Interactive adaptation of Hand-over Motion by a Robot Partner for Comfort of receiving

Recently various types of robot partners have been developed to conduct human-friendly physical support to people. In this study, we proposed a system that enables human beings to adapt interactively the hand-over motion of the robot for each human posture. Basically, it is difficult for a robot partner to plan such a human-friendly hand-over motions beforehand. Therefore, we focus on the degree of human comfortability on hand-over motion when the human receive an object. To evaluate the current hand-over motion, a human applies a force to the force sensor at the gripper when receiving an object. We apply fuzzy inference to adjust the position and direction of the robot gripper in hand-over motions according to the magnitude of applied force. And the position and direction at the next hand-over motion are determined. However, the adjustment is strongly dependent on the human postures. Therefore, we apply a neural network to classify the types of human postures. Finally, we show several experimental results of the interactive adaptation on hand-over motions, and discuss the effectiveness of the proposed method.