混合正規分布モデルを用いた自律移動型ロボットの自己位置推定問題の検討

抄録

A localization problem of autonomous robot is a very important issue in Robocup Middle-Size League. Using an omnidirectional camera image, lines in the field around a robot are detected and used as the image feature to identify the robot states. Because the localization must run in real-time and with high robustness, Monte-Carlo localization technique is often used. However, because of the symmetry of the field, particles are sometimes moved to a wrong position and evaluated to have 'not bad' state variables. Therefore the estimated distribution model must express the subpopulation of the particles and the representative values for the subpopulations must be used to decide the robot action. In this paper, the particle's distribution is expressed using gaussian mixture model and its parameters are estimated by expectation maximization algorithm. Using simulation results, the usefulness of the localization model is examined.