A Visual Based Extended Monte Carlo Localization for Autonomous Mobile Robots

An extended Monte Carlo localization (EMCL) method is proposed in this book chapter by introducing two validation mechanisms to apply a resampling strategy to conventional MCL. Two validation mechanisms, uniformity validation and over-convergence validation, are effectively used to check abnormity of the distribution of weight values of sample set, e.g., observation deviation or over-convergence problem. The strategy of employing different resampling processes is proposed to construct more consistent posterior distribution with observations. This new approach is aimed to improve localization performance particularly with smaller sample size in the non-modeled robot movements, and thus achieve global localization more efficiently. A vision-based extended MCL is further implemented, utilizing triangulation-based resampling from visual features in a constraint region of the pose space. Experiments conducted on a mobile robot with a color CCD camera and sixteen sonar sensors verify efficiency of the extended MCL method.