Decision-Theoretic Planning for Playing Table Soccer

Table soccer (also called “foosball”) is much simpler than real soccer. Nevertheless, one faces the same challenges as in all other robotics domains. Sensors are noisy, actions must be selected under time pressure and the execution of actions is often less than perfect. One approach to solve the action selection problem in such a context is decision-theoretic planning, i.e., identifying the action that gives the maximum expected utility. In this paper we present a decision-theoretic planning system suited for controlling the behavior of a table soccer robot. The system employs forward-simulation for estimating the expected utility of alternative action sequences. As demonstrated in experiments, this system outperforms a purely reactive approach in simulation. However, this superiority of the approach did not extend to the the real soccer table.