Text-to-humanoid Motion Generation on AMASS (test)
0FIDGround Truth
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ground Truth2025.11 | 0 | 61 | 3.804 | 8.238 | |
| Humanoid-LLAcomponents=LLM-generated tokens + vocabulary-directed controller, fine-tuned=humanoid feedback2025.11 | 2.626 | 44.7 | 4.911 | 7.122 | |
| LangWBCarchitecture=CVAE-based2025.11 | 6.171 | 32 | 5.587 | 6.031 | |
| UH-1backbone=decoder-only transformer2025.11 | 8.682 | 29.5 | 5.896 | 6.749 | |
| MDM+Retargetretargeting=kinematic2025.11 | 11.759 | 26.2 | 6.599 | 6.419 | |
| OmniH2Oimitation policy=true2025.11 | 17.159 | 22.2 | 8.021 | 5.868 |