Achievement of trapeze motion for humanoid robots based on realtime ego-motion / moving objects' motions recognition algorithm

Realtime ego-motion / moving objects' motions recognition function enables robots to behave under dynamic environments. Especially, visually guided multi-robot collaborative dynamic motion generation, such as trapeze motion, is focused as a challenging application of such realtime motion recognition function in this paper. For achieving that, we propose a realtime recognition method for ego-motion and moving objects' motions at the same time by integrating 3D optical flow and the inertia information. Furthermore, we applied this algorithm to a realtime motion recognition problem between two humanoid robots swinging a swing. As a result, we have achieved a successful trapeze motion executed by real humanoid robots.