arXiv cs.AI· Yukiya Horiba, Koshiro Aoki, Shunsuke Yasuki, Bum Jun Kim, Taiki Miyanishi·· 5 小时前AI 评分26
VLA 模型对抗性补丁防御的机制化方法：检测与抑制
Detect and Suppress: A Mechanistic Defense against Adversarial Patches in VLA Models
AI 导读
arXiv 论文用稀疏自编码器（SAE）对 Vision-Language-Action（VLA）模型内部表征做机制化分析，发现一个与对抗性补丁出现强相关的特征。该方法仅在线性探针检测到攻击时在推理阶段抑制该特征，无需微调模型，在 LIBERO-10 上的评估显示间歇性攻击下成功率提升；但持续施加同样的干预会显著降低策略性能，说明控制干预时机很关键。
正文 · 原文
Abstract:Adversarial patches can disrupt Vision-Language-Action (VLA) models by manipulating visual observations, leading to failures in robot control. However, it remains poorly understood which internal mechanisms underlie these failures and how targeted interventions can mitigate them. In this work, we mechanistically analyze VLA representations using a sparse autoencoder (SAE) and identify a feature whose activation strongly correlates with the presence of an adversarial patch. Based on this analysis, we suppress the identified feature at inference time only when a linear probe detects an attack. This intervention improves robustness without the cost of fine-tuning the VLA. We evaluate our method against VLA adversarial patch attacks on LIBERO-10. Conditional intervention improves success rate under intermittent attacks, whereas continuously applying the same intervention substantially degrades policy performance. These results show that attack-related internal representations can provide useful targets for VLA adversarial defense and that controlling when to intervene is important for limiting disruption to nominal policy behavior.
| Comments: | The Science of Physical AI Safety Workshop at CoRL 2026 |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.03498 [cs.RO] |
| (or arXiv:2610.03498v2 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.03498 arXiv-issued DOI via DataCite |
Submission history
From: Yukiya Horiba [view email]
[v1]
Fri, 2 Oct 2026 15:57:04 UTC (845 KB)
[v2]
Fri, 30 Oct 2026 13:17:07 UTC (845 KB)
来源：arXiv cs.AI · arxiv.org