Embodied Instruction Following on LangR
71Average Success RateVEGAS
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| VEGASBackbone=Qwen-2.5-VL-3B-Instruct, Policy Type=w/ CoT, Verification Strategy=Finetuned Verifier2026.05 | 71 | 99 | 52 | 92 | 62 | 82 | 97 | 34 | 48 | |
| w/ CoTBackbone=Qwen-2.5-VL-3B-Instruct, Policy Type=w/ CoT2026.05 | 65 | 98 | 50 | 85 | 59 | 64 | 97 | 25 | 42 | |
| w/ CoT + ZS VerifierBackbone=Qwen-2.5-VL-3B-Instruct, Policy Type=w/ CoT, Verification Strategy=Zero-shot Verifier2026.05 | 64 | 98 | 50 | 85 | 48 | 65 | 97 | 30 | 40 | |
| SemLangBackbone=LLaVA-1.5-7B2026.05 | 58 | 92 | 46 | 66 | 31 | 80 | 97 | 2 | 46 | |
| No-CoTBackbone=Qwen-2.5-VL-3B-Instruct, Policy Type=No-CoT2026.05 | 58 | 93 | 39 | 72 | 48 | 68 | 97 | 17 | 28 | |
| LLaRPBackbone=LLaMa-7B2026.05 | 46 | 92 | 34 | 32 | 26 | 47 | 95 | 0 | 39 |