Relative Topometric Localization in Globally Inconsistent Maps

Mobile robot localization is a mature field that over the years has demonstrated its effectiveness and robustness. The majority of the approaches, however, rely on a globally consistent map, and localize on it in an absolute coordinate frame. This global consistency cannot be guaranteed when the map is estimated by the robot itself, and an error in the map will likely result in the failure of the localization subsystem. In this paper we introduce a novel paradigm for localization, namely relative topometric localization, by which we forgo the need for a globally consistent map. We adopt a graph-based representation of the environment, and estimate both the topological location on the graph and the relative metrical position with respect to it. We extensively evaluated our approach and tested it against Monte Carlo localization on both simulated and real data. The results show significant improvements in scenarios where there is no globally consistent map.