As described in [5], Robotic Soccer is an exciting AI domain for many reasons. The fast-paced nature of the domain necessitates real-time sensing coupled with quick behaving and decision making. Furthermore, the behaviors and decision-making processes can range from the most simple reactive behaviors, such as moving directly towards the ball, to arbitrarily complex reasoning procedures that take into account the actions and perceived strategies of teammates and opponents. Opportunities, and indeed demands, for innovative and novel techniques abound.
Robotic Soccer systems have been recently developed both in simulation [6, 9, 12, 14] and with real robots [1, 4, 10, 11, 15, 13]. While robotic systems are difficult, expensive, and time-consuming to use, they provide a certain degree of realism that is never possible in simulation. On the other hand, simulators allow researchers to isolate key issues, implement complex behaviors, and run many trials in a short amount of time. While much of the past research has used Machine Learning in constrained situations, nobody has yet developed a full behavior based on learning techniques that can be used successfully in a game situation.
The Soccer Server [7], which serves as the substrate system for the research reported in this paper, captures enough real-world complexities to be a very challenging domain. This simulator is realistic in many ways: (i) the players' vision is limited; (ii) the players can communicate by posting to a blackboard that is visible (but not necessarily intelligible) to all players; (iii) each player is controlled by a separate process; (iv) each team has 11 members; (v) players have limited stamina; (vi) actuators and sensors are noisy; (vii) dynamics and kinematics are modelled; and (viii) play occurs in real time: the agents must react to their sensory inputs at roughly the same speed as human or robotic soccer players. The simulator, acting as a server, provides a domain and supports users who wish to build their own agents (clients).