Computation of the optimal relative pose between overlapping grid maps through discrepancy minimization

Abstract

Grid maps are a common environment representation in mobile robotics. Many Simultaneous Localization and Mapping (SLAM) solutions divide the global map into submaps, forming some kind of graph or tree to represent the structure of the environment, while the metric details are captured in the submaps. This work presents a novel algorithm that is able to compute a physically feasible relative pose between two overlapping grid maps. Our algorithm can be used for correspondence search (data association), but also for integrating negative information in a unified way. This paper proposes a discrepancy measure between two overlapping grid maps and its application in a quasi Newton optimization algorithm, with the hypothesis that minimizing such discrepancy could provide useful information for SLAM. Experimental evidence is provided showing the high potential of the algorithm.