Adaptive obstacle avoidance with a neural network for operant conditioning: experiments with real robots

Abstract

Gaudiano et al. (1996) have shown that a neural network model of classical and operant conditioning can be trained to control the movements of a wheeled mobile robot. The neural network learns to avoid obstacles as the robot moves around without supervision in a cluttered environment. The neural network does not require any knowledge about the quality or configuration of the sensors. In this article we report results using our neural network with the real mobile robot Khepera.