Fast loopy belief propagation for topological Sam

Abstract

SLAM has been one of the main focuses of attention in robotics research. In the last years, some new graphical solutions for this problem have been proposed, which are concerned about jointly determining the environment map and the robot localization history, a more specific problem known as Smoothing and Mapping (SAM). When applied to topological maps, Loopy Belief Propagation (LBP) provides an incremental and distributed solution to this problem, but eventually may incur time-consuming convergence. This work introduces the concept of starting points of belief propagation, a technique that can be used to reduce the convergence time of the LBP algorithm. We then propose an approach for determining starting points using information about the specific SAM graph structure in order to limit the number of iterations needed by LBP to provide an approximated global Maximum a Posteriori (MAP) estimate of the map and the robot trajectory. The experiments presented, performed with real-world data, confirm the adequacy of the proposed approach and encourage further investigation on it.