Continuous Control on DeepMind Control 3M steps
422Acrobot Swingup ScoreDrQv2
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DrQv2Observation type=pixel, Random seeds=5, Test rollouts=100, Evaluation window (training steps)=150K, Training steps=3M2022.10 | 422 | 503 | 873 | 932 | 913 | 239 | 494 | 905 | 228 | 954 | 944 | 616 | 668.6 | 744.66 | |
| SSPGObservation type=pixel, Random seeds=5, Test rollouts=100, Evaluation window (training steps)=150K, Training steps=3M2022.10 | 371 | 837 | 888 | 974 | 945 | 344 | 760 | 888 | 218 | 982 | 974 | 738 | 743.32 | 862.73 | |
| DrQObservation type=pixel, Random seeds=5, Test rollouts=100, Evaluation window (training steps)=150K, Training steps=3M2022.10 | 43 | 319 | 832 | 218 | 100 | 290 | 115 | 124 | 9 | 600 | 426 | 549 | 301.97 | 253.97 | |
| SACObservation type=pixel, Random seeds=5, Test rollouts=100, Evaluation window (training steps)=150K, Training steps=3M2022.10 | 10 | 174 | 7 | 211 | 100 | 0 | 56 | 52 | 2 | 100 | 15 | 26 | 62.77 | 39.08 | |
| CURLObservation type=pixel, Random seeds=5, Test rollouts=100, Evaluation window (training steps)=150K, Training steps=3M2022.10 | 6 | 530 | 589 | 310 | 172 | 223 | 170 | 154 | 7 | 645 | 651 | 449 | 325.41 | 266.34 |