Study of a neural network controller for robotic applications

Computer simulations have been performed for analysing the learning behaviour of a neural network when it is used as a controller in a robotic control system. Specifically, the effect of the backpropagation parameters on the convergence and stability of the network have been investigated. The neural network is configured as a feedforward inverse controller. Results indicated that increasing the number of neurons in the hidden layer will improve the convergence speed. However, beyond a certain limit additional neurons will cause system oscillations and finally instability. The main contribution of this paper is the derivation of a relationship between the rms error and the number of iterations used in the training of the neural network. Guidelines for selecting the neural network parameters are also given.< >