Argumentation Accelerated Reinforcement Learning for RoboCup Keepaway-Takeaway

Multi-Agent Learning (MAL) is a complex problem, especially in real-time systems where both cooperative and competitive learning are involved. We study this problem in the RoboCup Soccer KeepawayTakeaway game and propose Argumentation Accelerated Reinforcement Learning (AARL) for this game. AARL incorporates heuristics, represented by arguments in Value-Based Argumentation, into Reinforcement Learning (RL) by using Heuristically Accelerated RL techniques. We empirically study for a specific setting of the Keepaway-Takeaway game the suitability of AARL, in comparison with standard RL and hand-coded strategies, to meet the challenges of MAL.