Hebbian Multilayer Network in a Wheelchair Robot

Motivation to study Hebbian learning arises from its neurophysiological plausibility and its suitability for hardware implementation. Up to now, artificial Hebbian learning, embedded in a real system that performs adaptive motor control, has been restricted to one-layer networks. To overcome this limitation, a novel approach to adaptive preprocessing based on Hebbian learning is presented. It is shown how this network is integrated in an adaptive motor control system inspired by classical and operant conditioning models. Experimental results with a real mobile robot are described.