Off-Policy General Value Functions to Represent Dynamic Role Assignments in RoboCup 3D Soccer Simulation

摘要

Collecting and maintaining accurate world knowledge in a dynamic, complex,adversarial, and stochastic environment such as the RoboCup 3D SoccerSimulation is a challenging task. Knowledge should be learned in real-time withtime constraints. We use recently introduced Off-Policy Gradient Descentalgorithms within Reinforcement Learning that illustrate learnable knowledgerepresentations for dynamic role assignments. The results show that the agentshave learned competitive policies against the top teams from the RoboCup 2012competitions for three vs three, five vs five, and seven vs seven agents. Wehave explicitly used subsets of agents to identify the dynamics and thesemantics for which the agents learn to maximize their performance measures,and to gather knowledge about different objectives, so that all agentsparticipate effectively and efficiently within the group.