DiDrive：面向安全自动驾驶离线强化学习的风险感知分层扩散框架
原标题：DiDrive: A Risk-Aware Hierarchical Diffusion Framework for Safe Offline Reinforcement Learning in Autonomous Driving
AI 摘要
福州大学和中国科学院自动化研究所的研究人员提出了DiDrive，一个用于自动驾驶安全离线强化学习的风险感知分层扩散框架。DiDrive结合了风险感知分层扩散架构和3DICE策略优化范式，以应对分布偏移和重尾风险信号。在CARLA模拟基准测试中，DiDrive在成功率、平均奖励和车道偏离方面优于IQL、CQL和Diffusion-QL等基线，在60辆背景车辆的高密度场景中实现了85%的最高成功率和4295.68的最高平均奖励。
正文节选
[1]This work was supported by the Natural Science Foundation of Fujian Province, China under Grant 2026J001253. [1] 1]organization=College of Computer and Data Science, Fuzhou University, addressline=, city=Fuzhou, postcode=350116, state=Fujian, country=China 2]organization=Maynooth International College of Engineering, Fuzhou University, addressline=, city=Fuzhou, postcode=350116, state=Fujian, country=China 3]organization=Institute of Automation, Chinese Academy of Sciences, addressline=, city