Assistance for Telepresence Using Online Grasp Planning

This paper presents a user study evaluating teleoperated grasping performance and perceived workload of the human operator in a shared autonomy setup when working with different assistance modes and hand kinematics. The hands of a humanoid robot are operated using two approaches: direct mapping of human finger motions (telemanipulation), and "open/close" commands in combination with online grasp planning (shared autonomy). Human finger movements are measured with a data glove in both approaches. Grasp planning for the shared autonomy mode is based on the online calculation of reachable independent contact regions. In this approach, two visual assistance modes were tested: one indicating graspability in a binary manner (possible vs. impossible) and another one showing the potential contact regions for the fingertips. To analyze the influence of the hand kinematics on grasping performance and workload, two hands with different thumb positions are compared. The study shows that shared autonomy significantly decreases the task completion time and increases the grasp robustness compared to the direct mapping approach. The effect is more evident for a hand with optimized kinematics. The results reveal that choosing the appropriate control and assistance mode has a significant influence in telepresence performance.