Offline Reinforcement Learning on D4RL (Medium/Expert/Replay MuJoCo Tasks)
90.8Return (HalfCheetah, Medium-Expert)Flowformer
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Flowformer2023.05 | 90.8 | 109.9 | 108 | 42.2 | 66.9 | 71.7 | 34.7 | 75.5 | 62 | 73.5 | |
| cosFormer2023.05 | 85.5 | 98.1 | 100.5 | 42.1 | 59.8 | 71.4 | 32.8 | 59.3 | 60.5 | 67.8 | |
| Performer2023.05 | 85.1 | 93.5 | 72.6 | 42.1 | 59.7 | 63.3 | 31.7 | 64.6 | 61.3 | 63.8 | |
| Decision Transformer2023.05 | 83.8 | 104 | 107.7 | 42.4 | 64.2 | 70.6 | 34.6 | 79.7 | 62.9 | 72.2 | |
| Reformer2023.05 | 81.5 | 104.2 | 71.4 | 42.2 | 59.9 | 65.8 | 33.6 | 66.1 | 50.1 | 63.9 | |
| Linear Transformer2023.05 | 78.2 | 107.2 | 67.2 | 42.3 | 58.7 | 57.9 | 32.1 | 74.3 | 62.1 | 64.4 | |
| Primal.+DT2023.05 | 77.8 | 111.5 | 108.9 | 43 | 74.5 | 77.9 | 38.9 | 88.5 | 76.8 | 77.5 |