AAS 98-120

IMPLEMENTING TIME OPTIMAL ROBOT MANEUVERS USING REALISTIC ACTUATOR CONSTRAINTS AND LEARNING CONTROL

Ju Li, R. W. Longman - Columbia University; V. H. Schulz - University of Stuttgart, Germany; H. G. Bock - University of Heidelberg, Germany

Abstract

This paper is one in a series of papers on time optimal control of robots, aiming to bridge the gap between theoretical/numerical results and routine application of time-optimal robot path planning in a factory environment. The research is building toward a demonstration of increased productivity on a Mercedes Benz press chain, where one robot is the slowest to accomplish its task and hence determines the cycle time of the chain -- we aim to speed up that robot and thus make the chain operate faster. In this paper, we develop a set of realistic inequality constraints that must be satisfied by the optimized trajectory. These constraints are far more complicated than the typical torque limits appearing in the literature. The physical constraints are related to lifetime or wear considerations in the drive systems, lubrication in the bearings, contact stresses, ratcheting limits, etc., and result in a mixture of pointwise speed and torque limits as well as limits on weighted speed and torque, as well as speed and torque alone, averaged over a cycle. Once the limits are formulated, the issue of how to command the commercial robot feedback controllers so that they actually execute the desired trajectory. Here we suggest the use of learning control.

Back