Research map / Adversarial robustness and security / Attacks on text and graphs
When Robots Obey the Patch: Universal Transferable Patch Attacks on Vision-Language-Action Models
Hui Lu, Yi Yu, Yiming Yang, Chenyu Yi, Qixin Zhang, Bingquan Shen, Alex C. Kot, Xudong Jiang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add feature-space objective Jtr (L1 + Lcon) (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 85.75 -> 61.50 | Improved |
| add repulsive contrastive loss Lcon (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 70.75 -> 61.50 | Improved |
| add L1 deviation loss (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 64.00 -> 61.50 | Improved |
| add robustness-augmented universal patch attack (RUPA) (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 62.25 -> 61.50 | Improved |
| add Patch Attention Dominance loss (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 62.50 -> 61.50 | Improved |
| add Patch Semantic Misalignment loss (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 63.50 -> 61.50 | Improved |
| use joint action and direction text probes instead of action only (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 71.25 -> 61.50 | Improved |
| use joint action and direction probes instead of direction only (ablation) | UPA-RFAS · LIBERO transfer to OpenVLA-oft (physical) | task success rate (lower = stronger attack): 75.00 -> 61.50 | Improved |
| use UPA-RFAS instead of best baseline objective | OpenVLA-7B surrogate · LIBERO transfer to OpenVLA-oft-w (simulated) | task success rate (lower = stronger attack): 41.25 -> 5.75 | Improved |
| use UPA-RFAS instead of best baseline objective | OpenVLA-7B surrogate · LIBERO transfer to OpenVLA-oft-w (physical) | task success rate (lower = stronger attack): 65.00 -> 40.25 | Improved |
Closest papers by meaning
- BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization (NeurIPS 2025 · similarity 0.90)
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- Steal the Patch Size: Adversarially Manipulate Vision Language Models (ICML 2026 · similarity 0.90)
- ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time (ICLR 2025 · similarity 0.89)
- Attention! Your Vision Language Model Could Be Maliciously Manipulated (NeurIPS 2025 · similarity 0.88)
- Failures to Find Transferable Image Jailbreaks Between Vision-Language Models (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
