Reinforcement Learning on DMC Cheetah
338.8Run ScoreOurs-FT (FB-PbRL)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Ours-FT (FB-PbRL)Reward Specification=Preferences, Fine-tuning=true, Data Budget=2,000 preference pairs2026.05 | 338.8 | 335.4 | 924.6 | 982.8 | 645.4 | |
| FB-PbRL (Ours-FT)fine-tuning=true2026.05 | 249.5 | 333.9 | 920.3 | 983.3 | 621.7 | |
| PSMReward Specification=Ground-truth reward, Protocol=Zero-Shot RFRL, Data Budget=10,000 transitions2026.05 | 244.4 | 296.4 | 984.2 | 979 | 626 | |
| RLDPReward Specification=Ground-truth reward, Protocol=Zero-Shot RFRL, Data Budget=10,000 transitions2026.05 | 236.3 | 322.1 | 895.3 | 984.8 | 609.6 | |
| FB-PbRL (Ours-T)test-time CPTS=true2026.05 | 207.6 | 160.7 | 665.9 | 660.4 | 344.7 | |
| FBReward Specification=Ground-truth reward, Protocol=Zero-Shot RFRL, Data Budget=10,000 transitions2026.05 | 169.1 | 153.3 | 509.7 | 710.5 | 385.6 | |
| OPRL2026.05 | 115.1 | 99.9 | 445.1 | 445.5 | 276.4 | |
| LIRE2026.05 | 109.1 | 135.6 | 419.3 | 589.7 | 313.4 | |
| CLARIFY2026.05 | 106.7 | 93.6 | 446.5 | 439.3 | 271.5 | |
| OPPO2026.05 | 97.9 | 113.1 | 258.4 | 533.2 | 200.9 | |
| LaplaceReward Specification=Ground-truth reward, Protocol=Zero-Shot RFRL, Data Budget=10,000 transitions2026.05 | 96.3 | 106.4 | 409.2 | 654.3 | 316.5 | |
| DPPO2026.05 | 71.2 | 74.4 | 285.1 | 378.2 | 202.3 | |
| HILPReward Specification=Ground-truth reward, Protocol=Zero-Shot RFRL, Data Budget=10,000 transitions2026.05 | 68.2 | 38 | 318.3 | 349.6 | 193.5 |