Error Recovery in Wearable Robotic Co-Grasping: The Role of Human-Led Correction

Erin Y ChangWilson O TorresHannah S Stuart^*

• University of California, Berkeley, Berkeley, United States

Trust in automated systems influences the use and disuse of new technologies. Although recent advances in robotics have improved wearable devices designed to assist in grasping, perfectly reliable systems have yet to be achieved. In this work, we introduce a new strategy for wearable devices called Co-Grasping, where both body power and robotics can contribute to grasping, but the user controls the allocation of the human and robot roles. Our implementation of a Co-Grasping device successfully allows the human operator to intervene using body power during simulated robot errors, in order to aid in error recovery and continue performing grasping tasks without drops. Here, we also show that the presence of recoverable errors lowers trust perception and increases physical engagement behaviors. However, when the robot becomes reliable once again, trust rebounds and most behavioral metrics return to baseline as well. These results indicate that trust in faulty automation can be repaired and that enabling users to assume control over system actuation in response to such faults can prevent errors from negatively affecting overall device function. Facilitating human-led dynamic changes in human and robot role allocation through this Co-Grasping device lays a promising foundation for unique human-robot interactions that promote high performance and where trust can recover quickly, despite existing challenges in developing perfect automated systems.