Detect and Suppress: A Mechanistic Defense against Adversarial Patches in VLA Models
Detect and Suppress: A Mechanistic Defense against Adversarial Patches in VLA Models
This research paper introduces a defense mechanism for Vision-Language-Action (VLA) models against adversarial patches. By using a sparse autoencoder to identify internal features that correlate with attacks, the authors propose suppressing these features only when an attack is detected via a linear probe. Evaluated on the LIBERO-10 benchmark, this conditional intervention improves success rates under intermittent attacks while avoiding performance degradation associated with continuous suppression, demonstrating that targeted control of internal representations can enhance VLA robustness without fine-tuning.
The authors present a mechanistic analysis of Vision-Language-Action (VLA) models using sparse autoencoders to identify features correlated with adversarial patches, proposing a conditional suppression strategy that improves robustness on the LIBERO-10 benchmark without requiring fine-tuning.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
(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 s
Open source S3
uccess 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.
Open source S4
Detect and Suppress: A Mechanistic Defense against Adversarial Patches in VLA Models
Open source S1
arXiv:2610.03498v2 Announce Type: replace 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
Open source S2
Source：arXiv Robotics — research abstracts · arxiv.org