Subsectional Adaptive Monte Carlo Localization for Humanoid Soccer Robot

A subsectional adaptive Monte Carlo localization method is presented to overcome some shortcomings in regular Monte Carlo localization, such as particle degeneracy and the kidnap problem. Firstly, two feature variables are proposed to describe distribution of particle set and its difference from the real posture. Secondly, four states (global localization, local localization, local tracking and fault-tolerant localization) are identified by the combination of the variable values during the whole process of localization, and different strategies are designed for each state in order to adjust parameters and resampling rules adaptively. Finally, the results of physical and simulative experiments based on adult-size humanoid soccer robot system show that the proposed method is effective in achieving an accurate and real-time localization. Furthermore, this method can enhance the robustness of localization system by solving the kidnap problem efficiently.