"LI-HAE dataset"

"Hitch angle estimation is a critical perception task for the safe automation of articulated vehicles, enabling precise trailer-state feedback for path planning, stability control, and assisted reversing. Despite growing interest in learning-based approaches, the field lacks a publicly available benchmark that combines multi-modal sensing, diverse driving conditions, and high-fidelity ground truth across multiple trailer geometries. We introduce LI-HAE (LiDAR-Inertial-based Hitch Angle Estimation), a real-world dataset designed to address this gap.LI-HAE was collected from a tractor\u2013trailer system equipped with a trailer-mounted LiDAR and dual RTK-GNSS\/INS units rigidly attached to the tractor and trailer frames, respectively. The relative heading derived from differential RTK positioning provides centimeter-accurate hitch angle ground truth at each frame. Three trailer configurations with distinct geometries are included: Charger (Long\u2013Flat), Dummy (Short\u2013Tall), and Temporary (Compact). Each synchronized frame delivers five modalities: (1) a trailer-perspective LiDAR point cloud, (2) a five-step window of 3-axis IMU measurements (linear acceleration and angular velocity), (3) front-wheel steering angle, (4) longitudinal vehicle velocity.The dataset comprises 131,680 frames across 89 segments spanning approximately 79.6 km of real-world driving. Seven scenario types are covered: highway cruise, urban intersection, S-curve, uphill\/downhill (banked and inclined), reverse, rotary, and U-turn. Hitch angle excursions range from \u221271.3\u00b0 to +87.7\u00b0 in the Short\u2013Tall configuration, capturing large-angle dynamics during low-speed maneuvers such as reversing and tight cornering. Official train\/val\/test splits are provided for all three configurations. "