A feedforward/feedback neural control structure and its application to a robotic system

Abstract

In this paper, a feedforward/feedback neural control structure with learning capability is proposed, which consists of a neural identifier and three blocks of controllers: a feedforward controller for generating the computed torque, a PD controller, and a multilayer neural controller for compensation of error and disturbance. A higher order multilayer neural network is used for constructing the neural identifier and the neural controller combined with the conventional controllers in cascade. The control performance is compared with that of the conventional feedback/feedforward controller without neural networks. Computer simulation shows that the proposed control structure is very effective in controlling a robotic system.