It loses when the opponent managesto instigate danger. Such reasoning is similar to decisionmaking in strategy games like chess where the players notonly choose good moves for themselves but also moves thatprevent their opponent from winning. In order to maximally protect humans and robots, therobot should take into account all possible situations andplan for the worst. The goal of this work isto optimize robot motion such that it maintains maximalsafety regardless of any change in the environment includingthe most dangerous choices made by humans who share therobots workspace. This includes positive interactions such ascollaborative tasks and the absence of negative interactionssuch as human-robot collisions.
INTRODUCTIONIn order to deploy safe and flexible robots for serviceand automation, robots must act safely in close contactwith humans. We generateoptimized blocks and apply game theoretic tools to choose thebest action for the defender in the presence of an intelligentadversary. Our approach enables a simulated 7-DOFrobot arm to block known attacks in any sequence. Experimentally, we apply our model to simulatedrobot sword defense. By representing the domain as a MarkovGame, we enable the robot to not only react to the human butalso to construct an infinite horizon optimal policy of actionsand responses. In these cases, therobot can select optimal motions in response to human actionsand maximize safety. We focus on the casewhere human motion can be predicted. Our computational solutions are complemen-tary to passive and compliant hardware. Tobias Kunz, Peter Kingston, Mike Stilman and Magnus EgerstedtĪbstract We introduce and experimentally validate a novelalgorithmic model for physical human-robot interaction withhybrid dynamics. Dynamic Chess: Strategic Planning for Robot Motion