BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912387440836608 |
|---|---|
| author | Zhou, Xueyang Tie, Guiyao Zhang, Guowen Wang, Hechang Zhou, Pan Sun, Lichao |
| author_facet | Zhou, Xueyang Tie, Guiyao Zhang, Guowen Wang, Hechang Zhou, Pan Sun, Lichao |
| contents | Vision-Language-Action (VLA) models have advanced robotic control by enabling end-to-end decision-making directly from multimodal inputs. However, their tightly coupled architectures expose novel security vulnerabilities. Unlike traditional adversarial perturbations, backdoor attacks represent a stealthier, persistent, and practically significant threat-particularly under the emerging Training-as-a-Service paradigm-but remain largely unexplored in the context of VLA models. To address this gap, we propose BadVLA, a backdoor attack method based on Objective-Decoupled Optimization, which for the first time exposes the backdoor vulnerabilities of VLA models. Specifically, it consists of a two-stage process: (1) explicit feature-space separation to isolate trigger representations from benign inputs, and (2) conditional control deviations that activate only in the presence of the trigger, while preserving clean-task performance. Empirical results on multiple VLA benchmarks demonstrate that BadVLA consistently achieves near-100% attack success rates with minimal impact on clean task accuracy. Further analyses confirm its robustness against common input perturbations, task transfers, and model fine-tuning, underscoring critical security vulnerabilities in current VLA deployments. Our work offers the first systematic investigation of backdoor vulnerabilities in VLA models, highlighting an urgent need for secure and trustworthy embodied model design practices. We have released the project page at https://badvla-project.github.io/. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_16640 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization Zhou, Xueyang Tie, Guiyao Zhang, Guowen Wang, Hechang Zhou, Pan Sun, Lichao Cryptography and Security Artificial Intelligence 68T07 I.2.6; I.2.9 Vision-Language-Action (VLA) models have advanced robotic control by enabling end-to-end decision-making directly from multimodal inputs. However, their tightly coupled architectures expose novel security vulnerabilities. Unlike traditional adversarial perturbations, backdoor attacks represent a stealthier, persistent, and practically significant threat-particularly under the emerging Training-as-a-Service paradigm-but remain largely unexplored in the context of VLA models. To address this gap, we propose BadVLA, a backdoor attack method based on Objective-Decoupled Optimization, which for the first time exposes the backdoor vulnerabilities of VLA models. Specifically, it consists of a two-stage process: (1) explicit feature-space separation to isolate trigger representations from benign inputs, and (2) conditional control deviations that activate only in the presence of the trigger, while preserving clean-task performance. Empirical results on multiple VLA benchmarks demonstrate that BadVLA consistently achieves near-100% attack success rates with minimal impact on clean task accuracy. Further analyses confirm its robustness against common input perturbations, task transfers, and model fine-tuning, underscoring critical security vulnerabilities in current VLA deployments. Our work offers the first systematic investigation of backdoor vulnerabilities in VLA models, highlighting an urgent need for secure and trustworthy embodied model design practices. We have released the project page at https://badvla-project.github.io/. |
| title | BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization |
| topic | Cryptography and Security Artificial Intelligence 68T07 I.2.6; I.2.9 |
| url | https://arxiv.org/abs/2505.16640 |