A General Method for Kinematic Retargeting: Adapting Poses between Humans and Robots

This paper presents a method for kinematic retargeting that is general to a broad class of kinematic chains. Kinematic retargeting is the adaptation of a pose or motion from one kinematic embodiment to another. Our method distinguishes itself in its ability to adapt poses to new robots with very little configuration by the user. We accomplish this by defining two general metrics for retargeting and minimizing a cost function which is the weighted sum of these two metrics. This allows the method to automatically adapt poses between source and target chains that have different link lengths and degrees of freedom. These capabilities address a specific problem in HumanRobot Interaction (HRI), where behaviors are often defined in a robot-specific manner. The ability to automatically adapt behaviors from humans to new robots, and from one robot to another, will facilitate experimental repeatability. Through simulation and experiments, we demonstrate that our method is effective in adapting poses across chains with different numbers of joints, and in adapting socially expressive gestures from a human to two very different robots.