An integrative framework for global self-localization

Abstract

Concerning the robustness of mobile robot navigation, global self-localization is a key feature for many service applications. In this paper we describe an efficient Bayesian approach for hybrid topological/metric navigation, which is designed to exploit information from multiple sources of sensor data. Experiments with a combination of odometry/laserscans/computer vision show the system was able to generate initial position hypotheses, to cope with environmental ambiguities and to recover from severe position errors.