MDrive; MDrive-V2XPnP
收藏资源简介:
MDrive是由加利福尼亚大学洛杉矶分校团队创建的闭环协同驾驶基准数据集,旨在系统评估多智能体系统在真实驾驶环境中的性能。该数据集包含225个多样化场景,涵盖从真实世界V2X驾驶日志转换的Real2Sim场景以及基于NHTSA预碰撞类型生成的交互场景,数据来源包括V2XPnP数据集和智能体场景生成管道。数据集通过CARLA模拟器构建,采用Real2Sim转换和智能体生成流程,确保场景的真实性和交互复杂性。该数据集主要应用于自动驾驶领域,旨在解决现有评估协议在闭环驾驶中存在的不足,推动多智能体协同感知与决策协商系统的鲁棒性和泛化能力研究。
MDrive is a closed-loop collaborative driving benchmark dataset developed by the research team at the University of California, Los Angeles (UCLA), which aims to systematically evaluate the performance of multi-agent systems in real-world driving environments. This dataset encompasses 225 diverse scenarios, including Real2Sim scenarios converted from real-world V2X driving logs and interactive scenarios generated based on NHTSA-defined pre-crash types. Its data sources cover the V2XPnP dataset and an agent scenario generation pipeline. Constructed via the CARLA simulator, the dataset adopts Real2Sim conversion and agent generation workflows to ensure the authenticity and interactive complexity of the scenarios. Primarily applied in the field of autonomous driving, this dataset is designed to address the shortcomings of existing evaluation protocols in closed-loop driving, and promote research on the robustness and generalization capabilities of multi-agent collaborative perception and decision-making negotiation systems.




