Robust Monte Carlo Localization for humanoid soccer robot

Abstract

Most of Monte Carlo Localization (MCL) face kidnap problem. A novel method, called state-driven Monte Carlo localization (SDMCL) is presented to solve kidnap problem so that localization of humanoidrobot can be more efficient. In the proposed SDMCL, Focus and near are feature variables to divide the states of particles into four types: messy, approach, cluster and error. The state ‘error’ denotes some dramatic errors in location, such as kidnap. So the kidnap can be detected on line by monitoring the state of particles if the transition of real location of robot is large enough to be found by sensors. Based on the state detected, a novel strategy is proposed to reset the state of particles to avoid revising the particles gradually. The effectiveness of the proposed SDMCL is verified by RoboCup TeenSize humanoid soccer robot, Robo-Erectus Senior. The experimental results showe that the humanoid robot is able to localize itself accurately to perform humanoid soccer game. It also shows that the proposed SDMCL can recover from the kidnap problem quickly while holding its superior performance in the precision and stability of localization.