Está en la página 1de 4

A supervised path planner

A. Zolghadri, M. Monsion and B. Bergeon LARFRA-ENSERB, UniversitC Bordeaux I 33405 Talence, FRANCE and GR. Automatique, CNRS, Grenoble, FRANCE

Abstract- We present in this paper a supervised time : Xc(t) , Yc(t) which is the minimum-time travelling path planner for a Cartesian manipulator devoted path that the tool follows within the prescribed path deviation to contour tracking. While tracking a contour by bounds at comer points and subject to the limits on velocity a biaxial system, there is always a compromise and acceleration. The controller has to drive the motor between the quickness of the whole operation and position to follow the predetermined path. the accuracy subject to the contour errors : in In practice, the manipulator is not able to follow practice, the manipulator is not able to follow axactly the desired trajectories elaborated previously. The exactly the desired trajectories elaborated maximum velocity and acceleration are not constant previously. T h e maximum velocity and throughout the entire workspace; for some motions high acceleration are not constant throughout the velocities and accelerations can be achieved, while for others entire workspace; for some motions high the maximum velocity and acceleration should be decreased velocities and accelerations can be achieved, due to the actuator limitations, otherwise the contour error while for others the maximum velocity and would be intolerable. To prevent this drawback, the system acceleration should be decreased due to the proposed here is able to decide when and how the reference actuator limitations, otherwise the contour error trajectory must be reactualized, taking into account the would be intolerable. To prevent this drawback, situation detected by the sensors. the path-planner presented here is combined with The global project is concerned with the whole a supervision layer. The objective of this is to automation of the control design and its maintenance trough a calculate the possible local trajectories which supervision layer. The control scheme is a Multiple ensures the control loop linearity and increases Reference Model Robust Control (MRMRC) design [61, the path-tracking performances including the robust stability, robust disturbances rejection To demonstrate the efficiency of our supervised and robust tracking performance. path planner when the actuator limitations occur, The solution of the MTPP problem proposed in this the simulations are done with and without the paper aims at the objective of supervision and it is based on supervision layer. When the supervisor is on, the notion of local path planning like in Kim and Shin [2]. the performances are increased significantly. Its parameters are revisable when a failure situation is detected by the supervision layer. It is applicable to the fine path Keywords: Supervision, Path Planning, Contour tracking. planning as well as the rough motions, i.e. it is not necessary to assume that the maximum allowable velocity is achieved on each segment contrary to Luh [l] and Kim and Shin [2]. In I. INTRODUCTION this perspective the non-linear optimization techniques applied to the global path (Luh [ l ] and Imamura [3] ) are not The application is a two dimensional manipulator (biaxial available. These techniques are implemented off-line and do contouring system). The aim of control system is to track a not take into account the real movement of the robot. tool along a predesigned 2D contour by coordinating the movement of the DC motor of each axis and according to prespecified tolerances, depending on the final use of the tool. 11. A Flexible Path Planning Method for the Supervisory The overall system is depicted in Fig. 1 : Purposes of a Biaxial Contouring System.

A. Problem Statement and Notations

Fig. 1: Global scheme of the contour tracking system.

The task planner generates a desired (geometric) path in Cartesian space. This geometric path does not contain any timing information but includes only spatial positions, it is composed of a set of comer points p(i), i=1,...,M. The path planner will then convert the desired path as a function of

The problem is to track a two-dimensional contour as fast as possible subject to velocity and acceleration constraints and contour errors constraint. The real, physical constraints on the manipulators are the applied torques/forces at the axis. The actuator torque and force constraints can be arbitrary functions of the joint positions and velocities. To solve the path planning problem, one needs the upper bounds on joint acceleration. To fully utilize manipulator's capabilities, these bounds are driven from the arm dynamics and so the full manipulator dynamics should be modelled. However the dynamic model of

7803-0078/91/0600-1710$01.00 01991 IEEE

manipulator implies complex resolution methods and therefore a computation time incompatible with the on-line motion generation and its supervision. In this paper maximum constant velocity and acceleration bounds are assumed for each axis of the biaxial system. It is in practice a realistic assumption according to the fact that the x-axis and y-axis are decoupled prismatic links and there is no significant variation of inertia during a contouring operation. The velocity constraint is derived from the maximum allowable voltage and the maximum possible acceleration for each axis can be estimated from maximum motor torque, and the highest inertia and frictional forces expected to occur allowing a factor of safety. Mathematically, we can formulate the problem as follows: Given a geometric path composed of M segments s(i), i=1, ...,M formed by connecting M+l corner points p(i),
i= 1,...,M:

B . Path Planning Algorithm In order to solve LPP(i) problem, we suppose that a parabolic transition is made to connect one path segment to another. According to the value of the maximum allowable velocity, we can distinguish four cases:

Min I t i t
0

Subject to : j=1,2 j=1,2 o s [contourerrod s e(i) i=1,...,M (1) and are the limits on velocity and Where Jmm acceleration of the axis and e(i), i=l,..,,M-1 the maximum path deviation bounds at comer points. 02

vs(i+l) is attained

No

Yes

No

Yes

IYJI 5 Ymax j

qmax

This problem naturally leads to a non-linear programming problem with high dimensionality. If a traditional trapezoidal velocity profile is assumed for each segment, there will be 3M unknown parameters for the whole path planning. This is in contradiction with the objective of supervision of the path planner. In order to solve the problem discussed above, we propose dividing the global path planning problem (GPP) into a set of local path planning problems (LPP) as in Kim [2]:
M-1

The condition for attaining the maximum allowable velocity on s(i) can be obtained easily: 2 VS(1) (Vb - va) + 2(VS(i)2 - v i - vb) + VbVa xaxb 2 (7) 2a.m where (xa, va) and (Xb, Vb) are the states (position and velocity) of the tool at the end of the (i-l)th transition and at the beginning of the ith transition. This is a generalization of the particular case treated by Dombre and Khalil 141 with va=vb=O. Assuming that the vector e(i) is calculated at mid-time of the transition, then

where

GPP =

C LPP(i)

e(i) = q(i)r(i)+ef(i)r(i+l) at(i) = a,o(i)r(i)+atf(i)r(i+ 1)

(2)

i= 1

As an example, the fourth case can be expressed as below:

The vs(i) and as(i), the maximum allowable vclocity and the maximum allowable acceleration along S(i) respectively, can bc represented by : vs(i) = q(i) r(i) (3) as(i) =S(i) r(i) where r(i) represents a unit vector along: di). i.e. P(i- 1) Fig. 2: Local path planning The uppcr bound of the constant acceleration during the transition is set to:
(9) ymax*f (a(')) where a(i) is the angle between as(i) and as(i+l) and f is a factor caculated with respect to the constraints on speed and accelcration. If the maximum allowable velocity is not

and q(i)-and S(i) -denote the magnitude of the maximum allowable velocity and the maximum allowable acceleration in the direction of r(i) respectively. If we denote
1 2 vmax = (vmax Vmax

1' and ymax

1 2 (ymax ymax )t (5)

the vectors of the limits on velocity and acceleration of the motors, q(i) and 6(i) can be computed as:

1711

attained on s(i), this constant accelaration must verify the inequality: 2 (ato(i)%i))2 (adi-1Mi-1)) < &(i)s(i)-as(i) (ato(iMiI2+adi-1x i - 1)2) (10) Note that when tracking a contour with a biaxial contouring system, usually O.Omm I e(i) 5 0.3mm z(i)at(i) << vs(i) and so z(i+ l)at(i+l)<cvs(i+l) (11) and consequently we always need a deceleration phase before tracking the comer point and an acceleration phase at the end of the transition. For instance, we give the expressions of the positions xo , x1 ,x2 , x3 ,xf in the first case.(see figure 2).

b3) Compute the acceleration at(i) during transition by equation (9) and with respect to the inequality (10). b4) Test the inequality (7): whether the maximum allowable velocity is attained on s(i) and s(i+l). Choose one case among the four possible cases. b5) Compute the initial position xo(i+l)=xf(i)of the next segment. b6) If i<M then go to step bl. c) For the last segment (i=M): Compute the maximum deceleration as(M) toward p(M) so as to have a zero speed on this point.
111. Simulation Results.

xf = p(i) + Ldi)z(i) {.s(i)- atfti)z(i) ] 2 as(i+l) atO(i+ l)z(i+l) + a,o(i+l)z(i+l)(.s(i+l) + 1 2 as(i+l)


~

The simulations are done to demonstrate the efficiency of our flexible path planner when the actuator limitations occur. The function of the supervisory layer is to keep the motor from overloading too frequently. When the controller is tracking a local trajectory LPP(i) (as described above), if the supervisor detects a failure situation, i.e., u(kT) > [*urnax where [ is a factor of safety, the feedrate is set to a lower value for tracking the LPP(i+l). Otherwise, the feedrate keeps its optimal value determined by the path planner algorithm. The supervision intervention leads to the respect of the contour constraints, but the tracking time of some LPP(i) will be longer than their optimal value.

The velocity at the switching point on s(i) is : m - 1 3 0 (i) ~ ( i ) ~ 01 - as(i)) -

/
Fig. 3: Global scheme of supervised path planning solver.
As we can notice according to the simulation curves, when the supervisor of the path planner is out of service, the control signal saturations bring about intollerable contour errors. As soon as the supervision layer is on, the actuator saturations are compensated by an automatic reactualization of the synthesis parameters of the path planner on some local trajectories. In this example the total time for tracking the contour is increased to less than 8% while these adjustements are done, but the contour is tracked with respect to the contour constraints.

When using a discrete sample time Ts and a fixed acceleration as(i), it will not be possible in most cases to reach the desired velocity Vm exactly in a discrete number of samples. A method for solving the problem is to modify the acceleration

used :
a) Compute the non-discrete number of samples needed : N1 = v, / (as(i)*Ts) b) Round the value to the nearest upper integer, called N. c) Compute the modified acceleration : as(i)= vm / (N*Ts). The path-planning solver discussed thus far can be summarized by the following algorithm: a) Set i=O al) Compute the maximum acceleration a ( ) s0. a2) Compute the initial position for xo(1) on s(1). b) Set i=i+l bl) Compute the unit vectors r(i) and r(i+l). b2) Compute velocity bounds vs(i) and vs(i+l) and the maximum deceleration as(i) along s(i) and the maximum acceleration as(i+l) along s(i+1).

IV. Conclusion.

A supervised local path planner has been presented which allows real-time adjustment of its constraint

1712

parameters, so that linearity of the control loop could be guaranteed almost every time. The supervision layer keeps the actuators from overloading, and it can set the bounds of acceleration and velocity to a higher value if it decides that they are too conservative. To increase the performance level of the present supervision equipement, some progress has to be made. The necessity of inegrating some known logical functionings about the plant, estimator, control law and path planning algorithm, involves some techniques of artificial intelligence and decision theories. Implementation of the supervision layer using the S2D2 (Static, Simiotic and Dynamique Design) specification method [5] is in progress. The whole supervisory system will be then a real-time knowledge-based system working online with the plant.

REFERENCES [l] J. Y. S. Luh, C. S. Lin , Optimum path planning for mechanical manipulators, ASME, Vol 102, Juin 1981. [2] B. K. Kim, K. G. Shin , Minimum-time path planning for robot arms and their dynamics,-IEEE Trans. Sys., Man, Cyber., SMC-15 : pp 213-223, 1985. [3] F. Imamura and H. Kaufman, Time optimal countour tracking for machine tool controllers, Proceedings of ACC 90, San Diego, pp 1100-1108. [4] E. Dombre et W. Khalil, ModClisation et commande des robots, Editions H e r d s , 1988,Faris. [5] B. Bergeon, J. L. Ermine, A. Khaddad, M. Monsion, Generic Expert System for Identification of Industrial Processes, IFAC-IFORS Symposium on Identification and Parameters Estimation, Budapest, Hungary, July 8-12,1991. [6] S. Ygorra, B. Bergeon and M. Monsion, Design methodology for Robust Multiple Reference Model Contro1,ECC91 Grenoble, FRANCE.

-YllltS

y l s a t

20
15 10

.
60

5 0 5
-10 -15 -20 _.
~

40

20
KT

.
100 200 300 400 500 600

KT

X-motor control signal

Y-motor control signal

Feedrate

Output of path planner in XY plane:The tracked trajectory in XY plane:-

1
-10
-15

SUPERVISOR ON:
v a t l r , :

. KT

i
20

X-motor control signal

Y-motor control signal

100 200 300 400 500 600 70 Feedrate

Output of path planner in XY plane:The tracked trajectory in XY plane:-

1713