A new approach to global self-localization with laser range scans in unstructured environments

Global positioning is a key issue for mobile robotics. Laser range scanners are widely used on mobile robots. This paper presents a new algorithm which determines the position and orientation of the robot by matching an 180° laser range scan to a sensor based 2D world model deduced from previously recorded laser range scans. The matching is done on a special transformation of the laser range scans, which we will call gestalt features. Building up a pre-sorted database of these gestalt features for selected scans is essential to achieve real time speed. The algorithm is suitable for unstructured environments, because no assumption about the existence of straight, rectilinear lines, rectangular edges or similar within the environment are made. Furthermore, the algorithm is designed to be robust when distracting temporary obstacles are present in the laser range scan or even in the world model. Geometric error modeling and reasoning with intervals is essential to achieve good results.