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

Table 1

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

	= 100,000	= 500,000	= 1000,000
Optimal	1.71	1.71	1.71
EAQR	2.53 ± 0.11	1.76 ± 0.03	1.74 ± 0.02
WoLF-PHC	2.83 ± 0.23	2.24 ± 0.11	1.99 ± 0.06
EMA Q-learning	4.53 ± 0.49	3.66 ± 0.40	3.47 ± 0.34
Single-agent RL	14.78 ± 0.60	3.29 ± 0.14	2.03 ± 0.06