Actuator Nonlinearities Compensation Using RBF Neural Networks in Robot Control System

Abstract

In this paper, a compensation scheme is presented for general actuator nonlinearities. The compensator uses two neural networks, one to estimate the unknown actuator nonlinearities and another to provide adaptive compensation in the feedforward path. Since radial basis function network has the universal approximation property and can avoid the local minima problem, the compensator uses RBF neural networks to estimate the actuator nonlinearities and eliminate their effects. GL matrix and operator are introduced to help prove the stability of the system. Rigorous proofs of closed-loop stability for the compensator are provided and yield turning algorithms for the weights of the RBF neural networks. The whole scheme provides a general procedure for using RBF neural networks to compensate the actuator nonlinearities in robot control system. Simulation results verify the effectiveness of the designed scheme and the theoretical discussion.