HRDX

A Large-Scale Vector HD-Map Dataset
Sahith Reddy Chada, Isht Dwivedi, Nirav Savaliya
Honda Research Institute US, San Jose, CA
Request Dataset GitHub Paper

​​​​​​​

Introduction

HRDX is a large-scale dataset for vector HD-map construction, collected across diverse urban, suburban, and freeway environments in the San Francisco Bay Area. Spanning 40+ hours and 1,400 km of minimally overlapping drives — roughly six times larger than the next-largest public HD-map dataset — it is captured with a fully calibrated, time-synchronized multi-sensor suite and uniquely complemented by precisely aligned aerial orthoimagery.

  • Six synchronized surround cameras, a 128-beam LiDAR, and centimeter-level RTK GNSS/IMU, all cross-calibrated and time-synchronized for pixel-accurate fusion.
  • Precisely aligned aerial orthoimagery at up to 8 cm/pixel, sourced from public county databases — freely redistributable, unlike Google-tile augmentations.
  • 10 vector map classes: lane centerlines, lane lines, stop lines, road boundaries, crosswalks, bike lanes, parking spots, non-drivable areas, intersection centerlines, and road-surface text.
  • 20+ semantic and topological attributes, including lane color and style, intersection turn types, and a rich taxonomy of road-surface markings (arrows, STOP, YIELD, keep-clear).
  • Long, contiguous temporal sequences spanning hours — not fixed-length clips — enabling evaluation of online mapping under sustained drift and error accumulation.
  • Benchmarks and a Composite Score (CS) metric that jointly assesses geometric fidelity and attribute correctness.
40+
Hours of Drives
1,400km
Distance Covered
10
Vector Map Classes
20+
Semantic Attributes
6× Surround Cameras 128-Beam LiDAR RTK GNSS / IMU Aligned Aerial Imagery Attribute-Dense Labels Long Temporal Sequences

Video

Data Format

Drives are segmented into scenes. Each scene is a continuous, time-synchronized sequence with multi-camera images, LiDAR, ego pose, aligned aerial imagery, and continuous HD vector-map annotations.

DATA:
scene_xxx/                   sequence of synchronized sensor data
 |--images/
 |   |--cam{1..6}/*.jpg       six surround-view camera frames
 |--lidar/
 |   |--*.bin                 128-beam LiDAR scans (motion-compensated)
 |--pose/
 |   |--ego_pose.json         6-DoF pose (RTK GNSS/IMU + LIO)
 |--aerial/
 |   |--ortho.tif             aligned aerial orthoimagery (8-15 cm/px)
 |--calibration/
 |   |--sensors.json          intrinsics / extrinsics, time offsets
 |--map/
     |--annotations.json      vector map: polylines & polygons
                              # class, geometry, attributes (color, style,
                              #  turn type, text-on-road, lane neighbors, ...)

Download the Dataset

The dataset is available for non-commercial usage. You must be affiliated with a university and use your university email address to make the request.

Make a Download Request

Citation

This dataset corresponds to the paper "HRDX: A Large-Scale Vector HD-Map Dataset." In the current release, the data is available for researchers from universities. Please cite the following paper if you find the dataset useful in your work:

@article{chada2025hrdx,
  title   = {HRDX: A Large-Scale Vector HD-Map Dataset},
  author  = {Chada, Sahith Reddy and Dwivedi, Isht and Savaliya, Nirav},
  institution = {Honda Research Institute US},
  year    = {2025}
}