A cobot being in close contact to a human worker needs to adapt its trajectory by an online manner. Otherwise the robot has to wait for the human to clear its workspace, which would lead to a robot waiting most of the time. The local environment is captured with 3D perception systems and then transferred into a GPU-optimized environment model. The robot and its planned motions are also represented as GPU environment model. By taking advantage of the thousands of parallel threads on the GPU and the large memory it is possible to evaluate the current planned path (swept volumes) for collisions. This approach is independent of the complexity of the environment or planned path and can be executed with a fixed response time. The deterministic character is an important aspect in context of human robot safety. Thus, humans and robots can work together in the same workspace. The robot detects collisions in advance and can react and replan in a very fast and efficient manner, leading to less collisions and an increased productivity. The subsystem does not rely on prior knowledge of the environment and can quickly be transferred to new applications in the competence centres.
- Not specified