Self-localization addresses the problem of estimating the pose of mobile robots with respect to a certain coordinate system of their workspace. It is needed for various mobile robot applications like material handling in industry, disaster zone operations, vacuum cleaning, or even the exploration of foreign planets. Thus, self-localization is a very essential capability. This problem has received considerable attention over the last decades. It can be decomposed into localization on a global and local level. Global techniques are able to localize the robot without any prior knowledge about its pose with respect to an a priori known map. In contrast, local techniques aim to correct so-called odometry errors occurring during robot motion. In this thesis, the global localization problem for mobile robots is mainly addressed. The proposed method is based on matching an incremental local map to an a priori known global map. This approach is very time and memory efficient and robust to structural ambiguity as well as with respect to the occurrence of dynamic obstacles in non-static environments. The algorithm consists of several components like ego motion estimation or global point cloud matching. Nowadays most computers feature multi-core processors and thus map matching is performed by applying a parallelized variant of the Random Sample Matching (pRANSAM) approach originally devised for solving the 3D-puzzle problem. pRANSAM provides a set of hypotheses representing alleged robot poses. Techniques are discussed to postprocess the hypotheses, e.g. to decide when the robot pose is determined with a sufficient accuracy. Furthermore, runtime aspects are considered in order to facilitate localization in real-time. Finally, experimental results demonstrate the robustness of the method proposed in this thesis.
Where am I? Contributions to the Localization Problem of Mobile Robots
Wo bin ich? Beiträge zum Lokalisierungsproblem mobiler Roboter
2012
Sonstige
Elektronische Ressource
Englisch