Topological robot localization by training a vision-based transition detector

This paper presents a mobile robot localizer that detects topological transitions using a color vision-based classification system. After a two-step training process, the localization system is shown to achieve sufficient reliability to track the position of a mobile robot across two floors of an office building using purely discrete (non-probabilistic) update. Experimental results are based on an off-the-shelf electric wheelchair fitted with a parabolic color camera