Offline Reinforcement Learning on OGBench Visual
92.7Visual Cube Single Task 1 Success RateGSFlow (Ours)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GSFlow (Ours)Policy Category=Flow, Seeds=32026.03 | 92.7 | 42 | 100 | 31 | 88.67 | 70.9 | |
| ReBRACPolicy Category=Gaussian, Seeds=32026.03 | 83 | 4 | 98 | 26 | 88 | 59.8 | |
| FQLPolicy Category=Flow, Seeds=32026.03 | 81 | 21 | 98 | 33 | 94 | 65.4 | |
| IQLPolicy Category=Gaussian, Seeds=32026.03 | 70 | 34 | 97 | 0 | 7 | 41.6 | |
| FBRACPolicy Category=Flow, Seeds=32026.03 | 55 | 6 | 46 | 0 | 7 | 22.8 | |
| InFOM2025.06 | 52.1 | 11.2 | 72.4 | 6 | — | — | |
| IFQLPolicy Category=Flow, Seeds=32026.03 | 49 | 8 | 86 | 8 | 100 | 50.2 | |
| IQLModel=FB2025.06 | 31 | 1.3 | 74.7 | 0 | — | — | |
| ReBRACEncoder=DINO2025.06 | 15.3 | 5 | 26 | 0 | — | — | |
| TD InfoNCEConditioning=IS2025.06 | 12 | 5 | 9 | 0 | — | — | |
| ReBRAC2025.06 | 10.6 | 0 | 32 | 0 | — | — |