List your company
ResearchFormerly WizWand

Offline Reinforcement Learning on D4RL Adroit hammer-human

860Normalized ScoreFAC
-44.8190.1425659.9Nov 15, 2025Dec 1, 2025Dec 17, 2025Jan 2, 2026Jan 18, 2026Feb 3, 2026Feb 20, 2026
Updated 6mo ago

Evaluation results

MethodLinks
FACPolicy Type=Flow, Seeds=82026.02860
CQLPolicy Type=Gaussian, Seeds=82026.02440
EPQPolicy Type=Gaussian, Seeds=82026.02390
IQLPolicy Type=Gaussian, Seeds=82026.02250
IDQLPolicy Type=Diffusion, Seeds=82026.02200
CACPolicy Type=Diffusion, Seeds=82026.02200
SRPOPolicy Type=Diffusion, Seeds=82026.02100
FQLPolicy Type=Flow, Seeds=82026.02100
TD3+BCPolicy Type=Gaussian, Seeds=82026.0240
MCQPolicy Type=Gaussian, Seeds=82026.0230
ReBRACPolicy Type=Gaussian, Seeds=82026.0220
MXQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.1114.7
BCDomain=Adroit, Hyperparameter Tuning=individually tuned2025.1112.6
QQLDomain=Adroit, Hyperparameter Tuning=consistent2025.118.4
XQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.117.3
IQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.116.9
BCModel paradigm=Model-free, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.123
TD3+BCDomain=Adroit, Hyperparameter Tuning=individually tuned2025.112.4
ARMORModel paradigm=Conservative model-based, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.121.9
IQLModel paradigm=Model-free, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.121.8
MoMoModel paradigm=Conservative model-based, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.121.7
VIPOModel paradigm=Conservative model-based, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.121.1
EDACModel paradigm=Model-free, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.120.8
MOBILEModel paradigm=Conservative model-based, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.120.4
MOPOModel paradigm=Conservative model-based, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.120.3
CQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.110.3
ReBRACModel paradigm=Model-free, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.120.2
NEUBAYModel paradigm=Ours, pi_D=1.5, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.120
SAC-RNDPolicy Type=Gaussian, Seeds=82026.02-10