Internal model control of a robot using new neural networks

Abstract

The use of neural networks for control of a robot manipulator is presented in this paper. The control system consists of a neural model of the robot, a neural controller and a conventional PID controller. The control structure uses internal model control (IMC). The Alopex method is employed as a learning algorithm to train the networks. The standard backpropagation (BP) algorithm is also utilised for comparison with the Alopex learning algorithm (ALA). The proposed network is a recurrent hybrid network which is suitable for identification and control of robot manipulators. Compared to neural networks with pure nonlinear hidden processing elements, e.g., the diagonal neural network, the proposed recurrent hybrid network converges faster than taught to identify linear and nonlinear dynamics systems. Simulation results are presented to evaluate the performance of the IMC for the control of a SCARA-type robot manipulator.