Physical interaction as communication: Learning robot objectives online from human corrections

When a robot performs task next to human, physical interaction is inevitable: the human might push, pull, twist, or guide robot. The state of art treats these interactions as disturbances that should reject avoid. At best, robots respond safely while interacts; but after lets go, simply return their original behavior. We recognize human–robot (pHRI) often intentional: intervenes on purpose because not doing correctly. In this article, we argue when pHRI intentional it also informative: can leverage learn how complete rest its current even person go. formalize dynamical system, where has in mind an objective function they want optimize, does get direct access parameters objective: are internal human. Within our proposed framework become observations about true objective. introduce approximations from and real-time. all corrections perfect: users interact with noisily, so improve efficiency learning by reducing unintended learning. Finally, conduct simulations user studies robotic manipulator compare approach art. Our results indicate leads better performance improved satisfaction.