Högskolan i Skövde

A higher order recurrent connectionist architecture for adaptive control of autonomous robots is introduced in this paper. This architecture, inspired by Pollack's Sequential Cascaded Network, consists of two sub-networks: a function network for the coupling between sensory inputs and motor outputs, and a context network, which dynamically adapts the function network in order to allow a flexible mapping from percepts to actions. The approach taken here is compared to dynamics and algorithmic approach to autonomous robot control, and it is argued that the above architecture allows an integration of (a) the complex structure and control typical for the algorithmic approach, (b) the capacity to utilize systematically continuous state spaces, and (c) the self- organizing learning capacity of connectionist systems with a simple, but powerful mechanism for context-dependent adaptation of behaviour.