Research map / Adversarial robustness and security / Attacks on text and graphs
Phantom Menace: Exploring and Enhancing the Robustness of VLA Models Against Physical Sensor Attacks
Xuancun Lu, Jiaxiang Chen, Shilin Xiao, Zizhi Jin, Zhangrui Chen, Hanwen Yu, Bohan Qian, Ruochen Zhou, Xiaoyu Ji, Wenyuan Xu
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 medium-intensity Laser Blinding attack | OpenVLA · Libero-Spatial (simulator) | task success rate: 84.7 -> 68.2 (strong: 0.0) | Hurt |
| add strong Laser Blinding attack to OpenVLA-OFT | OpenVLA-OFT · Libero-Object (simulator) | task success rate: 98.4 -> 5.6 | Hurt |
| add EM Truncation attack (medium) | OpenVLA · Libero (simulator) | task success rate: 88.4 -> 0.6 on Object | Hurt |
| add camera attacks (EM Truncation medium) to pi0 | pi0 · Libero-Spatial (simulator) | task success rate: 96.8 -> 96.0 | No clear effect |
| add Voice Spoofing attack to OpenVLA-OFT | OpenVLA-OFT · Libero (simulator) | task success rate: 97.6 -> 0.0 on Spatial | Hurt |
| add Voice DoS attack to OpenVLA-OFT | OpenVLA-OFT · Libero-Goal vs Libero-Object | task success rate: 97.9 -> 10.2 (Goal); 98.4 -> 97.6 (Object) | Mixed |
| add adversarial training (medium Laser Color Strip) (ablation) | OpenVLA · Libero-Spatial (simulator) | task success rate: 44.6 -> 82.0 | Improved |
| add adversarial training (medium EM Color Strip) (ablation) | OpenVLA · Libero-Spatial (simulator) | task success rate: 34.8 -> 72.0 | Improved |
| add adversarial training on clean data (ablation) | OpenVLA · Libero-Spatial (no attack) | task success rate: 84.7 -> 81.8 | Hurt |
| add adversarial training (medium EM Truncation) (ablation) | OpenVLA · Libero-Object | task success rate: 0.6 -> 3.4 | No clear effect |
| add adversarial training (medium Light Projection) (ablation) | OpenVLA · Libero-Goal | task success rate: 26.4 -> 28.4 | No clear effect |
Closest papers by meaning
- BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization (NeurIPS 2025 · similarity 0.91)
- ReaLM: A Unified Red-Teaming Benchmark for Physical-World VLMs (NeurIPS 2026 · similarity 0.90)
- TRAP: Hijacking VLA CoT-Reasoning via Adversarial Patches (ICML 2026 · similarity 0.88)
- RedVLA: Physical Red Teaming for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- BadRobot: Jailbreaking Embodied LLM Agents in the Physical World (ICLR 2025 · similarity 0.87)
- AdvEDM: Fine-grained Adversarial Attack against VLM-based Embodied Agents (NeurIPS 2025 · similarity 0.87)
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.
