|Abstract / Kurzbeschreibung:
The area of autonomous mobile robots will play an increasing role in our daily life in the future. Especially in the field of service robotics and rehabilitation robotics, large progress is expected. In contrast to previous systems that only move in specially adapted surroundings, this new generation of robots should be used in an environment that was designed for people's needs and in which special changes for the use of such systems are either impossible or undesirable. Hence, service robots must be able to find their way in a human-oriented, natural environment, i.e., they have to navigate in it. In this thesis, an image processing method is presented that enables a robot to orientate in unchanged, existing environments. The approach is based on the use of one-dimensional 360° (panoramic) color images that are taken by a special optical sensor. From two of these images, the Panama algorithm that is presented in this thesis determines the spatial relationship between the positions where the images have been taken. The approach can be compared to methods that determine the optical motion flow but it addresses a different application: it is used to determine the spatial relations between positions that may be located several meters apart. This image processing method is embedded in a navigation technique in which the environment is represented as a network of routes along which the navigating agent can move. These routes are represented as sequences of panoramic images. This route knowledge is acquired by teaching. A teacher only presets the routes that should be learned but not the images that are required for their description. These images are independently selected by the autonomous system during the training, depending on the particular environment and the system's kinematic restrictions. The navigation method was implemented and tested on a simulated as well as on a real system. The real mobile system was an electric wheelchair -with the application rehabilitation robotics in mind. In addition to the navigation method, an approach for the use of the introduced image processing method for metrical self-localization is presented and its precision is analyzed. Furthermore, a second navigation method is presented that describes routes as sequences of basic behaviors. The recognition of routemark constellations triggers the switching between these behaviors. The method detects possible navigational errors and tries to recover from them by backtracking the previous way.