VLA推論は忠実か?自動運転モデルにおける因果連鎖の安全性の検証
Is VLA Reasoning Faithful? Probing Safety of Chain-of-Causation in Autonomous Driving Models
自動運転のVLAモデルが出力する推論の忠実性を初めて体系的に調査し、推論が実際のシーンと一致しない割合が高いことや、歩行者見落とし、軌道の脆弱性などを明らかにした。
詳しい要約
1. どんなもの?
2. 先行研究と比べてどこがすごい?
3. 技術・手法の肝は?
4. どうやって有効だと検証した?
5. 議論はある?
6. 次に読むべき論文は?
※ AIが要旨から生成した要約です。正確性は原文をご確認ください。
著者: Nicanor Mayumu, Xiaoheng Deng, Patrick Mukala
分類: cs.AI, cs.CV, cs.RO
原文アブストラクト
We present the first systematic study of faithfulness in Vision-Language-Action (VLA) driving models, analyzing 300 Alpamayo-R1-10B inferences across 100 diverse PhysicalAI-AV scenarios. Our main finding is that output natural-language rationales with trajectories may be significantly unfaithful: (i) overall reasoning fidelity is only 42.5%, with Chain-of-Causation matching scene reality less than half the time; (ii) 94 missed pedestrians in one-third of pedestrian-relevant scenes; (iii) 97.7% trajectory fragility under mild visual perturbations; and (iv) only 48.3% mean reasoning-action consistency, with 53.3% of inferences exhibiting low consistency, including 37.9% of stop-claimed cases where the model continues instead. We formalize faithfulness information-theoretically, define entity and action fidelity with verification criteria, and outline a four-component safety architecture aligned with these results.
