ROBOCUP : WHEN SOCCER MEETS COMPUTER

Abstract

RoboCup is an attempt to promote AI and robotics research by providing a common problem domain for evaluating various theories, algorithms, and agent architectures. Here, we focus on building a team of software agents to play in a simulated soccer field. Teamwork is imperative to win a game. Other multi-agent implementations often rely heavily on fixed plans which degenerate quickly when things go wrong. We propose an explicit teamwork model-Shared Mission--that allows agents to behave coherently in the dynamic and real-time environment. Obviating the need for the usual strong assumptions found in several models, we explore ways that allow our model to scale up to larger teams. Communication is a natured tool for facilitating the process of teamwork. We identity and propose solutions to address problems arising from the inherently noisy, low-bandwidth, and unreliable communication medium. Currently, the primitive skills provided are too raw to allow agents to perform high-level reasoning and planning. We augment such skills so that agents can pass, dribble, and intercept moving balls. We also examine ways to manage a renewable resource, the agent's stamina.