Using Reinforcement Learning Techniques to Select the Best Action in Setplays with Multiple Possibilities in Robocup Soccer Simulation Teams

Abstract

Set plays are predefined collaborative coordinate actions that players from any sport can use to gain advantage over its adversaries. Recently, a complete framework for creation and execution of this kind of coordinate behavior by teams composed of multiple independent agents was launched as free software (the Set play Framework). In this paper, an approach based on Reinforcement Learning(RL) is proposed, that allows the use of experience to devise the better course of action in set plays with multiple choices. Simulations results show that the proposed approach allows a team of simulated agents to improve its performance against a known adversary team, achieving better results than previously proposed approaches using RL.