Identification and control of a robot using a neural network

Abstract

This paper presents a closed-loop methodology for the identification of the forward dynamics of an actual industrial robot by using a multilayer feed-forward neural network. The problems encountered when using the open loop identification procedure is highlighted and the suggestion to overcome these identified problems are then made. The indirect control scheme is employed to control the robot arm. This control scheme is based on back-error-propagation algorithm which consist of neural network identification and neural network controller. Experimental results are presented to demonstrate the capability of the neural network controller to position the robot arm.