Kalman filter theory based mobile robot pose tracking using occupancy grid maps

In order to perform useful tasks the mobile robot's current pose must be accurately known. Problem of finding and tracking the mobile robot's pose is called localization, and can be global or local. In this paper we address local localization or mobile robot pose tracking with prerequisites of known starting pose, robot kinematic and world model. Pose tracking is mostly based on odometry, which has the problem of accumulating errors in an unbounded fashion. To overcome this problem sensor fusion is commonly used. This paper describes two methods for calibrated odometry and sonar sensor fusion based on Kalman filter theory and occupancy grid maps as used world model. Namely, we compare the pose tracking or pose estimation performances of both most commonly used nonlinear-model based estimators: extended and unscented Kalman filter. Since occupancy grid maps are used, only sonar range measurement uncertainty has to be considered, unlike feature based maps where an additional uncertainty regarding the feature/range reading assignment must be considered. Thus the numerical complexity is reduced. Experimental results on the Pioneer 2DX mobile robot show similar and improved accuracy for both pose estimation techniques compared to simple odometry.