The University of Hull

Artificial Intelligence aims to produce autonomous systems capable of demonstrating intelligent behaviour. One way of doing it is to attempt to replicate processes used in human intelligence. This thereby closely links Artificial Intelligence with Cognitive Science. This project is about the design, implementation and testing of an agent architecture based on cognitive models. It especially focuses on the use of affects in activity prioritisation. It is expected that the use of affective states arising out of a study of the theory of emotion and control states will provide a theoretical standing for this work. This will enable the use of adaptive control states (based on affect) to provide a systematic and efficient mean to deal with goal conflicts and other control issues in agent architecture. This project confirms this thesis through the implementation of affect-based agents. These agents were incrementally developed, in a systematic progression, in order to test the influence of different aspects of the design architecture. The testbed used for determining performance issues is a five-aside football simulation. This testbed enables both cooperative and competitive situations for a multi-agent society. Experimentation with the implemented agents allowed various forms of comparison between different and similar agent types and classes. It is shown that the affect-based architecture could be very efficient. The use of affect to direct perception (affordance) and control conflict resolution looks very promising. The thesis discusses some future work on improving the agents presented here with memory, learning and communication capabilities. The thesis finishes by considering the wider aspects of the work.