Robot Collaboration and Model Reliance Based on Its Trust in Human-Robot Interaction

Safety and performance are essential for fruitful teams consisting of a human and an autonomous robot. The collaboration in such a team requires that both parties are able to anticipate and understand each others' behavior. However, as both involved agents act autonomously, this form of collaboration has many sources of uncertainty, under which many decisions have to be made by both agents. In the case of interdependent set of actions, this also makes both agents vulnerable to each other's behavior. Depending on how critical the collaboration is, failing might significantly jeopardize safety and performance of the team. In interaction forms that contain uncertainty and vulnerability, trust plays an important role in making decisions, and consequently, in the outcome of the interaction. This applies for both agents, meaning that not only human trust is important but also robot trust. Having an adequate trust level can lead to safer and more natural collaboration. In this paper, we develop a robot trust model that allows the robot to reason about past interactions and use them to update its current belief about the human intentions. Using this model, the robot decides on how much to rely on the models it has and subsequently, trades-off between safety and performance. For evaluation, we conduct a human-subject experiment with 22 participants. The results show that including a sense of robotic trust in the decision-making of the robot enhances the interaction objectively and subjectively.