Extrapolative Generalization on Reacher-hard Unseen View
321.9Mean RewardDreamSAC
Evaluation results
| Method | Links | |
|---|---|---|
| DreamSACunsupervised pre-training steps=2M, task adaptation steps=500K2026.03 | 321.9 | |
| DreamerV3+RNDunsupervised pre-training steps=2M, task adaptation steps=500K2026.03 | 313.97 | |
| DreamerV3+Policyunsupervised pre-training steps=2M, task adaptation steps=500K2026.03 | 265.33 | |
| DreamSAC0-shot=true, unsupervised pre-training steps=2M2026.03 | 149.64 | |
| DreamerV3+RND0-shot=true, unsupervised pre-training steps=2M2026.03 | 79.31 |