Iterative Learning Controller for Trajectory Tracking Tasks Based on Experience Database

Abstract

An iterative learning controller based on experience database is proposed for a class of robotic trajectory tracking tasks. It is very general for supporting all types of iterative learning control schemes. The experience database consists of previously tracked trajectories and their corresponding control inputs. The initial control input of an iterative learning controller can be selected properly using a dynamic RBF neural network by properly considering the past experience of tracking various trajectories. Moreover, the RBF network can be created dynamically to ensure the network size is economical. Simulation results of trajectory tracking of a planar two-link manipulator indicate that the convergence speed of the iterative learning controller can be improved by using this method.