Improvement for the Rao-Blackwellized Particle Filters SLAM with MCMC Resampling

The ability to simultaneously locate a robot and accurately map its surroundings is considered to be a key prerequisite of truly autonomous robots. Rao-Blackwellized particle filters simultaneous localization and mapping can produce accurate results but it has the tendency to become over-confident. In this paper, the analysis on consistency is presented. The methodology of the Markov Chain Monte Carlo resampling is incorporated to prevent particle impoverishment. The algorithms are evaluated on accuracy and consistency using computer simulation. Experimental results show that the increased diversity of particles can improve the accuracy as well as consistency of RBPF SLAM.