Embodied Task Execution on VLABench Cross Category
61Primitive Success (PS)ERVLA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ERVLAmode=Zero-shot2026.06 | 61 | 47 | 66.4 | |
| Choice + No Knowledge TruncationAblation=Choice without KT, mode=Zero-shot2026.06 | 55.2 | 42.4 | 59.4 | |
| π0.5mode=Zero-shot2026.06 | 49.7 | 38.2 | 52 | |
| No Choice + Knowledge InsulationAblation=No choice and KI, mode=Zero-shot2026.06 | 47.6 | 36 | 53.8 | |
| No CoTAblation=No explicit chain-of-thought, mode=Zero-shot2026.06 | 44.8 | 34.8 | 51.6 | |
| No Choice (End-to-End)Ablation=End-to-End No Choice, mode=Zero-shot2026.06 | 42 | 33.6 | 49.6 | |
| ACoT-VLAmode=Zero-shot2026.06 | 38.9 | — | 54.1 | |
| π0-FASTmode=Zero-shot2026.06 | 38.2 | 31 | 47.8 | |
| π0mode=Zero-shot2026.06 | 33.6 | 21.2 | 44 | |
| X-VLAmode=Zero-shot2026.06 | 25.1 | — | — |