An improved FastSLAM using resmapling based on particle swarm optimization

摘要

FastSLAM is well known as a Rao-Blackwellised particle filter formulation of simultaneous localization and mapping. The accuracy of conventional FastSLAM degenerates over time due to the particle depletion in resampling phase. In this paper, an improved FastSLAM with resampling based on Particle Swarm Optimization (PSO) is proposed to address the problem. Firstly, instead of rejection and replication, PSO is employed in the resampling process for the pose convergence of the particle set. Then, in order to update the proposal distribution and the map, a special framework of FastSLAM is adopted in the proposed method. Finally, the result of computer simulation reveals that the modified method shows smaller error in both robot pose and feature estimations than conventional FastSLAM.