Multi Agent Reinforcement Learning for Gridworld Soccer Leadingpass

Abstract

Soccer robotics is an emerging field that combines artificial intelligence and mobile robotics with the popular sport of soccer. Robotic soccer agents need to cooperate to complete tasks or subtasks, one way is by learning to coordinate their action. Leadingpass is considered as a task that had to be performed successfully by the team, or opponent could intercept the ball that leads the team to lose the game. This paper describes how Reinforcement Learning (RL) methods are applied to the learning scenario, that the learning agents cooperatively complete the leadingpass task in the Gridworld soccer environment. Not only RL algorithms for single agent case, but also for multi agent case. © Springer-Verlag Berlin Heidelberg 2013.