查看文章

Human-robot interaction (HRI) in the real world often bars robots from key information on which their decisions may hinge. Existing HRI planning methods either neglect the robot's ability to learn and adapt to human intents at runtime, leading to overly conservative robot motion, or optimistically assume (cooperative) human behaviors, potentially resulting in loss of safety. In order to simultaneously achieve safety and efficiency for HRI in uncertain, non-lab environments, my work leverages principles from game theory and safety analysis and proposes a novel HRI planning framework that jointly reasons about the physical states and the robot's internal representation of the human uncertainty in closed loop, leading to scalable computation of safe robot policies in real time.