i

i

An Architecture for Adaptive

Control Behaviour

Mark Nelson
[email protected]

Department of Computer Science
University of Western Australia, Nedlands 6009.

ABSTRACT

The fundamental motivation for building mobile robots is to enable them to perform useful tasks autonomously in real world environments. Because of occlusion objects in these environments are usually only partially perceivable from most viewpoints. Thus a robot cannot be expected to ever possess complete knowledge of its environment, and therefore its navigational control system must be capable of reacting to unexpected events and adapting its behaviour to comply with the continually changing constraints imposed by the environment. Previous approaches towards the development of mobile robot control systems have not achieved a truly adaptive capability. It is conjectured that a capability for adaptive behaviour may lead to the simulation of aspects of intelligence, which is the objective of this research.

This work presents a control system that has been evolved in terms of a sequence of navigational behaviours which (it is argued) probes increasingly deeper into the cognitive processes of intelligence. This incremental development is applied to a distributed architecture whose philosophy states that intelligent behaviour is a consequence of simple computational processes interacting in special ways. This approach has been implemented in a computer simulation from which experimental results indicate that a complex, adaptive behavioural capability has been achieved.

Keywords and phrases: mobile robot navigation, control system, reactive and adaptive behaviour, evolving intelligence.