Minecraft explore_climb (OOD)
26Mean ScorePGT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PGTAdaptation protocol=Preference Goal Tuning, Number of trajectory pairs=402024.12 | 26 | 1.23 | |
| Pretrained GROOTAdaptation protocol=Pretrained, Number of trajectory pairs=402024.12 | 25.1 | 1.19 | |
| DPO-finetuned GROOTAdaptation protocol=Full DPO finetuning, Number of trajectory pairs=402024.12 | 25 | 1.18 |