The piecewise monte carlo localization system for a humanoid soccer robot

This paper presents the piecewise Monte Carlo localization method that changes the size of sample set and resampling rules dynamically driven by the state of filter at each recursive update step. The states of filter is divided into global localization and local tracking based on two feature variables: focus and near. Physical walking toward target experiments and simulative kidnap problem experiments were performed on the humanoid soccer robot platform. Experimental results show that the piecewise MCL can acquire useful localization data in a short time and track its position timely with an acceptable errors in both distance and orientation. Moreover, it can transfer between global localization state and local tracking state correctly.