ADAPTIVE PARTICLE FILTER FOR SELF-LOCALIZATION OF ROBOCUP 3D SOCCER ROBOTS

摘要

Accurate predictions of positions and orientations are especially challenging for a biped robot due to the noisy measurement of the sensors as well as complexity and stochasticity in the robot motions. In this paper, we implement an adaptive particle filter (PF) to localization of our biped soccer robots. We first design a simple way to model the kinematics of biped walking through the supervised learning approach. Subsequently, the adaptive PF is implemented in the RoboCup 3D simulation soccer robot game to predict accurate positions and orientations of the Nao biped robots. The experimental results show significant improvement in performance against the localization method based on Kalman filter.