arXiv cs.AI· Everett Richards·· 3 天前AI 评分18
面向自动驾驶危险检测的视觉语言模型任务对齐稳定性分析
Task-Aligned Stability Analysis of Vision-Language Models for Autonomous Driving Hazard Detection
AI 导读
研究用 BDD100K 道路场景的受控损坏实验，检验 CLIP 图像-文本相似度导出的危险分数变化能否由嵌入漂移预测。结果显示两者关系高度依赖损坏类型：部分损坏族中表征漂移与决策漂移强耦合，另一些则在嵌入变化不大时仍引发危险决策不稳定。多数损坏族通过漏检抑制危险识别，遮挡则触发误报，表明鲁棒性基准应纳入任务对齐的稳定性指标。
正文 · 原文
Abstract:Vision-language models (VLMs) are increasingly used for scene understanding in autonomous driving, but robustness analysis often relies on task-agnostic embedding stability alone. We study whether corruption-induced embedding drift predicts changes in a task-aligned hazard score derived from CLIP image-text similarities. Using controlled corruptions on BDD100K road scenes, we compare embedding drift against margin drift, defined as the change in hazard score under perturbation. The relationship is highly corruption-dependent: some families exhibit strong coupling between representation drift and decision drift, while others induce hazardous decision instability despite relatively modest embedding change. Furthermore, corruption families differ in failure direction: most suppress hazard detections via false negatives, while occlusion instead triggers false alarms, suggesting that benchmark design should account for asymmetric failure modes, not just overall instability rates. These results suggest that robustness benchmarks should include task-aligned stability measures in addition to embedding-level perturbation statistics.
| Comments: | 8 pages (5 main body + 3 references / appendices). ICML 2026 Workshop on Combining Theory and Benchmarks (CTB) |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO) |
| ACM classes: | I.2.10; I.4.8 |
| Cite as: | arXiv:2606.11889 [cs.CV] |
| (or arXiv:2606.11889v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2606.11889 arXiv-issued DOI via DataCite |
Submission history
From: Everett Richards [view email]
[v1]
Wed, 10 Jun 2026 10:20:14 UTC (5,906 KB)
[v2]
Mon, 12 Oct 2026 14:48:01 UTC (5,906 KB)
来源：arXiv cs.AI · arxiv.org