Matteo Cercola, Katie Z Luo, Simone Formentin, Marco Pavone · 2026 · arXiv
DOI: 10.48550/arXiv.2610.28184 · open-access PDF
Model pengemudi VLA (Vision-Language-Action) dapat melihat dan memahami bahaya-bahaya yang penting bagi keselamatan namun tidak dapat menerjemahkan bukti tersebut menjadi tindakan-tindakan yang aman. Kami menggambarkan kegagalan ini sebagai kesenjangan yang mengikat persepsi-tindakan: informasi bahaya tersedia di dalam model tetapi tidak memegang kendali yang cukup atas generasi trajektori. Menggunakan simulasi simulasi dan pengulangan pengemudi yang tercatat, kami menunjukkan bahwa bahaya-bahaya di jalan sebenarnya dikodekan dalam representasi visual, tetapi rencana longitudinal didominasi oleh sejarah kecepatan ego dan hanya merespon lemah terhadap kedekatan bahaya atau t
Machine translation (Meta M2M100); the English original follows.
Reasoning-augmented vision-language-action (VLA) driving models can perceive and reason about safety-critical hazards yet fail to translate that evidence into safe actions. We characterize this failure as a perception-action binding gap: hazard information is available inside the model but exerts insufficient control over trajectory generation. Using matched simulated counterfactuals and logged driving replay, we show that in-path hazards are actually encoded in the visual representation, but longitudinal plans are dominated by ego-speed history and respond only weakly to hazard proximity or t
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.28184, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange