Covisibility-based map learning method for mobile robots

Abstract

In previous work, we proposed a unique landmark-based map learning method for mobile robots based on the "co-visibility" information i.e., very coarse qualitative information on "whether two objects are visible together or not". In this paper, we introduce two major enhancements to this method: (1) automatic optimization of distance estimation function, and (2) weighting of observation information based on reliability. Simulation results show that these enhancements improve the performance of this proposed method dramatically, not only in the qualitative accuracy measure, but also in the quantitative measure. © Springer-Verlag Berlin Heidelberg 2004.