vehicle collision trajectory dataset
收藏资源简介:
本研究构建了一个基于真实世界车辆事故报告的车辆碰撞轨迹数据集。该数据集通过物理模拟器CARLA生成,利用事故报告中的物理线索和情境信息,模拟出车辆碰撞后的轨迹。数据集用于微调大型语言模型LLaMA,使其能够根据用户描述生成与现实世界一致的车辆碰撞轨迹。数据集的应用领域主要在于增强自动驾驶系统对安全关键事件的应对能力,解决现有数据集中碰撞场景的稀缺问题。
This study constructs a vehicle collision trajectory dataset based on real-world vehicle accident reports. This dataset is generated using the CARLA physics simulator, which simulates post-collision vehicle trajectories by leveraging physical cues and contextual information extracted from accident reports. This dataset is utilized to fine-tune the Large Language Model LLaMA, enabling it to generate vehicle collision trajectories consistent with real-world conditions based on user-provided descriptions. The primary application scope of this dataset lies in enhancing the response capabilities of autonomous driving systems towards safety-critical events, addressing the shortage of collision scenarios in existing datasets.




