Human-robot action teams: How robot proactivity affects teamwork

Abstract

Many researchers in human-AI and human-robot interaction are interested in robots supporting teams that collaborate under intense conditions, such as action teams. Action teams must make de-cisions quickly, under uncertainty, and under high workload. Here, a helpful autonomous system would be one that can act proactively, as people do not have time to engage in autonomy manage-ment. However, there is limited understanding of how proactive robot behaviors may impact team dynamics and human-robot collaboration, as prior work has primarily focused on either dyadic interactions or separately studied proactive dialogue and proactive actions. In this paper, we ex-plored this within the context of an escape room game, where action teams of three people and a physical robot (Stretch) collaboratively complete tasks. We varied Stretch's behaviors (proactive or passive) in each escape room to facilitate teamwork. Our results reveal insights into how teams can differ in their preferences for proactive and passive robot behaviors, which are affected by a robot's perceived helpfulness, capabilities, and a team's desire for a sense of control over the robot. We discuss several suggestions for designing future autonomous system teaming behaviors, which consider: the team's sense of control and resilience to robot errors, the robot's status in the team, and the robot's communication style. We also explore the ethical implications of how proactive robot help may mislead people during human-robot collaboration. These contributions lay the foundation for designing robots that can support action teams, specifically with regard to the nuances of designing proactive behaviors, and moderating user expectations of proactive robot teammates.