An Augmented Dataset of Autonomous Vehicle Collisions in California

The rapid advancement of autonomous vehicles (AVs) and the emergence of robotaxi services have the potential to transform urban mobility. However, public concerns regarding AV safety remain a significant barrier to widespread adoption. While extensive AV testing has been conducted in controlled environments, real-world accident data is crucial for understanding safety risks and enhancing public trust. This study addresses the gap in AV-specific accident datasets by presenting a comprehensive augmented dataset of AV collisions in California, covering all reported AV-involved accidents from January 1, 2019, to December 31, 2024. The dataset integrates information from California DMV accident reports, geographical data derived using Geographic Information System (GIS) tools, and semantic information extracted via Large Language Models (LLMs). The resulting tabular dataset supports a wide range of applications, including AV crash pattern analysis, contributing factor identification, risk assessment, safety algorithm refinement, regulatory policy development, and urban infrastructure planning.自动驾驶汽车（Autonomous Vehicles, AVs）的快速发展与机器人出租车服务的兴起，有望重塑城市出行格局。然而，公众对自动驾驶汽车安全性的担忧仍是其大规模普及的重大阻碍。尽管已有大量自动驾驶汽车在受控环境中开展测试，但真实世界的事故数据对于厘清安全风险、提升公众信任至关重要。本研究针对自动驾驶汽车专属事故数据集的空白，发布了一套覆盖2019年1月1日至2024年12月31日加州所有上报涉自动驾驶汽车事故的全面增强型碰撞数据集。该数据集整合了加州车辆管理局（California DMV）的事故报告信息、通过地理信息系统（Geographic Information System, GIS）工具提取的地理数据，以及经由大语言模型（Large Language Models, LLMs）提取的语义信息。最终生成的表格型数据集可支撑广泛的应用场景，包括自动驾驶汽车碰撞模式分析、事故诱因识别、风险评估、安全算法优化、监管政策制定以及城市基础设施规划。