Exploiting Qualitative Spatial Constraints for Multi-hypothesis Topological Map Learning

Abstract

Topological maps are graph-based representations of space and have been considered as an alternative to metric representations in the context of robot navigation. In this work, we seek to improve on the lack of robustness of current topological mapping systems against ambiguity in the available information about the environment. For this purpose, we develop a topological mapping system that tracks multiple graph hypotheses simultaneously. The feasibility of the overall approach depends on a reduction of the search space by exploiting spatial constraints. We here consider qualitative direction information and the assumption that the map has to be planar. Qualitative spatial reasoning techniques are used to check the satisfiability of individual hypotheses. We evaluate the effects of absolute and relative direction information using relations from two different qualitative spatial calculi and combine the approach with a topological mapping system based on Voronoi graphs realized on a real robot.