Visual Reinforcement Learning on DMControl VDCS Markov-temporal perturbations (test)
864Cartpole Swingup ScoreACO-MoE
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| ACO-MoETraining steps=1M2026.04 | 864 | 492 | 882 | 910 | 339 | 700 | 963 | 738 | 736 | |
| Q2Training steps=1M2026.04 | 683 | 379 | 23 | 310 | 165 | 309 | 443 | 57 | 296.1 | |
| SVEATraining steps=1M2026.04 | 649 | 961 | 60 | 69 | 41 | 221 | 837 | 279 | 389.6 | |
| SODATraining steps=1M2026.04 | 615 | 307 | 59 | 475 | 44 | 158 | 171 | 128 | 244.6 | |
| SimGRLTraining steps=1M2026.04 | 613 | 778 | 31 | 603 | 3 | 237 | 845 | 188 | 412.3 | |
| SGQNTraining steps=1M2026.04 | 354 | 948 | 287 | 5 | 6 | 166 | 792 | 229 | 348.4 | |
| FTRTraining steps=1M2026.04 | 342 | 545 | 105 | 659 | 109 | 4 | 532 | 232 | 316.1 | |
| DrQ-v2Training steps=1M2026.04 | 49 | 2 | 68 | 4 | 3 | 6 | 45 | 25 | 25.3 |