Neural network based iterative learning controller for robot manipulators

Abstract

An efficient neural network based learning control scheme is proposed to solve the trajectory tracking controI problem of robot manipulators. The proposed approach has four distinctive characteristics: 1) good tracking performance can be achieved during the first learning trial; 2) learning algorithm for adjusting neural network weights is independent of the manipulator dynamic model, thus displays strong robustness to torque disturbances and model parameter uncertainty; 3) no acceleration measurement or estimation is needed; and 4) real-time implementation with a higher sampling rate is readily possible. Simulation results on a 3 degree-of-freedom manipulator are presented to show its validity.