Agreeing to Interact: Understanding Interaction as Human-Robot Goal Conflicts

In human-robot interaction (HRI) people have studied user preferable robot actions in various social situations. The role of the robot is often designed, and situations are assumed that the robot will interact with the human. However, there are also situations where either the human or robot may not be willing to interact. In such situations, the human and robot are under a goal conflict and must first agree to begin an interaction. In this paper, we re-explore interaction beginnings and endings as a confliction and agreement between human and robot goals - the willingness of whether to interact or not. Through our discussion, we categorize conflict/agreement interactions into nine situations. Using a probabilistic analysis approach and 93 HRI recordings, we evaluate the different human behaviors in different interaction situations. We further question whether learning from typical existing HRI would benefit other scenarios when a robot has physical task capabilities. We conclude that the benefits of understanding different agreement situations would largely depend on a robot»s task capability as well as the human»s expectation toward these capabilities; however, conflict and agreement should not be neglected when applying interaction capability to physical-task-capable robots. Our research also suggests the probabilistic drawbacks of robot speech in situations where both the human and robot are not willing to interact.