Metraipolis › archive › dc_e266acdf0f41
Does Adversarial Training Improve Generalization in Multi-View VLAs? Revealing and Mitigating View Collapse — arXiv:2609.33707
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.33707
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T20:57:06.521Z
- SHA-256 of the archived text
74bc0d6b1c95ae11c684903fc4e53da347292f85c567750124ca0d11e774d2c2- Stable URL of this page
https://metraipolis.com/documento/dc_e266acdf0f41
How to cite
Does Adversarial Training Improve Generalization in Multi-View VLAs? Revealing and Mitigating View Collapse — arXiv:2609.33707 — arXiv (en), https://arxiv.org/abs/2609.33707 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e266acdf0f41 (Metraipolis archive dc_e266acdf0f41, sha256 74bc0d6b1c95ae11c684903fc4e53da347292f85c567750124ca0d11e774d2c2)
BibTeX
@misc{dce266acdf0f41,
title = {Does Adversarial Training Improve Generalization in Multi-View VLAs? Revealing and Mitigating View Collapse — arXiv:2609.33707},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.33707},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e266acdf0f41; sha256 74bc0d6b1c95ae11c684903fc4e53da347292f85c567750124ca0d11e774d2c2},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.33707. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Vision-language-action (VLA) models adapt pretrained vision-language models (VLMs) for closed-loop robot control, transferring their perceptual and semantic capabilities to action prediction. Despite strong in-distribution performance, however, VLAs often degrade under deployment shifts. Adversarial training (AT) offers a model-adaptive approach to robustness without explicitly anticipating individual shifts, but its effect on natural distribution-shift generalization in multi-view VLAs remains unclear. We study this question using a multi-view VLA directly adapted from a pretrained VLM and evaluate generalization across seven LIBERO-Plus shift axes. Direct AT substantially improves Camera Viewpoint and Sensor Noise, the two shifts affecting only the third-person view, yet produces mixed or negative effects on other shifts. Controlled view interventions reveal a surprising failure mode that we term view collapse: Direct AT can shift cross-view reliance so strongly that the policy becomes dominated by the wrist view. This exposes a \textit{robustness shortcut}: apparent robustness to a shifted view can arise from reduced use of that view rather than more robust perception of it. This motivates a distinction between robust perception, extracting reliable information under within-view shifts, and robust fusion, adapting reliance across views according to their reliability. To reduce fixed view reliance, we use a simple View Swap intervention and then re-evaluate AT. With View Swap, AT further improves Camera Viewpoint, Sensor Noise, and Robot Initial State, while its effects remain mixed on other shifts. Our results show that multi-view robustness requires separating improved perception from changes in cross-view reliance, and that AT provides selective rather than generic distribution-shift benefits.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Robotics (cs.RO).
Pagina del lavoro: https://arxiv.org/abs/2609.33707 — 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 2026-10-07T20:29:01.350Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).