Sensor-based manipulation planning as a game with nature

This paper uplores a game-theoretic approach to automatic planning of sensor·based robotic manipulation programs. To win the game, the robot must provably attain a specified task swc. The robot moves by cboosing 1 control signal. and nature moves by cboosing a sensor signal. Planning is accomplished by searching lhe game O"ee. In some task domllnJ. lhe llppf'OII:h provides a slrllght!orwan1 method or reasoning about uncenalnty. non-detenninistic actions, and imperfect sensors. We cScmonstrate lhe approach in two dift'erent wlc domains: orienting an object using an instnunent.ed tilling tray: and orienting and and gnuping 111 object with an insuumen:ed panllel-jaw gripper.