Efficient and Reliable Monte Carlo Localization with Thinning Edges

Abstract

The global convergence of MCL is time-consuming because of a large number of random samples. Moreover, its success is not guaranteed at all times. This paper presents a novel approach to reduce the number of samples of MCL and one heuristic approach to detect localization failure using thinning edges extracted in real time. Random samples are drawn only around the neighborhood of the thinning edges rather than over the entire free space and localization quality is estimated through the probabilistic analysis of samples added around the thinning edges. A series of experiments verified the performance of the proposed scheme.