Feed-forward learning with frequency adaptation towards the control of series elastic actuators

Abstract

In this paper, a novel method towards approaching perfect tracking performance in periodic motions for robotic joints with serial elastic actuators, is proposed. The method is in an adaptive feed-forward scheme which has the ability to learn the required controlling signal, leading to reduced tracking error. Ordinary learning feed-forward methods do not have the capability of learning frequency of the motion; but here the method first learns the frequency of a motion adaptively and then basis functions are created based on the learned frequency. Finally the magnitudes of the injected basis functions are found and the tracking error will be lowered. In the simulations the effectiveness of the proposed method for a two DOF planar manipulator is verified. The results for two cases of tracking linear and circular periodic trajectories of end-effector prove satisfactory.