Sensorbasierte, echtzeitfähige Online-Bahnplanung für die Mensch-Roboter-Koexistenz

Human-robot cooperation is a very active field of research, aiming to combine the specific capabilities of robots and humans to enable new forms of labor. One requirement for successful cooperation or coexistence is the safety of the human operator in the presence of heavy-weight and dangerous robot arms. As a second goal, high availability of the robot system is important to facilitate its adoption in industry. Thus two goals can be identified: first, detection of the human to enable safety measures, and second, path planning to avoid blockades by evasive behaviour of the robot arm. Both aspects are covered in this work. Detection of the human operator and other obstacles is accomplished by a network of static cameras, monitoring the common workspace of the human and the robot. The static cameras allow to use an efficient background subtraction method and – through calibration – the use of models of known workspace object. The latter provides a means to detect occlusions of unknown objects (such as the human) by known objects. These are considered and handled correctly in the image-based collision detection and distance calculation methods presented in this work. Additionally, methods are described to cope with dynamic workspace backgrounds, which are common in industrial work spaces and impose difficulties on background subtraction methods. Using a distance calculation method, distance-based velocity control of the robot is realized to ensure safe human-robot coexistence. The path planning method investigated here uses the available sensor data to react to current obstacle situations in each system cycle (100 ms). The developed concept combines global path planning on a randomized, static graph in a high-dimensional configuration space with the limitation of costly collision tests in the vertices and edges of the graph. Unknown collision or distance information in the graph is estimated from known information, which is stored over system cycles. A framework is presented to adapt this information according to the workspace activity. Thus, with this concept it is possible to meet real-time demands in the presence of dynamic obstacles while achieving statistical completeness over system cycles in static environments Two path planner variants are investigated: a path-optimizing variant based on the collision test and a time-optimizing variant based on the distance measurement. The latter is especially well suited for human-robot cooperation as it also generates paths with extended clearance increasing the ergonomy of shared work spaces.