List your company
ResearchFormerly WizWand

Offline Reinforcement Learning on D4RL Adroit pen (cloned)

117.4Normalized ReturnMXQL
-7.29625.07757.4589.823Jan 30, 2023Aug 20, 2023Mar 9, 2024Sep 27, 2024Apr 17, 2025Nov 5, 2025May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MXQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.11117.4
QQLDomain=Adroit, Hyperparameter Tuning=consistent2025.11115.2
IQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.11114.1
XQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.11112.6
FACPolicy Type=Flow, Seeds=82026.02103.2
ReBRACModel paradigm=Model-free, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.12102.8
BCDomain=Adroit, Hyperparameter Tuning=individually tuned2025.1199.1
EPQPolicy Type=Gaussian, Seeds=82026.0291.8
ReBRACPolicy Type=Gaussian, Seeds=82026.0291.8
NEUBAYModel paradigm=Ours, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1291.3
IQLModel paradigm=Model-free, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1283.4
IQLPolicy Type=Gaussian, Seeds=82026.0277.2
SPAR-PROJLearning Paradigm=Ours2026.0576.2
DPPOsupervision=preference only, seeds=52023.0175.1
MoMoModel paradigm=Conservative model-based, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1274.1
FQLPolicy Type=Flow, Seeds=82026.0274
TD3+BCLearning Paradigm=Policy Gradient Guidance2026.0571.4
VIPOModel paradigm=Conservative model-based, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1271.1
MOBILEModel paradigm=Conservative model-based, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1269
EDACModel paradigm=Model-free, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1268.2
QCS-R2024.0266.5
Diff-QLLearning Paradigm=Policy Gradient Guidance2026.0566.2
IDQLPolicy Type=Diffusion, Seeds=82026.0264
IDQLLearning Paradigm=In-Support Learning2026.0563.6
IQLLearning Paradigm=In-Support Learning2026.0562.2
TD3+BCPolicy Type=Gaussian, Seeds=82026.0261.4
SRPOPolicy Type=Diffusion, Seeds=82026.0261
CACPolicy Type=Diffusion, Seeds=82026.0256
BCQLearning Paradigm=Policy Gradient Guidance2026.0555
LAPOLearning Paradigm=In-Support Learning2026.0555
MOPOModel paradigm=Conservative model-based, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1254.6
TD3+BCDomain=Adroit, Hyperparameter Tuning=individually tuned2025.1152.7
BCModel paradigm=Model-free, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1251.9
ARMORModel paradigm=Conservative model-based, pi_D=68.7, Number of seeds=6, Evaluation episodes=20, Evaluation step=final step2025.1251.4
IQLsupervision=task rewards, seeds=52023.0151.3
BaseLearning Paradigm=Ours2026.0550.4
DCMethod Category=RCSL, Evaluation=official codebase2024.0250
MCQPolicy Type=Gaussian, Seeds=82026.0249.4
AWACLearning Paradigm=In-Support Learning2026.0549.3
PLASLearning Paradigm=Policy Gradient Guidance2026.0549
EQLLearning Paradigm=In-Support Learning2026.0546.9
PT+IQLsupervision=preference only, seeds=52023.0142.9
CQLsupervision=task rewards, seeds=52023.0142.4
CQLLearning Paradigm=Policy Gradient Guidance2026.0540.3
CQLMethod Category=Value-Based2024.0239.2
CQLPolicy Type=Gaussian, Seeds=82026.0239.2
IQLMethod Category=Value-Based2024.0237.3
DTMethod Category=RCSL, Evaluation=official codebase2024.0228.7
PT+CQLsupervision=preference only, seeds=52023.0118.3
CQLDomain=Adroit, Hyperparameter Tuning=individually tuned2025.1114.7
SAC-RNDPolicy Type=Gaussian, Seeds=82026.022.5
SPAR-MLPLearning Paradigm=Ours2026.050.1
CQL-AWLearning Paradigm=In-Support Learning2026.05-2.5