Embodied Task Planning on VirtualHome (unseen domains)
80.16Success RateTMoW
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TMoWBackbone=Llama-3.2-1B2026.01 | 80.16 | 13.2 | |
| SayCanPaySay Backbone=Llama-3.2-3B, Pay Backbone=Llama-3.2-1B2026.01 | 49.53 | 18.55 | |
| LLM+FTBackbone=Llama-3.2-1B, Training=Fine-tuned2026.01 | 44.24 | 21 | |
| FLAREBackbone=Llama-3.2-3B2026.01 | 40.07 | 22.57 | |
| LLM-PlannerBackbone=Llama-3.2-3B, Adaptation=Few-shot2026.01 | 36.05 | 22.93 | |
| ZSPBackbone=Llama-3.2-3B, Adaptation=Zero-shot2026.01 | 7.32 | 28.22 |