A neural system for qualitative mapping and navigation in visual environments

Abstract

We describe a real-time, view-based neural system for unsupervised exploration, adaptive map-making, and navigation within 3D environments defined by a spatially distributed set of visual landmarks inspired by on analogy to learning aspect graphs of 3D objects, the system comprises two neurocomputational architectures which respectively model place learning and cognitive mapping in the rat hippocampus. The first architecture performs unsupervised place learning by combining the "What" with the "Where", namely through conjunctions of landmark identity, pose, and egocentric gaze direction within a local sensory view of the environment. The second associatively learns action consequences by incorporating the "When", namely through conjunctions of learned places and coarsely coded robot motions. Together, these networks approximate a Markovian decision process for an environment, and consequently provide a neural substrate for exploration, environment recognition, and behavioral route planning. Preliminary results from ongoing implementations on a mobile robot called MAVIN (the Mobile Adaptive VIsual Navigator) demonstrate the potential for these capabilities.