Research map / Adversarial robustness and security / Attacks on text and graphs
Exploring the Adversarial Vulnerabilities of Vision-Language-Action Models in Robotics
Taowen Wang, Cheng Han, James Liang, Wenhao Yang, Dongfang Liu, Luna Xinyu Zhang, Qifan Wang, Jiebo Luo, Ruixiang Tang
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 |
|---|---|---|---|
| use UADA objective instead of UMA (negated cross-entropy) baseline | adversarial patch on OpenVLA · LIBERO (Simulation setup, DoF1 NAD) | NAD (%): 14.1 -> 21.0 | Improved |
| use UADA instead of UMA | adversarial patch on OpenVLA-7b · LIBERO (Physical setup, avg FR) | failure rate (%): 73.8 -> 96.8 | Improved |
| attack multiple DoFs (DoF1~3) rather than DoF1 only with UADA (ablation) | UADA · LIBERO (Physical setup, avg FR) | failure rate (%): 96.8 -> 100 | Improved |
| use random noise patch instead of learned adversarial patch | patch baseline · LIBERO (avg FR) | failure rate (%): 23.5 -> 28.3 | No clear effect |
| target DoF4 (x-axis rotation) with TMA (ablation) | TMA · LIBERO Simulation (avg FR) | failure rate (%): 99.5 -> 67.6 (DoF1 vs DoF4) | Hurt |
| increase inner-loop steps for UADA (ablation) | UADA · LIBERO | NAD | Mixed |
| increase patch size from 3% to 10% of image (ablation) | TMA · LIBERO (avg of 4 suites) | failure rate | Improved |
| transfer patches from Bridge V2 physical setup to LIBERO | UADA · LIBERO transfer | failure rate (%): 23.5 -> 100 | Improved |
| apply four prior defense techniques | adversarial patch · LIBERO | failure rate | No clear effect |
Closest papers by meaning
- BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization (NeurIPS 2025 · similarity 0.93)
- TRAP: Hijacking VLA CoT-Reasoning via Adversarial Patches (ICML 2026 · similarity 0.90)
- ReaLM: A Unified Red-Teaming Benchmark for Physical-World VLMs (NeurIPS 2026 · similarity 0.89)
- AdvEDM: Fine-grained Adversarial Attack against VLM-based Embodied Agents (NeurIPS 2025 · similarity 0.89)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · 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.
