Failure Is an Option: How the Severity of Robot Errors Affects Human-Robot Interaction

Abstract

Just as humans are imperfect, even the best of robots will eventually fail at performing a task. The likelihood of failure increases as robots expand their roles in our lives. Although failure is a common problem in robotics and human-robot interaction (HRI), there has been little research investigating people's tolerance to said failures, especially when there is a risk of property damage and bodily harm. Safety is an important concern for human-robot interaction, and robot designers need to understand how people calibrate their levels of trust and adapt their behavior around robots that could expose them, and property, to physical harm.

To explore this issue, we performed an experiment where people were exposed to failure in a study with actual personal and property risk. Participants observed a Baxter robot while it performed a grocery packing task, and were given opportunities to react to and assist the robot in multiple failure cases. The study revealed important factors that influence trust, perception of safety, and whether participants would assist the robot after witnessing failure. Some of these findings were that the severity and recency of failures are among the most influential factors that influence human reports of trust in a robot. We also observed lower ratings of trust in the robot from female participants relative to male participants. While the majority of the participants assisted the robot when it failed, they were more likely to assist when the participants had not observed other failures prior to assisting the robot.

By understanding how people respond to robot failure and aspects of robot behavior that influence their trust, better understanding and design can be incorporated into robots. This should increase human comfort levels and willingness to interact and work with robots.