Continuous Control on Hopper (Perturbation Robustness Metrics)
962.5Performance (p_act=0.1)Self-Paced
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Self-PacedTraining steps=1 million, Evaluation episodes=1002025.11 | 962.5 | 665.6 | 469.4 | 372.6 | 291.5 | 830.8 | 359.4 | 258 | 184.2 | 159.9 | 1,280.3 | 806.1 | 653.5 | 540.2 | 473.6 | |
| SPACETraining steps=1 million, Evaluation episodes=1002025.11 | 795.5 | 589.8 | 428.1 | 343.6 | 276 | 556.7 | 223.9 | 157.7 | 120.2 | 107.1 | 1,012.3 | 726.3 | 566 | 480.2 | 447 | |
| PPOTraining steps=1 million, Evaluation episodes=1002025.11 | 789.5 | 553.9 | 401.4 | 308.9 | 248.3 | 465.3 | 186.8 | 136.4 | 112.4 | 95.2 | 1,082.6 | 755.6 | 562.2 | 496.2 | 415.7 | |
| LinearTraining steps=1 million, Evaluation episodes=1002025.11 | 716.5 | 479 | 346.1 | 237.4 | 181.6 | 156.5 | 96.4 | 93.9 | 93.2 | 94.1 | 894.1 | 702.3 | 585 | 485.6 | 412.8 | |
| ACCELTraining steps=1 million, Evaluation episodes=1002025.11 | 451.2 | 358.9 | 302 | 260.7 | 229 | 275.5 | 180 | 139.8 | 110.6 | 88.8 | 484 | 403.9 | 363.8 | 329.2 | 319.3 | |
| FixedTraining steps=1 million, Evaluation episodes=1002025.11 | 352.8 | 237.7 | 135.6 | 81.1 | 64.5 | 39.4 | 30.2 | 25.5 | 21.2 | 20.3 | 528.6 | 470.6 | 393.6 | 272.8 | 266.5 | |
| Dom. Rand.Training steps=1 million, Evaluation episodes=1002025.11 | 342.8 | 219.8 | 165.9 | 143 | 131 | 128.8 | 95.3 | 77.7 | 61.5 | 48 | 540 | 490.5 | 382.9 | 337 | 312.8 |