Evolving Emergent Team Strategies in Robotic Soccer using Enhanced Cultural Algorithms

Cultural Algorithm has been applied to efficiently solve different types of problems from different engineering fields. More specifically, Cultural Algorithm has been used in different multiagent systems to model the emergent cooperative behavior of agents. The goal of this work is to use a modified and enhanced Cultural‐Algorithm‐based evolutionary training, to generate emergent behaviors, and hence effective plays, for a controlled team of robotic soccer players. This should serve to substitute implementing some AI controllers for the teams in a soccer videogame. The system is targeted to be used as an efficient tutorial for beginners on the use of Cultural Algorithms for the coordination of a group of agents in a complex dynamic environment. This modified version of Cultural Algorithm was able to successfully learn different types of plays, including active and passive characters, within a reasonable number of generations. Experimental results show how this configured version of Cultural Algorithms can be used to effectively compete with other types of case‐based plays and learning strategies. Moreover, it shows how such controlled behavior can result in a noticeable improvement on the values of metrics that were used to measure the efficiency of the robots, in terms of their score, control time, and offensive and defensive plans.