Reasoning on Human Beliefs and Decisions and Integrating their Anticipation in a Human-Aware Robot Task Planner

We address the task planning problem for cognitive and interactive robots collaborating with humans to achieve a shared task or assisting human in a task. Two main aspects will be discussed: (1) the ability to reason about Theory of Mind and potential divergence of beliefs between the Human and the robot and to plan corrective actions and communications if needed; and (2) the elaboration of a concurrent and compliant joint action model based on social and collaborative signals. This model captures subtle possible agents' coordination and the human's inherent uncontrollability. We illustrate how this model is used to explore relevant courses of action and guide our planning approach. The result is a behavioral policy capturing the best robot actions to perform to be congruent and compliant with any online human's decision and action, including being passive. The policy also aims to best satisfy an estimation of human preferences.