Learning In RoboCup Keepaway Using Evolutionary Algorithms

Abstract

Manually coordinating the efforts of many autonomous agents can be a formidable challenge, so the idea of using machine learning techniques (such as evolutionary algorithms) to produce such coordination is attractive. We present a study using evolutionary algorithms to train autonomous agents to play the game of keepaway — a sub-problem of the RoboCup robotic soccer league. Our results exceed those previously produced using other methods.