Robot Inference of Human States: Performance and Transparency in Physical Collaboration

To flexibly collaborate towards a shared goal in human-robot interaction (HRI), a robot must appropriately respond to changes in their human partner. The robot can realize this flexibility by responding to certain inputs, or by inferring some aspect of their collaborator and using this to modify robot behavior-approaches which reflect design viewpoints of robots as tools and collaborators, respectively. Independent of this design viewpoint, the robot's response to a change in collaborator state must also be designed. In this regard, HRI approaches can be distinguished according to the scope of their design objectives: whether the design goal depends on the behavior of the individual agents or the coupled team. This chapter synthesizes work on physical HRI, largely in manufacturing tasks, according to the design viewpoint and scope of objective used. HRI is posed as the coupling of two dynamic systems; a framework which allows a unified presentation of the various design approaches and, within which, common concepts in HRI can be posed (intent, authority, information flow). Special attention is paid to predictability at various stages of the design and deployment process: whether the designer can predict team performance, whether the human can predict robot behavior, and to what degree the human behavior can be modelled or learned.