Moritz Bühler, Jürgen Adamy, Thomas Weisswange, "Theory of Mind Based Assistive Communication in Complex Human Robot Cooperation", IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022) - Workshop on Human Theory of Machines and Machine Theory of Mind for Human-Agent Teams (TOM4HAT), 2022.

When cooperating with a human, a robot should not only care about its environment and task but also develop an understanding of the partner’s reasoning. To support its human partner in complex tasks, the robot should share information. However simply communicating everything it knows might annoy and distract humans as not all information might be relevant or novel in a given situation. To decide when and what type of information should be communicated, one needs to take into account knowledge and situation awareness of the partner. We previously proposed the concept of Theory of Mind-based Communication which selects information sharing actions based on evaluation of relevance and an estimation of human beliefs. Here, we integrate this into a communication assistant in a cooperative human-robot setting and evaluate performance benefits. Human belief is estimated based on task progression and gaze positions. Using this belief estimate, the system predicts the most likely next actions and evaluates the impact on future rewards. If a simulated belief update from a robot communication action will lead to a future plan with higher expected reward, while including an explicit cost of communication, the system will choose to assist the human with information. We designed a task that is challenging for the human and generates situations where humans could potentially profit from additional information communicated by the robot. Naive human participants performed the task with the robot to generate data which was then used to evaluate the influence of the human-centric communication concept on a range of performance measures. Compared to the condition without information exchange, assisted participants could recover from unawareness much earlier. The approach respects the costs of communication and balances interruptions better than other approaches. Belief inference enables an assistance concept that does not patronize but enables humans to make good decision themselves.