Extrapolative Generalization on Cheetah-run Unseen Friction
120.23Mean RewardDreamSAC
Evaluation results
| Method | Links | |
|---|---|---|
| DreamSACunsupervised pre-training steps=2M, task adaptation steps=500K2026.03 | 120.23 | |
| DreamerV3+Policyunsupervised pre-training steps=2M, task adaptation steps=500K2026.03 | 118.79 | |
| DreamerV3+RNDunsupervised pre-training steps=2M, task adaptation steps=500K2026.03 | 97.43 | |
| DreamSAC0-shot=true, unsupervised pre-training steps=2M2026.03 | 27.53 | |
| DreamerV3+RND0-shot=true, unsupervised pre-training steps=2M2026.03 | 9.42 |