摘要
While vision-language models (VLMs) demonstrate remarkable capabilities in autonomous driving applications, they face significant challenges in safety-critical environments due to their insufficient risk awareness, susceptibility to visual distractions, and challenges in effectively balancing safety requirements with learning efficiency. To mitigate these challenges, we propose safe-GRPO, a safe-aware group relative policy optimization framework specifically designed for safe decision-making in VLM-based driving systems. Safe-GRPO introduces three key innovations: i) an uncertainty-aware, safety-sensitive reward mechanism that dynamically adjusts reward and penalty scales according to both risk severity and model uncertainty; ii) an uncertainty-guided noise-enhanced training method that adaptively improves robustness against visual interferences; and iii) a risk-aware dynamic clipping strategy that controls policy updates based on scenario-specific risks. Extensive experiments on an enhanced real-world scenarios with driving theory tests dataset (3926 risk-annotated scenarios) show that safe-GRPO significantly outperforms existing approaches in decision accuracy, robustness, and risk prediction accuracy-achieving a 75.14% safe decision rate in high-risk scenarios, demonstrating notable advancements towards reliable real-world autonomous driving applications. © 2015 Chinese Institute of Electronics.
| 源语言 | 英语 |
|---|---|
| 页(从-至) | 1493-1506 |
| 期刊 | Chinese Journal of Electronics |
| 卷 | 35 |
| 期 | 4 |
| DOI | |
| 出版状态 | 已出版 - 1 7月 2026 |
基金资助
National Natural Science Foundation of China, NNSF, (62225202, 62302022)
关键词
- Autonomous vehicles
- Computer vision
- Decision theory
- Machine learning
- Robustness (control systems)
- Safety engineering
- Statistical tests
- Uncertainty analysis
- Autonomous driving
- Group relative policy optimization
- Language model
- Learning frameworks
- Policy optimization
- Reinforcement learnings
- Risk awareness
- Safety reinforcement learning
- Uncertainty
- Vision-language model
- Risk perception
- Vision-language models (VLMs)
收录
- Scopus
- EI
学术指纹
探究 'A Safety-Enhanced Reinforcement Learning Framework for Vision-Language Models in Autonomous Driving' 的科研主题。它们共同构成独一无二的学术指纹。引用此
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver