Offline Reinforcement Learning on Meta-World medium-replay
3,967BP->DC*CoWorld
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CoWorldSource Selection=Best-Source2023.05 | 3,967 | 3,623 | 4,521 | 4,570 | 4,845 | 3,889 | 4,241 | |
| CoWorldSource Selection=Multi-Source2023.05 | 3,864 | 3,573 | 4,507 | 4,460 | 4,678 | 3,626 | 4,094 | |
| DV2 FinetuneStrategy=Finetune2023.05 | 3,500 | 2,456 | 3,467 | 3,702 | 4,273 | 3,499 | 3,781 | |
| LOMPOStrategy=Offline2023.05 | 2,883 | 446 | 2,983 | 2,230 | 2,756 | 1,961 | 1,712 | |
| Offline DV2Strategy=Offline2023.05 | 2,143 | 3,142 | 3,921 | 278 | 3,899 | 3,002 | 2,730 | |
| CQLStrategy=Offline2023.05 | 1,984 | 867 | 683 | 988 | 577 | 462 | 927 | |
| CURLStrategy=Offline2023.05 | 1,972 | 51 | 281 | 986 | 366 | 189 | 641 | |
| DV2 Finetune + EWCStrategy=Finetune, Regularization=EWC2023.05 | 1,566 | 167 | 978 | 528 | 2,048 | 224 | 918 | |
| DrQ + BCStrategy=Offline2023.05 | 567 | 587 | 623 | 1,203 | 134 | 642 | 626 | |
| LOMPO FinetuneStrategy=Finetune2023.05 | 259 | 95 | 142 | 332 | 3,698 | 224 | 792 |