FreezeVLA: Action-Freezing Attacks against Vision-Language-Action Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918147173384192 |
|---|---|
| author | Wang, Xin Li, Jie Weng, Zejia Wang, Yixu Gao, Yifeng Pang, Tianyu Du, Chao Teng, Yan Wang, Yingchun Wu, Zuxuan Ma, Xingjun Jiang, Yu-Gang |
| author_facet | Wang, Xin Li, Jie Weng, Zejia Wang, Yixu Gao, Yifeng Pang, Tianyu Du, Chao Teng, Yan Wang, Yingchun Wu, Zuxuan Ma, Xingjun Jiang, Yu-Gang |
| contents | Vision-Language-Action (VLA) models are driving rapid progress in robotics by enabling agents to interpret multimodal inputs and execute complex, long-horizon tasks. However, their safety and robustness against adversarial attacks remain largely underexplored. In this work, we identify and formalize a critical adversarial vulnerability in which adversarial images can "freeze" VLA models and cause them to ignore subsequent instructions. This threat effectively disconnects the robot's digital mind from its physical actions, potentially inducing inaction during critical interventions. To systematically study this vulnerability, we propose FreezeVLA, a novel attack framework that generates and evaluates action-freezing attacks via min-max bi-level optimization. Experiments on three state-of-the-art VLA models and four robotic benchmarks show that FreezeVLA attains an average attack success rate of 76.2%, significantly outperforming existing methods. Moreover, adversarial images generated by FreezeVLA exhibit strong transferability, with a single image reliably inducing paralysis across diverse language prompts. Our findings expose a critical safety risk in VLA models and highlight the urgent need for robust defense mechanisms. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_19870 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | FreezeVLA: Action-Freezing Attacks against Vision-Language-Action Models Wang, Xin Li, Jie Weng, Zejia Wang, Yixu Gao, Yifeng Pang, Tianyu Du, Chao Teng, Yan Wang, Yingchun Wu, Zuxuan Ma, Xingjun Jiang, Yu-Gang Computer Vision and Pattern Recognition Vision-Language-Action (VLA) models are driving rapid progress in robotics by enabling agents to interpret multimodal inputs and execute complex, long-horizon tasks. However, their safety and robustness against adversarial attacks remain largely underexplored. In this work, we identify and formalize a critical adversarial vulnerability in which adversarial images can "freeze" VLA models and cause them to ignore subsequent instructions. This threat effectively disconnects the robot's digital mind from its physical actions, potentially inducing inaction during critical interventions. To systematically study this vulnerability, we propose FreezeVLA, a novel attack framework that generates and evaluates action-freezing attacks via min-max bi-level optimization. Experiments on three state-of-the-art VLA models and four robotic benchmarks show that FreezeVLA attains an average attack success rate of 76.2%, significantly outperforming existing methods. Moreover, adversarial images generated by FreezeVLA exhibit strong transferability, with a single image reliably inducing paralysis across diverse language prompts. Our findings expose a critical safety risk in VLA models and highlight the urgent need for robust defense mechanisms. |
| title | FreezeVLA: Action-Freezing Attacks against Vision-Language-Action Models |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2509.19870 |