Strategic play for a pool-playing robot

Abstract

Playing pool is not a trivial task for an autonomous robotic system: In order to excel, the tight coupling of accurate perception, planning and highly dynamic while precise manipulation is required. To investigate scientific challenges arising from this tight coupling requirement an anthropomorphic robotic system has been developed, which is capable of playing pool. Focusing in this paper on the planning part, a well-calibrated pool simulator is developed in order to predict the outcome of possible strokes. An optimization based framework applying a discounted return term is presented for suggesting an optimized next stroke by planning several strokes ahead. Different to existing planners so far, our planner considers the player's and opponent's skill. Therefore, strategic considerations (e.g. safety shots) can now be evaluated depending on the outcome of the next strokes for both player and opponent.