Long-Horizon Evolution in Hybrid Embodied Tasks on HTEWorld (test)
31RCBDWEM
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| WEMWorld-ego definition=Semantic-based view, Protocol=Autoregressive multi-turn rollout2026.05 | 31 | 87 | 57 | 54 | 52 | 89 | |
| Motion-basedWorld-ego definition=Motion-based view, Protocol=Autoregressive multi-turn rollout2026.05 | 29 | 87 | 53 | 54 | 51 | 89 | |
| Intention-basedWorld-ego definition=Intention-based view, Protocol=Autoregressive multi-turn rollout2026.05 | 28 | 86 | 53 | 52 | 50 | 88 | |
| PAN-styleProtocol=Autoregressive multi-turn rollout2026.05 | 27 | 86 | 49 | 50 | 46 | 88 | |
| Cosmos-14BProtocol=Autoregressive multi-turn rollout2026.05 | 26 | 83 | 51 | 48 | 49 | 85 | |
| Cosmos-2BProtocol=Autoregressive multi-turn rollout2026.05 | 24 | 83 | 50 | 47 | 48 | 86 | |
| WoW-7BProtocol=Autoregressive multi-turn rollout2026.05 | 23 | 83 | 49 | 45 | 47 | 85 |