Sonar-based Pose Tracking of Indoor Mobile Robots

Abstract

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 the problem of mobile robot's local localization or pose tracking with prerequisites of known starting pose, robot kinematics 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 a simple odometry calibration method and compares two fusion methods of calibrated odometry data and sonars' range data based on Kalman Filter theory. One fusion method is based on standard Extended Kalman Filter and another one, proposed in this paper, on the Unscented Kalman Filter. Occupancy grid map is used as the world model, which is beneficial because only sonars' range measurement uncertainty has to be considered. If a feature based map is used as the world model then an additional uncertainty regardingthe feature/range reading assignment must be also considered. Experimental results obtained with the Pioneer 2DX mobile robot (manufacturer ActivMedia Robotics) show that better accuracy of pose estimation and smoother robot motion can be obtained with Unscented Kalman Filter.