Hybrid, metric-topological representation for localization and mapping

Abstract

This chapter describes an approach for indoor spatial representation, which is used to model the environment for the navigation of a fully autonomous robot. The metric and topological paradigms are integrated in a hybrid system for both localization and map building: A global topological map connects local metric maps avoiding the requirement of global metric consistency. This allows for a compact environment model, which permits both precision and robustness and allows the handling of loops in the environment during automatic mapping by means of the information of the multimodal topological localization. The presented implementation uses a 360 laser scanner to extract corners and openings for the topological approach and lines for the metric method. This hybrid approach has been tested in a 50 × 25 m 2 portion of the institute building with the fully autonomous robot Donald Duck. Experiments are of four types: Maps created by a complete exploration of the environment are compared to estimate their quality; Test missions are randomly generated in order to evaluate the efficiency of the approach for both the localization and relocation; The fourth type of experiments shows the practicability of the approach for closing the loop. © 2007 Springer-Verlag Berlin Heidelberg.