Improved Particle Filter Localization in Crowded Environments for Mobile Robots

In dynamic crowded environments, the localization performance of traditional map-matching algorithms for mobile robot will be significantly decreased, even the localization will completely fail, because of severe changes of the observation information. In this paper, an improved particle filter localization algorithm is proposed based on localizability estimation. On one hand, this algorithm estimates the belief of laser range finder observations using the localizability matrix of observation model. On the other hand, it estimates the belief of the odometer data using the covariance matrix of prediction model. Then based on these two indicators, the predicted robot pose is modified according to the observation information. Experiments of localization and navigation under different typical corridor environments are designed to compare the proposed algorithm with classical particle filter algorithms. The result demonstrates the validity of the proposed localization algorithm under complex environments.