Accurate vision based position tracking between places in a topological map

This paper presents a method for accurately tracking the position of a mobile robot which moves between places in a previously learned topological map. Places in the map are represented by sets of visual landmarks extracted from panoramic images. Probabilistic localisation methods and the landmark representation enable position tracking within places. A sensor model is presented which improves the accuracy of local position estimates, and is robust in the presence of occlusion and data association errors. Position tracking between places requires the recognition of place transition events and the passing of local position estimates between places. This paper presents such a system and reports real world position tracking results from paths through topological maps.