Offline Reinforcement Learning on Maze2D medium v1
166.82Normalized ReturnInverter
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InverterK=16, #NN-passes=38, ms/NN-pass=1.62 ms, other=11.2 ms, sum=72.9ms, × slower=1.0×2026.05 | 166.82 | 1.2 | |
| Diffuser#NN-passes=256, ms/NN-pass=14.54 ms, other=14.5 ms, sum=3.74s, × slower=51.3×2026.05 | 130.07 | 22.7 | |
| ReBRAC#NN-passes=600, ms/NN-pass=0.21 ms†, sum=126.0ms†, × slower=1.7×2026.05 | 105.11 | 31.67 | |
| SAC-N#NN-passes=600, ms/NN-pass=1.80 ms, sum=1.08s, × slower=14.8×2026.05 | 88.61 | 18.72 | |
| CQL#NN-passes=600, ms/NN-pass=1.84 ms, sum=1.10s, × slower=15.1×2026.05 | 86.11 | 9.68 | |
| TD3+BC#NN-passes=600, ms/NN-pass=1.67 ms, sum=1.00s, × slower=13.8×2026.05 | 59.45 | 36.25 | |
| EDAC#NN-passes=600, ms/NN-pass=1.74 ms, sum=1.04s, × slower=14.3×2026.05 | 57.04 | 3.45 | |
| AWAC#NN-passes=600, ms/NN-pass=1.28 ms, sum=765.0ms, × slower=10.5×2026.05 | 52.88 | 55.12 | |
| PrivORL-jRL Algorithm=IQL, epsilon=102025.12 | 49.3 | — | |
| PrivORL-jRL Algorithm=TD3PLUSBC, epsilon=102025.12 | 38 | — | |
| PrivORL-jRL Algorithm=EDAC, epsilon=102025.12 | 35 | — | |
| IQL#NN-passes=600, ms/NN-pass=1.69 ms, sum=1.02s, × slower=14.0×2026.05 | 34.85 | 2.72 | |
| DP-TransformerRL Algorithm=IQL, epsilon=102025.12 | 32.8 | — | |
| DT#NN-passes=600, ms/NN-pass=3.57 ms, sum=2.14s, × slower=29.4×2026.05 | 31.71 | 26.33 | |
| PrivORL-jRL Algorithm=EDAC, epsilon=12025.12 | 31.5 | — | |
| PrivORL-jRL Algorithm=TD3PLUSBC, epsilon=12025.12 | 31.1 | — | |
| DP-TransformerRL Algorithm=TD3PLUSBC, epsilon=102025.12 | 26.6 | — | |
| PrivORL-jRL Algorithm=IQL, epsilon=12025.12 | 23.4 | — | |
| DP-TransformerRL Algorithm=EDAC, epsilon=102025.12 | 19 | — | |
| DP-TransformerRL Algorithm=IQL, epsilon=12025.12 | 18.1 | — | |
| PrivORL-j-URL Algorithm=EDAC, epsilon=102025.12 | 16.4 | — | |
| BC-10%#NN-passes=600, ms/NN-pass=1.25 ms, sum=749.8ms, × slower=10.3×2026.05 | 14.25 | 2.33 | |
| PrivORL-j-URL Algorithm=IQL, epsilon=102025.12 | 14.1 | — | |
| PrivORL-j-URL Algorithm=EDAC, epsilon=12025.12 | 13.5 | — | |
| PrivORL-j-URL Algorithm=TD3PLUSBC, epsilon=102025.12 | 10.8 | — | |
| DP-TransformerRL Algorithm=EDAC, epsilon=12025.12 | 10.2 | — | |
| DP-TransformerRL Algorithm=TD3PLUSBC, epsilon=12025.12 | 7.6 | — | |
| PrivORL-j-URL Algorithm=TD3PLUSBC, epsilon=12025.12 | 6.7 | — | |
| PrivORL-j-URL Algorithm=IQL, epsilon=12025.12 | 5.5 | — | |
| BC#NN-passes=600, ms/NN-pass=1.70 ms, sum=1.02s, × slower=14.0×2026.05 | 0.79 | 3.25 |