notesum.ai

Published at November 29

RMIO: A Model-Based MARL Framework for Scenarios with Observation Loss in Some Agents

cs.MA

Released Date: November 29, 2024

Authors: Shi Zifeng¹, Liu Meiqin¹, Zhang Senlin¹, Zheng Ronghao¹, Dong Shanling¹

Aff.: ¹Zhejiang University

Arxiv: http://arxiv.org/pdf/2411.19639v1

Refer to caption

SMAC
Maps	MES	REIS	RMIO	MAG	MAMBA	MAPPO	QMIX
2s_vs_1sc	300	50k	94(5)	90(8)	91(15)	9(12)	0(0)
8m		75k	94(3)	87(9)	77(10)	32(15)	63(9)
3s_vs_3z		75k	96(3)	93(6)	90(10)	12(16)	0(0)
2s3z		150k	98(1)	81(11)	78(13)	46(23)	23(17)
3s_vs_4z		200k	96(1)	71(13)	63(37)	0(0)	0(0)
1c3s5z		200k	95(3)	78(8)	70(9)	52(17)	46(19)
3s_vs_5z		500k	95(2)	65(15)	65(8)	58(21)	0(0)
corridor		500k	66(19)	55(25)	46(16)	0(0)	0(0)
MaMuJoCo
Scenarios	MES	REIS	RMIO	MAG	MAMBA	MAPPO	FACMAC
4 agent Swimmer	300	200k	98(3)	94(4)	91(6)	52(7)	32(6)
10 agent Swimmer		200k	102(2)	96(3)	67(29)	-20(7)	-38(6)
2 agent HalfCheetah		1m	2042(48)	1814(58)	1759(64)	823(98)	531(102)
6 agent HalfCheetah		1m	2098(42)	1957(47)	1836(58)	946(104)	620(94)