10 CBA 6 Cong. Latinoamericano de Controle Automático-PREPRINT 1 IMC Scheme Using Neural Networks for Robot Arm

This work explores the use of the internal model paradigm for control (IMC) implemented using neural networks (ANN). A Motion Control problem involving a two-degrees manipulator robot arm is considered. The purpose of the control system is to make the output of the plant track a desired angle. As the study is performed using simulation methodology the robot arm is replaced by its mathematical model. This model is used in a feedback control structure to generate couples of input / output pairs to be used as training set for the neural network. The neural network has a feedforward topology and the learning algorithm used is backpropagation. The individual neurons are modeled as static systems and the dynamics of the network is introduced by delays between pairs of input neurons. So, the trained network constitutes a dynamical neural model of the arm. It is used as internal model in an IMC strategy. Several tests were performed to study the performance of the system. Some simulation results are shown