Robotic soccer is a very good example of a PTS domain: teams can coordinate before the game, at half-time, and at other break points, but communication is limited during play . Introduced by Mackworth , the robotic soccer domain has been gaining popularity as an AI and robotics test-bed, with systems having been recently developed both in simulation and with real robots . Robotic soccer is a multi-agent domain with heterogeneous (at least two types of agents even if each team is homogeneous), communicating agents. For a survey of research issues arising in this type of domain, see . The research presented in this article was first developed in simulation and some of it has also been successfully used on our real robot team .
The soccer server [25, 1], which serves as the substrate simulator for the research reported in this article, 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 modeled; 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 soccer server was successfully used as the basis for the RoboCup-97 simulator competition in which 29 teams participated .
As presented in  The soccer server models a communication environment appropriate in a time-pressured, crowded environment. All 22 agents (11 on each team) use a single, unreliable communication channel. When one agent speaks, agents on both teams can hear the message immediately along with the (relative) direction from which it came. The speaker is not inherently known. Agents have a limited communication range, hearing only messages spoken from within a certain distance. They also have a limited communication capacity, hearing a maximum of 1 message every 200 msec (actions are possible every 100 msec, so if all other agents are speaking as fast as they can, only 1 of every 42 messages will be heard). Thus communication is extremely unreliable. We use version 3 of this simulator for the research reported in this article.
Table 2: Characteristics of the soccer server communication model.
This section details the full implementations of both main contributions of this article: the flexible teamwork structure and the novel communication paradigm. Both implementations are unified within the same robotic soccer system: the CMUnited-97 simulator team .