Multi-robot coordination in the robot soccer environment

Abstract

MAPS (Multi-Agent Planning system) is a system developed for multi-agent coordination developed for use in the robot soccer domain. The system operates without explicit communication of strategy between agents, relying upon observation of team members to produce meaningful coordinated behaviour. Each robot is coordinated with the others by MAPS choosing goal actions that most benefit the team. The choice of goal action is based on the robot's observations of the rest of the team and their behaviours. The various observations are modelled by superposition of potential fields, where the fields represent the influence of the team, other robots, and components of the environment. This paper outlines the system as it functioned in the UQ RoboRoos entry in the real robot small-size league of RoboCup '98. Testing and competition results show that this coordination model keeps the robots in good field position in both attack and defence. The improved field position is shown to provide better opportunities for improving ball position and shooting at goal, as well as preventing collisions between robots. 1