A neural network-based controller for a two-link robot

Abstract

A case of a multilayer perceptron (MLP) used for position control of a two-link robot is reported. Simulation results as well as the computational burden on neurocontrollers designed for robot control are presented. Such issues as the number of layers and number of nodes per layer are discussed. It is concluded that a neural network can be used to approximate a dynamical model of a robot. However, the error associated with this model is not nearly as good as that of conventional controllers, specifically a computed torque controller.<>