mode number, mi is natural frequencies. If we only consider the vertical effective of the 1st. mode, the dynamic equation of the beam at the operating point c can be approximated as
IG+ dfi

+ kq = c f(t)
d(c) =%!??!?% ,
Q(C)

(6)

I(c) = & Q(c)

where I, d and k are coefficients which change with respect to the operating point position c.

(b)
Fig.2 Robots contact task on a flexible cantilever

2.1 Approximated Model of Flexible Object


The dynamic equation of the cantilever[6] is Fig.3 The relation between z and 0 On the other hand, the relation between z and 0 is shown in Fig.3, where 01-Ci is the tangent line and B-CI-C? is the normal line of the object at the contact point Ci. Shifting 01-Ci to the cross point 0, we get a new parallel line 0-C?. It is easy to show that

at?

pA ax2&

pA at4

pA

(I)

where z(x,t) is the vertical displacement of the beam as shown in Fig.2(b), E1 denotes bending stiffness, pA is mass per unit length and 5 is damping coefficient. 6(x) is delta function; and c is the location of the contact force f(t). According to the expansian theorem, the solution of eq.(1) can be expressed as

e = etan0 -

cos0

(8)

If the object is very stiff, then e+O, tan0=0 and cos)+ 1. Therefore,

equations:

(5) Here, 0 is the rotation angle of the contact force f, when

513

there is no friction on the object's surface, and p,=c is eq.(12) with respect to time t, the following equations are obtained, length of the object curve. Note that, the above parameters I, d and k are varying with . . respect to the operating point position c --- the robot's endx = Jx, (13) effecter position. It is the first difficulty in the robot's hybrid X = JX, + JXc (14) control. = [ cos e +sine 2.2 Cartesian Coordinate in Space of Reference, Object and Task sin 8 pccos 8

(15)

Let (xb,yb),(x,,y,) be the reference frame fixed relative to the robot basis and the object frame fixed relative to the fixed side of the beam as shown in Fig.4. To establish the dynamic equations of robot-object relatively to the contact point C, we define @,e) the task frame as follows. as

where J is the Jacobian matrix from the task frame (p,8) to the reference frame (xb,yb).

2.3 Dynamic Model of Robot Manipulator


As shown in Fig.4, the dynamic equation of the robot manipulator in the reference space are

where x = [x,,ybIT is the robot's end effecter position, M(x) is the robot's inertia matrix, p(x,x) is the vector of Coriolis and centrifugal forces, g(x) is the verctor of gravitational force, Fe is the robot's driving force, F, is the contact force between the robot's endeffecter and the object. Specifying the following nonlinear feedback compensation Fig.4 Local approximated model of the object at fixed operating point First, we translate the frame (x,,y,) to the contact point C , and then rotate it such that its yo axis directs to the direction of the contact force F, ( F, is the force exerted on robot end-effecter by the beam.). Because it is assumed that there is no friction on the object's surface, the direction of the contact force Fn coincides with the normal vector n of the object's surface at the contact point C. We further move the , frame (x,,y,) along the direction of the x and the distance p, (p, is the length of the arc from the origin of the frame (x,,y,) to the contact point C.). The frame (p,8) transformed finally is defined as the task frame. Let x = [x,,y,IT, x, = [pc,elTbe the robot end-effector's position relative to the reference frame and task frame, respectively. The geometric relationship between x and x, , can be expressed as

Fe = M(x)JX, + p(x,X) + g(x) + F,,

+ M ( x ) J M ~ ' ( u U,) -~
into eq.(16),we get

(17)

where U, = [0, p,f,IT; U, = [u1,u2ITis the new control input vector, f, is the magnitude of the contact force F,, and M, = diag[m, ,m2] denotes a desired inertia matrix. Eqs.( lo), (18) and (19) represent the simplified robotobject dynamics. Although the position loop is independent from the contact force f, in eq.(18), the force loop is dependent on the position p, in eq.(19). The object's dynamic parameters I(p,), qp,) and k(p,) are varying with respect to p,.

Hybrid Control Design

where x,, is the distance from the origin of the frame (x,,y,) to the origin of the frame (xo,yo). By differentiating

Because of the two difficulties mentioned above, we have to consider the problem how to design position/ force hybrid control when the position and force control loops are

514

coupled. Due to that the position control loop is independent on the contact force fn, we can design the position control loop previously to the force control loop. This choice ensures to manage the influence of the position p on the , force control loop. Based on the choice, we can regard the position variable p as a exogenous parameter in the force , control loop, which are unknown a priori values but can be measured on-line, and hence we enable translate the system, eq.(lO) and (19), to the linear parameter-vary system with the gain scheduling parameter p, [9],[12]. In this section, we first design the position control loop using the optimal state feedback. and then establish the gain scheduled controller for the force control loop.

Fig.
3.

Optimal state feedback for position control loop


Force Control Loop
and (19) in the state-spaceform (26)

Rewrite the qs.(lO)

3. 1 Position Control Loop


First, we set m, = 1, andrewrite eq.(lX) in the following state-space form

(27)

where

bp=[

3,

e:=[

:,]
(22)
ydt) = fn(t). This is a linear parameter-varying (LPV) system with the scheduling parameter p , . Gain scheduled control design is shown in Fig.6.

The desired end-effector state vector is given as


Xpd

= [pd, pdlT

As proposed in [4] and 151, the position loop's control input is designed as Ul(t) = Xpd - R' bpPXp(t) which optimize the following performance index [7].
1 T

(23)

Matrix P is the solution of the following Riccati equation.


PAp + A;fP - PbPR-'b;fP + Q = 0 (25)

' I
Fig.6 Controller of the force control loop

As larger as Q is selected, as faster as the position response


is.

First, select the state-space form of the weight Gw(p,) as

where II'lIL' represents the L' norm and

4 Simulation Results

To examine the performance of the approach described above, the computer simulation of a two degree of freedom planar robot manipulator slides on a cantilever has been done. The desired displacement and contact force are specified a? following form.

where,

The design parameters in the position and force controller is chosen as follows: 1. in the position controller we set m l = 1, Q = I , R = 1, then P is solved as

Then, based on Watanabe's control design[l 11, if there exist a symmetric matrix Q(p,) for E>O and 6>0, Q(pc) satisfies the following conditions:

2 . in the force controller 6 = 1x10'" E = 0.145 ,

, ,

y = 0.01 , vmax 30.0 =


-

0.4332 -0.3316 -0,2606

-0.3316 0.6516 0.5335 -3.3800 7.3070

-0.2606 0.5335 0.4516 -0.1686 2.0758

6.9480 -3.3800 -0.1686 6.4170 7.5960

1.0214 7.3070 2.0758 7.5960 3.9560

Qo =
6.9480

1
then. the feedback control

1.0214

Here, Qo is the optimal solution of eq.(23) which satisfies the following conditions.

(34) stabilizes the system (eqs.(30),(31)), and Fig.7 and 8 show the simulation results. Fig.7 shows the (35) robot's position responses, and Fig.8 is the contact force responses along the normal of the beam.

W L 2 , WZ0 )

516

Displacement(m)
....... .......

Japan (#07750534, #07243 105, #06452253).

1.50

References
[ 11 M.H.Raibert and J.J.Craig, "Hybrid PositionForce

0.50

'"V
2.00

0.00 0.00

I
4.00
6.00

Control of Manipulators" ASME J. Dyn. Sys. Meas. Contr., Vol. 102,pp. 126- 133,1985

8.00

10.0

Time(sec)

[2] Z.Luo and M.Ito, "Control Design of Robot for Compliant Manipulation on Dynamic Environments" 01.9, IEEE Trans., on Robotics and Automation, V NO.3, pp.286-296,1993
[3] T.Yoshikawa and A.Umeno, "Dynamic Hybrid Control of Manipulators Considering Object Dynamics" J. Rob. Soc. of Japan, Vol.11, No.8, pp. 1229-1235, 1993

Fig.7 Response of the robot's position


Force(N) Measured Values
1.50

............ Desired Values

[4] J.Wu, Z. Luo, M. Y amakita and K. Ito, "Robot Manipulator's Contact T asks on Uncertain Flexible Objects" Proc. of KACC, 1995

[SI J.Wu, Z. Luo, M. Y


0.50

y
2.00 4.00
6.00

1
8.00
10.0

amakita and K. Ito, "Adaptive PositionForce Hybrid Control of Manipulator on Uncertain Flexible Objects Proc. IFAC'96 (To appear )
"

0.00 1 0.00

Timdsec)

A.Hac, "Repetitive Control of Bilinear Systems with Application to Semi-active Isolation of V ibration" Int. J. Control, Vo1.61, No.6, pp.1265-1296, 1995 B.O.Anderson and J.B .Moore,"Optimal Control: Linear Quadratic Methods" U.S.A.Prentice - Hall, 1990 [8] H.Kaufman, I.Bar-kana and K.Sobe1, "Direct Adaptive Control Algorithms" New York: Springer-Verlag,1994
[9] J. S. Shamma and M. Athans, "Analysis of Gain Scheduled Control for Nonlinear Plant IEEE Trans. on Automatic Control, Vo1.35, No.8, 1990
"

Fig.8 Response of the robot's contact force

5 Conclusions

This paper proposed a dynamic position/force hybrid control using a prevail strategy: position control loop is designed previously to force control loop, for the contact task on a flexible object in which it is difficult to decompose robot's control positron and force control loop, independently The coordinated controller consists of two parts: the position controller which is designed by optimal state feedback and the contact force controller which is designed by gain scheduling approach. The effectiveness of the approach is illustrated through computer simulations with the contact task on a flexible cantilever. As the further research, we will consider the case where the flexible object's dynamic parameters are unknown. W e will also study the influence of the position variation to the performance of the force control loop.

[ 101 P.Apkarian and P.Gahinet, "A Convex Characterization

of Gain-Scheduled H cu controllers" IEEE Trans. on Automatic Control, Vo1.40, NOS, 1995


[ 1 11 R.Watanabe, K.Uchida, M.Fujita and E.Shimemura"Hw

Control of Linear Systes with Scheduling Parameter" SICE, V01.31, N0.4, pp.481 - 488, 1995
[ 121 J. S. Shamma and M. Athans, "Gain Scheduling:

Potential Hazards and Possible Remedies" IEEE Contr . Sys., pp.101- 107, June, 1992 [13] A. S. Banach and W . T . Baumann, "Gain-Scheduled Control of Nonlinear Partial Differential Equations" Proc. of the 29th. CDC, pp.387 - 392, 1990
[ 141 M.K.F.Knoop and A.M.Perez, "Nonlinear PI-controller

Acknowledgments

Design for a Continuous-flow Furnace via Continuous Gain Scheduling" J. Proc. Cont., V 01.4, No.3, 1994

This work was supported through the scientific research fund from the Ministry of Education, Science and Culture of

517

Вам также может понравиться