Evaluation of Algorithms for indoor mobile robot self-localization through laser range finders data

In this paper an evaluation of several indoor mobile robot localization algorithms is presented. Two-dimensional laser range scans are used to estimate motions without a kinematic model of the robot and, moreover, without using odometry. Features extracted from each scan are used as landmarks of the environment to estimate relative robot position in an unknown environment. Different known feature extraction algorithms have been deeply tested in order to evaluate their performances. A data association method uses simultaneous observation of features from two consecutive scans to determine correct features associations. Two different strategies are tested to estimate relative robot position using common features of two consecutive scans. All of these algorithms have been tested with two different angular resolutions of the laser scanner, and a comparison between all of them, in term of number of features identified and execution-time, is presented.