ユーザの修正経路教示に基づく協働搬送ロボットの目的地推論

抄録

For natural human-robot collaboration, a collaborative delivery robot should infer the user’s desired goal position to transport instruments, where the user’s convenience and the surrounding environment are considered. However, it is difficult to describe all such user preferences and complex human-robot coexisting environments in advance. Therefore, we focus on the user’s corrective path of the robot position, and using this correction, the robot sequentially improves inference of the user’s desired goal position. We generate multiple position samples from the corrective path, and each position sample is weighted based on the implicit intention of the correction to learn both the desired and undesired positions. Furthermore, the inferences of desired and undesired positions are integrated. Consequently, the robot sequentially improves the goal inference in fewer trials. The effectiveness of the proposed method was evaluated using a simulator that simulates human-robot collaborative environments.