Function approximation based multi-agent reinforcement learning

The paper presents two new multi-agent based domain independent coordination mechanisms for reinforcement learning. The first mechanism allows agents to learn coordination information from state transitions and the second one from the observed reward distribution. In this way, the latter mechanism tends to increase region-wide joint rewards. The selected experimented domain is Adversarial Food-Collecting World (AFCW), which can be configured both as single and multi-agent environments. Experimental results show the effectiveness of these mechanisms.