Real-Time Control Of A Two Link Manipulator Using Multi-Layered Perceptrons

Abstract

As the demands for versatility and speed in modern robotics increase, the need for accurate position and speed control becomes more stringent accordingly, to the extent that conventional control strategies can no longer provide satisfactory performances. The use of artificial neural network {ANN} -based feedforward controllers appears to be an interesting and efficient approach to solving the problem, provided efficient inverse dynamics learning is performed. This paper deals with the case of a two-link robot, for which a high performance mixed control architecture is based upon neural networks as well as conventional PID control blocks.