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

Abstract

Mobile robot pose tracking is mostly based on odometry. However, with time, odometric pose tracking accumulates errors in an unbounded fashion. This paper describes a way to decrease the odometry error by using an extended Kalman filter (EKF) for fusion of calibrated odometry data and sonar readings. Common approaches for calibrated odometry and sonar fusion use a feature based map which has two uncertainties in the measurement process. One uncertainty is related to the sonar range reading and the other one to the feature/range reading assignment. Our approach is adapted to an occupancy grid map which has only the sonar range reading uncertainty in the measured process. Experimental results on the mobile robot Pioneer 2DX show improved accuracy of the pose estimation compared to the calibrated odometry.