Yours Sincerely

iii iv Traditional robot control involves following a structure of perception, planning and then action. In the perception and planning phases a world model is built up and subsequently used to determine action. Behaviour based robotics departs from this organisation by removing the world model and employing a structure where the robot is controlled through a combination of parallel behavioural modules. This project involved the design and implementation of a software framework for the development of behaviour based applications for use with Eyesim. Basic behaviours and a simple controller were developed to demonstrate the framework. Adaptive ca-pabilities were implemented to extend the system and allow the robot to successfully react to changes in the environment. The adaptive capabilities allow for improved navigation through environments of varying densities. The main adaptive controller was built around the Q-Learning algorithm. This controller was trained using carefully chosen training environments in order to optimise its performance. Q-Learning has been demonstrated to provide a marked improvement in performance. v v