Reinforcement Learning of Player Agents in RoboCup Soccer Simulation

摘要

Multiagent systems have emerged as an active sub field of artificial intelligence. Machine learning techniques have played a significant role by handling the inherent complexity of such systems. Robotic Soccer is a typical multiagent system, wherein the challenge is to develop and hone the skills of the agents that take part in the game. For an in-depth and sophisticated understanding of the game, soccer-playing agents must possess the capability to learn and acquire low-level skills. These skills can later be put together and used to emulate the expertise of experienced players. This paper describes the use of reinforcement learning, a machine learning technique, to acquire the base level skills of intercepting a moving ball. Results of simulation runs using the Robocup Soccer server have also been presented.