GEM-CAN: Real-World CAN-Bus Attack Scenarios on an Autonomous Vehicle for Intrusion Detection

This dataset provides labeled CAN-bus traffic captured from a GEM e6 autonomous vehicle under normal operation and controlled cyberattacks. It includes ~143K frames spanning nominal driving (~100K), DoS floods using ID 0x00000000 (~41K), and data-tampering injections targeting brake and steering functions (~1.3K). Each entry records timestamp, arbitration ID, DLC, payload bytes, and a Normal/Attack label, with metadata detailing attack windows, bus load, bitrate, and test conditions. Data were collected via PCAN-View/PCAN-USB on a closed track while the vehicle operated autonomously. The dataset provides real-vehicle evidence of availability and integrity attacks—supporting reproducible evaluation of lightweight automotive intrusion detection systems. A JSON metadata file is provided to summarize experimental conditions, attack definitions, logging parameters, and labeling policy, complementing the CSV files.本数据集收录了GEM e6型自动驾驶车辆在正常运行与受控网络攻击场景下捕获的带标注控制器局域网总线（CAN-bus）流量数据。该数据集共包含约14.3万个数据帧，涵盖正常行驶场景的约10万个数据帧、使用ID 0x00000000发起的拒绝服务（Denial of Service，DoS）泛洪攻击数据约4.1万个，以及针对制动与转向功能的数据篡改注入攻击数据约1300个。每条数据均记录了时间戳、仲裁ID、数据长度码（Data Length Code，DLC）、有效载荷字节数据与正常/攻击类别标签；同时附带元数据，详细说明攻击时段、总线负载、比特率及测试环境等信息。 数据集采集于封闭赛道场景，采集工具为PCAN-View与PCAN-USB，采集期间车辆处于自动驾驶状态。本数据集提供了真实车辆场景下可用性与完整性攻击的实测证据，可支撑轻量化车载入侵检测系统的可复现性评估。数据集配套提供JSON格式元数据文件，用于汇总实验条件、攻击定义、日志参数与标注规则，以补充CSV格式的数据文件。