Goal-conditioned Reinforcement Learning on OGBench State-based
82PointMaze Medium Navigation Success RateQRL
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| QRLMethod Architecture=Non-Hierarchical2026.05 | 82 | 86 | 88 | 75 | 14 | 21 | 5 | 5 | 1 | 5 | 1 | 0 | 32 | |
| HIQLMethod Architecture=Hierarchical2026.05 | 79 | 58 | 96 | 91 | 65 | 89 | 49 | 15 | 6 | 38 | 12 | 7 | 50 | |
| DualGCIQLMethod Architecture=Goal Representation approach (Dual)2026.05 | 78 | 36 | 73 | 41 | 2 | 37 | 4 | 89 | 51 | 60 | 57 | 31 | 47 | |
| DualGCIVLMethod Architecture=Goal Representation approach (Dual)2026.05 | 76 | 46 | 75 | 28 | 0 | 29 | 3 | 89 | 60 | 72 | 5 | 23 | 42 | |
| Ms.PRMethod Architecture=Ms.PR2026.05 | 64 | 54 | 96 | 96 | 48 | 55 | 41 | 65 | 12 | 49 | 75 | 47 | 59 | |
| GCIVLMethod Architecture=Non-Hierarchical2026.05 | 63 | 45 | 72 | 16 | 0 | 24 | 2 | 53 | 36 | 42 | 6 | 13 | 31 | |
| GCIQLMethod Architecture=Non-Hierarchical2026.05 | 53 | 34 | 71 | 34 | 0 | 27 | 2 | 68 | 40 | 51 | 95 | 26 | 42 | |
| DualCRLMethod Architecture=Goal Representation approach (Dual)2026.05 | 33 | 39 | 93 | 87 | 21 | 57 | 18 | 60 | 24 | 44 | 6 | 2 | 41 | |
| CRLMethod Architecture=Non-Hierarchical2026.05 | 29 | 39 | 95 | 83 | 16 | 60 | 24 | 19 | 10 | 19 | 3 | 0 | 33 | |
| GCBCMethod Architecture=Non-Hierarchical2026.05 | 9 | 29 | 29 | 24 | 0 | 8 | 1 | 6 | 1 | 5 | 2 | 0 | 15 |