Data underlying the publication: Learning collision risk proactively from naturalistic driving data at scale

This dataset includes the resulting data of the research: Learning collision risk proactively from naturalistic driving data at scale. It is organised into two zipped files: one from the PreparedData folder and another from the ResultData folder. The PreparedData archive contains the processed and segmented training samples from highD, ArgoverseHV, and SHRP2 NDS, along with the checkpoints and loss logs generated during GSSM posterior inference. The ResultData archive gathers the outcomes of the full experimental pipeline, including test set preparation, first-stage safety evaluations, and second-stage conflict and risk analyses. Overall, this dataset supports the research aimed at learning collision risk from naturalistic driving interactions, where a context-aware, scalable, and generalisable method is proposed. The scripts that produced these data are open-sourced at https://github.com/Yiru-Jiao/GSSM

本数据集收录了题为《大规模自然驾驶数据下主动学习碰撞风险》的研究产出数据。该数据集被整理为两个压缩包，分别对应PreparedData文件夹与ResultData文件夹。其中PreparedData压缩包包含来自highD、ArgoverseHV及SHRP2 NDS的经处理与分割后的训练样本，同时附带GSSM后验推断过程中生成的模型检查点（checkpoint）与损失日志。ResultData压缩包汇集了完整实验流程的全部产出结果，涵盖测试集构建、第一阶段安全性评估以及第二阶段冲突与风险分析。总体而言，本数据集可支撑面向自然驾驶交互场景下碰撞风险学习的相关研究，该研究提出了一种具备上下文感知能力、可扩展且可泛化的方法。生成该数据集的代码脚本已在https://github.com/Yiru-Jiao/GSSM 开源。