Robust Monte Carlo localization based on vector model

This paper proposed an enhanced Monte Carlo localization algorithm, which is more effective, stable and robust than traditional localization algorithm by using many strengthening mechanisms, such as vector model, re-initialization and reverse convergence. The vector model redefines the pattern of environment map, so that the localization result is not limited by the resolution of map. Re-initialization gives second chance when the algorithm is missing the right location and can't jump out the local solution. Reverse convergence, the most important in this paper, can let the algorithm spread particle swarm moderately. It is simple but very useful, especially for the case within noise or sensing distance limitations in the sensors. The simulation results also show the excellent performance of proposed algorithm.