Research map / Adversarial robustness and security / Attacks on text and graphs
PhysPatch: A Physically Realizable and Transferable Adversarial Patch Attack for Multimodal Large Language Models-based Autonomous Driving Systems
Qi Guo, Xiaojun Jia, Shanmin Pang, Simeng Qin, Lin Wang, Ju Jia, Yang Liu, Qing Guo
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 potential-field-based mask update instead of fixed 120x120 square patch (ablation) | PhysPatch · nuScenes stop-sign target | ASR (%): 37.9 -> 40.3 (GPT-4o) | Improved |
| add SVD-based local alignment loss instead of standard local alignment (ablation) | PhysPatch · nuScenes stop-sign target | ASR (%): 38.8 -> 40.3 (GPT-4o) | Improved |
| add patch-guided crop-resize instead of random cropping (ablation) | PhysPatch · nuScenes stop-sign target | ASR (%): 37.5 -> 40.3 (GPT-4o) | Improved |
| add SVD-based local alignment loss (ablation) | PhysPatch · nuScenes stop-sign target | ASR (%): 14.0 -> 14.5 (Claude-4) | Improved |
| add potential-field mask update (ablation) | PhysPatch · nuScenes stop-sign target | ASR (%): 22.9 -> 25.4 (Gemini-2.5) | Improved |
| use PhysPatch instead of FOA-Attack (patch of ~1% image area) | MLLM-based AD systems · nuScenes stop-sign, QVQ-Plus | ASR (%): 20.5 -> 29.2 | Improved |
| use PhysPatch instead of FOA-Attack under PAD defense | GPT-4o · nuScenes stop-sign, PAD defense | ASR (%): 6.7 -> 16.8 | Improved |
| PhysPatch vs FOA-Attack on Qwen2.5-VL-72B | Qwen2.5-VL-72B · nuScenes stop-sign target | ASR (%): 14.4 -> 15.4 | No clear effect |
| PhysPatch vs FOA-Attack on generation time | attack optimization · Generation time | seconds: 174 -> 152 | Improved |
| PhysPatch vs lightweight baselines on generation time | attack optimization · Generation time | seconds: 79 -> 152 (AttackVLM) | Hurt |
Closest papers by meaning
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- MAGIC: Mastering Physical Adversarial Generation in Context Through Collaborative LLM Agents (AAAI 2026 · similarity 0.86)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.86)
- AdvEDM: Fine-grained Adversarial Attack against VLM-based Embodied Agents (NeurIPS 2025 · similarity 0.86)
- SlowTrack: Increasing the Latency of Camera-Based Perception in Autonomous Driving Using Adversarial Examples (AAAI 2024 · similarity 0.85)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.85)
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.
