EAQR: A Multiagent Q-Learning Algorithm for Coordination of Multiple Agents

Table 3

Maximal steps for 4-agent/12-vertex box-pushing (evaluation episodes = 50,000).

	= 100,000	= 500,000	= 1000,000
EAQR	2.77	1.85	1.81
WoLF-PHC	3.45	2.55	2.18
EMA Q-learning	5.90	4.66	4.61
Single-agent RL	15.89	3.66	2.20