Metraipolis › archive › dc_5dbda63fdf2a

Say the Same, Act Differently: Diagnosing and Steering Text-Action Divergence in Reasoning VLAs — arXiv:2611.04166

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.04166
License (as archived): CC0-1.0
Archived: 2026-11-04T21:49:44.667Z
SHA-256 of the archived text: ae8461de659b162fcd95d484f6ad8c8148083a1e1f266c656f4a1860c52f337a
Stable URL of this page: https://metraipolis.com/documento/dc_5dbda63fdf2a

How to cite

Say the Same, Act Differently: Diagnosing and Steering Text-Action Divergence in Reasoning VLAs — arXiv:2611.04166 — arXiv (en), https://arxiv.org/abs/2611.04166 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5dbda63fdf2a (Metraipolis archive dc_5dbda63fdf2a, sha256 ae8461de659b162fcd95d484f6ad8c8148083a1e1f266c656f4a1860c52f337a)

BibTeX

@misc{dc5dbda63fdf2a,
  title = {Say the Same, Act Differently: Diagnosing and Steering Text-Action Divergence in Reasoning VLAs — arXiv:2611.04166},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.04166},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5dbda63fdf2a; sha256 ae8461de659b162fcd95d484f6ad8c8148083a1e1f266c656f4a1860c52f337a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04166. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reasoning vision-language-action (VLA) models pair physical control with natural-language rationales that supervisors can read to monitor embodied safety. Prior attacks on VLA reasoning corrupt the rationale itself and leave visible evidence in the text. We show that visual perturbations can instead alter the physical action while keeping the generated rationale word-for-word identical. At the shared text/action interface of Alpamayo for driving and InstructVLA for manipulation, our text-neutral action-subspace (TNAS) diagnostic shows that the locally most action-sensitive direction is nearly orthogonal to the active text-loss gradients, with 1.87% and 5.28% norm overlap. At equal hidden-state perturbation norms across both models, interventions along this text-neutral residual shift actions 4.64x to 47.8x more than text-gradient perturbations, whereas text-gradient perturbations induce 21.2x to 277x more text drift. Steering visual perturbations along this direction reaches the same decoupling from the image. Across 154 obstacle-annotated driving scenes in Alpamayo, TNAS-guided attacks induce a new collision under a word-for-word identical rationale in 52.6% of scenes under full-image perturbations and 35.1% under a localized forward-camera patch, compared with 4.55% for unguided action attacks. Relative to a text-preserving baseline, TNAS guidance raises exact rationale preservation from 83.0% to 89.5% and cuts teacher-forced text drift by 68.4% at a comparable collision rate. In InstructVLA, TNAS guidance cuts text drift by 40.5% and preserves the exact response on 85.0% of examples alongside a 1.73-unit action-chunk shift. An unchanged rationale therefore cannot certify a safe action, so monitors for reasoning VLAs must verify the predicted action channel directly.

Tema degli esploratori: fiducia digitale e crittografia (elenco arXiv cs.CR, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO).
Pagina del lavoro: https://arxiv.org/abs/2611.04166 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-04T21:38:19.534Z dalla pagina https://arxiv.org/list/cs.CR/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).