Dual Population Coding for Path Planning in Graphs with Overlapping Place Representations

Abstract

Topological schemes for navigation from visual snapshots have been based on graphs of panoramic images and action links allowing the transition from one snapshot point to the next; see, for example, Cartwright and Collett[5] or Franz et al.[9]. These algorithms can only work if at each step a unique snapshot is recognized to which a motion decision is associated. Here, we present a population coding approach in which place is encoded by a population of overlapping “firing fields”, each of which is activated by the recognition of an unspecific “micro-snapshot” (i.e. feature), and associated to a subsequent action. Agent motion is then computed by a voting scheme over all activated snapshot-to-action associations. The algorithm was tested in a large virtual environment (Virtual Tübingen[24]) and shows biologically plausible navigational abilities.