A data-driven method for determining natural human-robot motion mappings in teleoperation

Though many aspects of teleoperation systems have been fine-tuned through research, the mapping between the operator's movement and the robot's movement is almost always pre-defined as a simple scaling and offset. We believe that implementing nontraditional data-driven motion mappings has the potential to further improve the usability of teleoperation platforms, making it easier for a human to remotely complete challenging tasks. This paper presents a new paradigm for determining data-driven human-robot motion mappings for teleoperation: the human operator mimics the target robot as it autonomously moves its arm through a variety of trajectories. We analyze the resulting motion data to find the human's chosen mapping combined with the systematic errors the human made when relying on proprioception to execute these arm movements. We report results from a study in which nine human subjects repeatedly mimicked the planar arm motions of a simulated PR2. We propose three nontraditional motion mappings (similarity, affine, and variable similarity), and we fit each of the proposed models to these data sets, verifying within and across trials and subjects. As hypothesized, the newly proposed mappings are found to outperform the traditional motion mapping model.