Task Generalization on Meta-World ML-10 (test)
97.5Success RateMZ+Recon
Evaluation Results
| Method | Links | |
|---|---|---|
| MZ+ReconEvaluation protocol=finetune, Input source=pixels, Environment steps=10M2021.11 | 97.5 | |
| MuZeroEvaluation protocol=finetune, Input source=pixels, Environment steps=10M2021.11 | 94.1 | |
| Q-LearningEvaluation protocol=finetune, Input source=pixels, Environment steps=10M2021.11 | 80.1 | |
| RL2Evaluation protocol=few-shot, Input source=states, Environment steps=400M2021.11 | 35.8 | |
| MAMLEvaluation protocol=few-shot, Input source=states, Environment steps=400M2021.11 | 31.6 | |
| MuZeroEvaluation protocol=zero-shot, Input source=pixels, Environment steps=50M2021.11 | 26.5 | |
| MZ+ReconEvaluation protocol=zero-shot, Input source=pixels, Environment steps=50M2021.11 | 25 | |
| Q-LearningEvaluation protocol=zero-shot, Input source=pixels, Environment steps=50M2021.11 | 6.8 |