Asymptotic convergence of feedback error learning method and improvement of learning speed

Abstract

Deals with the improvement of learning speed based on the analysis of the convergence of the feedback error learning method. The authors derive the condition for the asymptotic convergence of the feedback error learning method for each trial. This condition is the relationship between the learning rate and the /spl alpha/ function, which is calculated from the input-output relationship of the system. Using the /spl alpha/ function obtained above, a high-speed learning method for a trajectory control system is obtained. Simulations results are given for the trajectory control of a 1 link robot manipulator in two cases: (1) using a general feedback error learning method and (2) using the proposed high-speed learning method. The simulation results show the effectiveness of the proposed conditions and learning method.