Safe Navigation of a Wall-Climbing Robot - Risk Assessment and Control Methods

Safe navigation on vertical concrete structures is still a great challenge for mobile climbing robots. Although there is a large field of application it is hard to find the optimum of applicability and safety. Applicability comes with demands for fast navigation speed, high maneuverability, easy handling by the user, and high payload (e.g. inspection sensors, tools), whereas safety requires a more defensive system behavior. This thesis addresses the problem of safe navigation in the range of wall-climbing robots. So far, closed-loop controlers for adhesion and locomotion are not sufficient to avoid a drop-off in certain situations. Therefore, additional measures are needed to improve system's safety. For adhesion safety a behavior-based network has been developed combining closed-loop control behaviors and deliberative components. After all, the adhesion control network is analyzed online to determine the current state of the system via an evaluation function, which is optimized using a genetic algorithm and training examples. The functionality of the developed approaches and the benefit for safety is proven in real-world experiments as well as in a simulated environment. It is shown that the prototypic robot is able to detect and avoid risky surfaces and that navigation safety could be improved tremendously.