Approaches to Simulation of Mouse Behaviour in the Morris Water Maze

Abstract

As computer games have become more complex emphasis has shifted from graphics to the use of artificial intelligence to make the behaviour of non-player characters (NPCs) richer and more believable. In order to do this successfully it is necessary to understand what realistic behaviour actually is. In the world of neurological science, the Morris Water maze is frequently used as a way of testing whether mouse or rat behaviour is affected by brain lesions, drugs etc. It provides a very clean and simple environment for assessing the search patterns undertaken by a mouse swimming to find a hidden platform. Our aim is to find a simple approach to providing a realistic approximation to real mouse behaviour, in order to gain insight into general creature behaviour. A solution that combines basic behaviour with A searching showed similar behaviour to mice during training, but tended to be fragile. A genetic algorithm approach based on criteria used for assessing real mouse behaviour fails to converge to anything realistic. We have found that an ant colony model can be made to work reasonably well in replicating the real mouse training, but lacks the wider search behaviour employed by real mice when the platform is absent.