Résumé
Mammals are able to process the location-based (site-dependent) information provided by the spatial layout of landmarks to return to an inconspicuous place (the goal) whose location has been memorized exocentrically by reference to the landmarks. We have developed an algorithm accounting for this place navigation ability. This algorithm is based on the angular information provided by the bearings and angular sizes of landmarks, rather than on metric distances. It is implemented by a system that combines elementary computing procedures and two types of simple associative memory network. One type of network enables the animal to learn how the environmental perspective changes when it makes a one-step translation in any direction, whereas the other type enables the animal to learn how the current panorama is oriented with respect to an overall reference direction. This system highlights the fundamental role of exploration of the environment prior to place navigation, and assigns specific functions to hippocampal place cells and postsubicular head-direction cells.