Generalization of Iterative Learning Control for Multiple Desired Trajectories in Robotic Systems

Abstract

Iterative learning controllers are found to be effective for trajectory tracking tasks in the robotic systems especially when the system model is not known. One of the drawback of iterative learning control is its slow convergence and high tracking errors in the initial iterations because of zero knowledge about the system for each new desired trajectory. In this paper, importance of the initial control input in the convergence of error is highlighted. Experience of iterative learning controller for different desired trajectories is modelled using neural network. For a new desired trajectory, this neural network generates the initial control input which is used by the learning controller. This approach is proved to be very effective in improving the convergence of the tracking error. The proposed method is very general and applicable to most of the iterative learning controller without modifying their simple learning structures.