タングルトポロジーを用いたロボットハンドによる人間の持ち替え動作の模倣

抄録

This paper proposes a method for imitating human regrasping motion by a robot. The method is based on the learning from observation (LFO) paradigm, in which human motion is recognized by a task model and robot motion is reproduced from the recognized task sequence. For designing a task model for regrasping, we focused on a topological criterion, Gauss linking integral (GLI), which represents a tangle state of two strands. Fingers and an object are represented by strands in this study, and the relationship among them is described in a topological way based on GLI. In this paper, first, a task model for regrasping is proposed, and then, a method for recognizing human regrasping motion using the task model is described. Next, the proposed method is validated by reproducing a regrasping movement in a robotic hand. Human regrasping movements of a pen-like object are considered. The successful reproduction of the regrasping movement verifies the proposed methodology to be useful and proved that it is feasible to control a robotic hand by imitating human.