Real-time self-localization in unknown indoor environment using a panorama laser range finder

Abstract

This paper deals with self-localization of a mobile robot on the condition that no a-priori knowledge about the environment is available. The applied method features accurate, robust, independent of any artificial landmarks and feasible with such a moderate computational effort that all necessary tasks can be executed in real-time on a standard PC. The perception system used is a panorama laser range finder which takes scans of its present environment. A modified dynamic programming algorithm provides pattern matching and pattern recognition on the preprocessed panorama scans and thereby renders a qualitative fusion of the sensory data. For an exactly quantitative estimate of the robot's current position, a robust localization module is employed. The knowledge gained on the environment is stored in a self-growing, graph based map which combines geometrical information and topological restrictions. Preliminary experiments in an ordinary office environment proved the reliability and efficiency of the system.