Dynamic role assignment using general value functions

Collecting and maintaining accurate world knowledge in a dynamic, complex, competitive, and stochastic environment such as RoboCup 3D Soccer Simulation is a challenging task. Knowledge should be learned in real-time with time constraints. We use recently introduced Off-Policy Gradient Descent algorithms in Reinforcement Learning that illustrate learnable knowledge representations for dynamic role assignments. The results show that the agents have learned competitive policies against the top teams from RoboCup 2012 for three vs three, five vs five, and seven vs seven agents. We have explicitly used subset of agents to identify the dynamics and the semantics for which the agents learn to maximize their performance measures, and to gather knowledge about different objectives so that all agents participate effectively within the team.