AutoRace Dataset v1: A Multi-Modal Perception Dataset from a Custom Autonomous Racing Simulator

AutoRace Dataset v1 is a multi-modal perception dataset for autonomous racing, recorded in a custom simulator built on ROS 2 Humble and Gazebo Fortress. The vehicle is a BMW M4 GT4 model driven along a section of the Spa-Francorchamps circuit, covering straight segments, medium-radius corners, and 111 m of elevation change. The sensor suite comprises four cameras (1280×720), three 16-beam LiDARs, a 500 Hz IMU, and a dual-antenna GNSS with a 1.30 m baseline, together with wheel odometry and ground-truth pose and velocity. Data is distributed as a ROS 2 rosbag2 recording in MCAP format. Each sensor is characterized against the recorded data rather than its configuration, and the accompanying documentation includes a sim-to-real gap analysis describing where the simulation diverges from real hardware. Full documentation, sensor tables, calibration and extrinsics (https://github.com/SaeidAbdollahi/autorace_dataset)