The Impacts of Unreliable Autonomy in Human-Robot Collaboration on Shared and Supervisory Control for Remote Manipulation

This work compared human-robot shared and supervisory control of remote robots for dexterous manipulation, and examined how the reliability of robot autonomy affects human operator performance, workload, and preference for robot assistance. Specifically, we implemented two human-robot collaboration (HRC) paradigms for remote manipulation: (1) shared control, where humans controlled gross manipulation and the robot autonomy controlled precise manipulation actions, and (2) supervisory control, where the robot autonomy controlled both gross and precise manipulation actions but relied on humans to detect and correct errors. We conducted two user studies: one to compare the effectiveness of the two HRC paradigms when assistive autonomy is reliable, and the other to examine the impact of error type and frequency on tasks and human operators in the two HRC paradigms when assistive autonomy is unreliable. Our results show that: (1) the interface with a higher level of reliable autonomy yields significantly better performance, lower workload, and higher user preference but lower engagement, and (2) the frequency and type of the error have significant impacts on the task performance and human workload but only partially affects the operator's preference and usage of autonomy.