Experimental studies on robustness of a learning method with a forgetting factor for robotic motion control

摘要

P-type learning control algorithms for manipulators are quite simple and easily implemented compared with the D-type, since differentiation of velocity signals is unnecessary. When initialization errors, fluctuations of dynamics, and measurement noise exist, the convergence of trajectories to a neighborhood of a given ideal trajectory is uncertain in the P-type algorithm. However, manipulator motion trajectories in P-type learning control that includes a forgetting factor are uniformly bounded. Moreover, if command input data in a long-term memory are updated selectively after every few operational trials, output trajectories converge to a neighborhood of the desired one. In this paper, experimental results are presented, which show the robustness and convergence of this proposed method, and the best choice of a forgetting factor is discussed based on these experimental results.