Fusion Diskreter Planung mit dynamischer Pfadplanung auf Basis von Circular Fields

Kurzfassung

In the near future robots are thought to become an integral part of human everyday life. Also in industrial settings robotic Co-Workers are expected to become a commodity. Even though the particular application areas may vastly change, a robot always needs to act in a dynamic and partially unknown environment. It shall reactively generate motions and prevent upcoming collisions. Furthermore, knowledge of the obstacle geometry, distance, and structure shall be incorporated into strategic decision making. Earlier research at DLR showed that Circular Field (CF) planners, which base on dynamical systems and their reaction to virtual electromagnetic forces associated to the environment, show promising calculation speed and convergence properties. However, at the same time the variety of parameter choices of the technique can lead to very different global behavior, which might not necessarily lead to a good choice. In turn, well selected adaptation properties may also be exploited for online strategy determination on decisional level. The goal of the thesis was to handle this inherent complexity of possible context dependent strategy parameterization. For this, based on initial dynamic reachability analysis, several approaches where developed to find suitable and robust paths in real-time that take into consideration the environmental state. More specifically, a predictive multi-agent system for strategy evaluation and selection based on suitable cost functions was designed. The developed algorithm takes possible future collisions due to obstacle motion into account and incorporates this knowledge into the strategy finding. Together with an object binding algorithm that clusters and organizes the environment representation and is handled by an environment observer, this system is able to solve complex motion planning problems in real-time. Finally, the 6D MATLAB/Simulink/Python implementation of a hybrid Circular and Potential Fields approach is used to perform the experimental analysis for static multi-object parcours and to successfully avoid dynamically moving humans in a 6D task motion.