TruckDrive

TruckDrive: Long-Range Autonomous Highway Driving Dataset Torc Robotics · Princeton University · CVPR 2026 Filippo Ghilotti, Edoardo Palladin, Samuel Brucker, Adam Sigal, Mario Bijelic, Felix Heide TruckDrive is a long-range autonomous highway driving dataset designed for heavy-truck safety, perception, prediction, and planning research. It targets high-speed highway operation, where reliable scene understanding hundreds of meters ahead is required for anticipatory planning and safe braking. This repository hosts the TruckDrive Devkit together with release documentation. License See LICENSE-NONCOMMERCIAL.txt for the full non commercial license text. See LICENSE-COMMERCIAL for the full commercial license text. Downloading the TruckDrive Dataset To download the dataset from Hugginfg Face use the provided python script. Download the full dataset: python download_truckdrive.py \ out /PATH/TO/TruckDrive \ all-scenes \ all-modalities To download only one scene: python download_truckdrive.py \ out /PATH/TO/TruckDrive \ scene scene_28_1 \ all-modalities To download only specific folders, replace --all-modalities with options such as: camera --lidar --radar --poses --calibration --annotations --accumulated-gt-depth To automatically unzip the downloaded files add: --unzip. To automatically remove the downloaded files after unzipping add: --remove-zips-after-unzip. Highlights 475k synchronized multimodal samples 165k densely annotated frames up to 1,000 m for 2D benchmark annotations up to 400 m for 3D benchmark annotations 7 long-range FMCW LiDARs, 3 short-range LiDARs, 10 4D radars, and 11-15 cameras supports perception, tracking, depth estimation, prediction, planning, and end-to-end driving Dataset Overview TruckDrive targets long-range, highway-scale autonomous driving for semi-trucks and other heavy commercial vehicles. It is designed to stress perception and planning systems beyond the short-range assumptions common in urban autonomous-driving benchmarks. Item Paper-rep