Robust self-localization of mobile robot

Abstract

A large sample size is needed for Monte Carlo localization (MCL) in multi -robot dynamic envi- ronment due to the frequent robot kidnap phenomenon making too little samples locate in the regions where the value of desired posterior density function (PDF) is large. A new localization method named genetic Monte Carlo localization (GMCL) is proposed. The crossover and mutation operations in evolutionary computation are introduced into MCL to make samples move towards regions with large value of PDF, so the sample set of GMCL can represent the desired PDF better. Experiment results show that GMCL needs fewer samples and is more precise and robust in the dynamic environment.