1987 Volume 5 Issue 2 Pages 139-149
Learning control based on the repetitive operation of robots is one of the useful methods to realize high speed and high precision control for robots. It betters the next operation of a robot by using previous operation's data. From the practical viewpoint, the trial number of learning should be small. However, most previous studies forcuseed on the proof of learning convergence, and few study considered about the convergence rate for learning control.
In this paper, we propose a learning control scheme with high convergence rate. This proposed learning control scheme takes advantage of linear approximation of the inverse robot dynamics. In this learning algorithm, modified control inputs are generated by using position, velocity, and acceleration errors, multiplied position servo gains, velocity servo gains, and motor inertia, respectively. This method is very practical because it does not necessitate the identification of phisical parameters of a robot. Some simulation results are given in order to examine the applicability of this proposed learning scheme to various kinds of industrial robots.