Adaptive Temporal Difference Learning of Spatial Memory in the Water Maze Task

The Morris water maze task is a spatial memory task in which an association between cues from the environment and position must be learned in order to locate a hidden platform. This paper details the results of using a temporal difference (TD) learning approach to learn associations between perceptual states, which are discretized using a Self Organizing Map (SOM), and actions necessary for a robot to successfully locate the hidden platform in a "dry" version of the water maze task. Additionally, the adaptability of the temporal difference learning approach in non-stationary environments is explored.