Topological Mapping and Navigation for Mobile Robots with Landmark Evaluation

A topological mapping and navigation system based on natural landmarks for mobile robots is presented. The landmarks are extracted from natural scenes based on visual saliency, whose relationships are captured by hidden Markov model to distinguish similar scenes. Based on those, the vertex of topological map is built. A competitive learning is designed to evaluate the availability of landmarks according to its utility during the period of mapping and navigation. The landmarks with low evaluation are removed from the map. A similar landmark combination strategy based on clustering is also designed. Experiments show that the system has high accuracy of localization, increases the utility of landmarks effectively and supports navigation in long duration.