New attack method targets Transformer vulnerabilities in autonomous driving systems

Researchers have developed a new gray-box attack framework called Adversarial Flow Matching (AFM) that targets vulnerabilities in Transformer modules used by end-to-end autonomous driving systems. AFM can generate visually imperceptible adversarial examples in a single step by manipulating the generative latent space and a neural average velocity field. Experiments show AFM effectively degrades the performance of both Vision-Language-Action (VLA) and modular autonomous driving agents while maintaining high visual imperceptibility and demonstrating robust cross-model transferability. AI

影响 Introduces a novel attack method that could impact the safety and robustness of autonomous driving systems relying on Transformer architectures.

排序理由 This is a research paper detailing a novel attack framework for autonomous driving systems. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

• Adversarial Flow Matching
• arXiv
• Autonomous Driving
• Transformer
• Vision-Language-Action

• 论文
• 安全

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →