Offline Reinforcement Learning on D4RL Medium-Replay HalfCheetah
95.8Normalized ScoreAdaptive Policy Selection and Fine-Tuning
Evaluation results
| Method | Links | |
|---|---|---|
| Adaptive Policy Selection and Fine-TuningPhase=Online, Online Interaction Budget=160 K2026.05 | 95.8 | |
| BestPhase=Offline, Online Interaction Budget=160 K2026.05 | 84.7 | |
| OEPhase=Offline, Online Interaction Budget=160 K2026.05 | 84.7 | |
| PSPOAveraged across=4 random seeds2026.05 | 78.4 | |
| RAMBOMethod Category=Model-free2025.12 | 77.6 | |
| VIPOMethod Category=Conservative model-based2025.12 | 77.2 | |
| SUMOMethod Category=Conservative model-based2025.12 | 76.2 | |
| FTPhase=Online, Online Interaction Budget=160 K2026.05 | 75.4 | |
| RRPINumber of random seeds=42026.03 | 74.4 | |
| MoMoMethod Category=Conservative model-based2025.12 | 72.9 | |
| MOPOMethod Category=Conservative model-based2025.12 | 72.1 | |
| NEUBAYMethod Category=Ours2025.12 | 72.1 | |
| MOBILEMethod Category=Conservative model-based2025.12 | 71.7 | |
| PMDBNumber of random seeds=42026.03 | 71.7 | |
| PMDBMethod Type=Model-based2023.10 | 71.7 | |
| PMDB2026.05 | 71.7 | |
| Count-MORLMethod Type=Model-based2023.10 | 71.5 | |
| CROPMethod Type=Model-based2023.10 | 70.4 | |
| MAPLEMethod Category=Bayesian-inspired2025.12 | 69.5 | |
| RAMBOMethod Type=Model-based2023.10 | 68.9 | |
| RAMBONumber of random seeds=42026.03 | 68.7 | |
| RAMBO2026.05 | 68.7 | |
| OPEPhase=Offline, Online Interaction Budget=160 K2026.05 | 68.4 | |
| ADMPOMethod Category=Conservative model-based2025.12 | 67.6 | |
| AMGNumber of random seeds=42026.03 | 67.6 | |
| ADM2026.05 | 67.6 | |
| MoDAPMethod Category=Bayesian-inspired2025.12 | 67.3 | |
| CBOPMethod Category=Bayesian-inspired2025.12 | 66.4 | |
| LEQMethod Category=Conservative model-based2025.12 | 65.5 | |
| APE-VMethod Category=Conservative model-based2025.12 | 64.6 | |
| EPQNumber of random seeds=42026.03 | 62 | |
| EPQ2026.05 | 62 | |
| EDACMethod Category=Model-free2025.12 | 61.3 | |
| EDACMethod Type=Model-free2023.10 | 61.3 | |
| COMBOMethod Category=Model-free2025.12 | 55.1 | |
| COMBO2025.12 | 55.1 | |
| COMBOMethod Type=Model-based2023.10 | 55.1 | |
| DMGNumber of random seeds=42026.03 | 51.4 | |
| DMG2026.05 | 51.4 | |
| ATACMethod Type=Model-free2023.10 | 48 | |
| CQL2023.06 | 45.5 | |
| CQLModel Category=Value-based2026.02 | 45.5 | |
| CQL2023.06 | 45.5 | |
| CQLMethod Category=Model-free2025.12 | 45.3 | |
| CQLNumber of random seeds=42026.03 | 45.3 | |
| CQL2026.05 | 45.3 | |
| IDQL2025.12 | 45.1 | |
| TD3+BCModel Category=Value-based2026.02 | 44.6 | |
| IQLAccess=True reward functions2024.04 | 44.5 | |
| MOReLNumber of random seeds=42026.03 | 44.5 | |
| MOReL2026.05 | 44.5 | |
| CABI+TD3-BCMethod Type=Model-based2023.10 | 44.4 | |
| IQL2023.06 | 44.2 | |
| IQLMethod Type=Model-free2023.10 | 44.2 | |
| Planner2025.12 | 44 | |
| MPDiffuser+Rankranking=true2025.12 | 43.5 | |
| MPDiffuserranking=false2025.12 | 43.4 | |
| PTAccess=Scripted preferences2024.04 | 42.3 | |
| Diffuser2023.06 | 42.2 | |
| Diffuser2025.12 | 42.2 | |
| TT2023.06 | 41.9 | |
| TT2023.06 | 41.9 | |
| OPRLAccess=Scripted preferences2024.04 | 41.5 | |
| Decision Stacks2023.06 | 41.1 | |
| D-MPC2025.12 | 41.1 | |
| DCModel Category=SSM-based2026.02 | 41.1 | |
| DMMModel Category=SSM-based2026.02 | 41.1 | |
| TAP2023.06 | 40.8 | |
| DMambasource=Lv et al., 20242023.06 | 40.8 | |
| CAMPAccess=Scripted preferences2024.04 | 40.5 | |
| MOREL2023.06 | 40.2 | |
| OPPOAccess=Scripted preferences, Implementation=Reported2024.04 | 39.8 | |
| DMModel Category=SSM-based2026.02 | 39.8 | |
| DMambasource=Ota, 20242023.06 | 39.6 | |
| DDImplementation=reproduced2023.06 | 39.4 | |
| DDImplementation=Original2023.06 | 39.3 | |
| Decision Diffuser2025.12 | 39.3 | |
| DTModel Category=Transformer-based2026.02 | 39.1 | |
| MambaDM2023.06 | 39.1 | |
| RATE2023.06 | 39 | |
| EDTModel Category=Transformer-based2026.02 | 37.8 | |
| BC2023.06 | 36.6 | |
| DT2023.06 | 36.6 | |
| BC2025.12 | 36.6 | |
| DT2025.12 | 36.6 | |
| DT2023.06 | 36.6 | |
| BCAccess=Scripted preferences2024.04 | 35.7 | |
| QLDTModel Category=Transformer-based2026.02 | 35.6 | |
| OPPOAccess=Scripted preferences, Implementation=Re-implementation2024.04 | 33.6 | |
| DARA2025.12 | 28.9 | |
| DROCO2025.12 | 27.9 | |
| BOSA2025.12 | 27.8 | |
| IQL*variant=tuned/reproduced2025.12 | 26.1 | |
| IGDF2025.12 | 24.2 | |
| CQL*variant=tuned/reproduced2025.12 | 21.3 | |
| OTDF2025.12 | 19.7 | |
| DS4Model Category=SSM-based2026.02 | 15.2 |