notesum.ai

Published at November 11

Imitation from Diverse Behaviors: Wasserstein Quality Diversity Imitation Learning with Single-Step Archive Exploration

cs.LG

cs.AI

Released Date: November 11, 2024

Authors: Xingrui Yu¹, Zhenglin Wan², David Mark Bossens¹, Yueming Lyu¹, Qing Guo¹, Ivor W. Tsang¹

Aff.: ¹Centre for Frontier AI Research (CFAR), A*STAR, Singapore; ²Centre for Frontier AI Research (CFAR), A*STAR, Singapore; School of Data Science, The Chinese University of Hong Kong, Shenzhen, China

Arxiv: http://arxiv.org/abs/2411.06965v1

PPGA-trueReward	$6.75\times 10^{6}$	94.08	8,942	2,871	$3.64\times 10^{6}$	77.04	5,588	1,891	$5.71\times 10^{6}$	49.96	9,691	4,570
	HalfCheetah				Walker2d				Humanoid
	QD-Score	Cov(%)	Best	Avg	QD-Score	Cov(%)	Best	Avg	QD-Score	Cov(%)	Best	Avg
mCWAE-WGAIL-Bonus	$\mathbf{3.80\times 10^{6}}$	98.28	4,553	1,547	$\mathbf{3.07\times 10^{6}}$	87.39	4,142	1,407	$\mathbf{6.40\times 10^{6}}$	75.05	6,875	3,447
WAE-WGAIL-Bonus	$2.66\times 10^{7}$	76.89	4,788	1,246	$2.55\times 10^{6}$	73.93	3,310	1,381	$4.97\times 10^{6}$	67.27	8,620	3,09
mCWAE-WGAL	$3.78\times 10^{7}$	87.76	5,728	1,623	$2.37\times 10^{6}$	71.61	3,611	1,326	$3.54\times 10^{6}$	66.96	8,485	2,114
WAE-WGAIL	$3.74\times 10^{6}$	93.77	5,463	1,615	$2.11\times 10^{6}$	73.20	3,327	1,148	$3.98\times 10^{6}$	68.08	8,553	2,359