Controlador neural com camada oculta definida por meio de algoritmo genético aplicado ao posicionamento de um manipulador robótico

Abstract

This work presents the position control of a robotic manipulator with two degrees of freedom. Such control is accomplished by applying a multilayer neural network, emulating a PD2 controller. The choice of the number of neurons in the hidden layer is made with the aid of a Genetic Algorithm, whose role is to generate populations where individuals will be the number of neurons in the hidden layer and evaluate the performance of the control system when subjected to different references signals. The two best individuals of each generation will give origin to the population of the next iteration, noting the probabilities of crossover and mutation, using the criterion of elitism as a warranty of optimization. The evaluation of the overall performance of the control system, in turn, is based on trajectory following response (sinusoidal reference signals) and on position response (step reference signals).