STONE

<div align="center"> <h1> <img src="assets/images/STONE_dataset_logo_v2.png" alt="STONE Logo" width="70" valign="middle"> STONE Dataset </h1>

A Scalable Multi-Modal Surround-View 3D Traversability Dataset

for Off-Road Robot Navigation

</div> <p align="center"> <img src="assets/images/teaser.png" width="420" alt="STONE Teaser"> </p> <p align="center"> <a href="https://konyul.github.io/STONE-dataset/assets/paper/final_paper_compressed.pdf"> <img src="https://img.shields.io/badge/Paper-PDF-red?style=for-the-badge&logo=arxiv"> </a> <a href="https://konyul.github.io/STONE-dataset"> <img src="https://img.shields.io/badge/Homepage-Project_Page-blue?style=for-the-badge&logo=googlechrome"> </a> </p> <br> <p align="center"> <b>📷 Example Scenes from the STONE Dataset</b> </p> <p align="center"> <img src="assets/images/teaser_mid.png" width="800"> </p> <br>

📢 Updates

• [2026-03] We opened the STONE Dataset GitHub.
• [2026-03] We opened the STONE Dataset Website.
• [2026-02] Our paper has been accepted to ICRA 2026.

<br>

🔜 Upcoming

• [2026] The final dataset will be released. Download will be available via a Google Form. <br>

🌍 Overview

STONE is a large-scale multi-modal dataset designed for off-road navigation and 3D traversability prediction.

Key features of STONE include:

• Trajectory-guided 3D traversability maps generated by a fully automated labeling pipeline
• Multi-modal surround-view sensing with 128-channel LiDAR, six RGB cameras, and three 4D imaging radars
• Diverse environments and conditions, including grasslands, farmlands, construction sites, lakes, and both day and night scenarios
• Geometry-aware labeling based on terrain attributes such as slope, elevation, and roughness
• Benchmark for voxel-level 3D traversability prediction with single-modal and multi-modal baselines

<br>

🤖 Robot Platform & Sensor Setup

<p align="center"> <img src="assets/images/robot_setup.png" width="800"> </p>

Platform

• UGV: Bunker Pro
• Operating System: Ubuntu 22.04
• Framework: ROS 2 Humble

Sensors

• 360° Rotating LiDAR: 1 × Hesai OT128
• Multi-view RGB Cameras: 6 × Basler ACE2 2A1920-51gcPRO
• 4D Imaging Radars: 3 × Continental ARS 548 RDI
• GNSS/INS: NovAtel PIM222A dual-antenna GNSS/INS
• IMU: EPSON G366P

<br>

📊 Dataset

The STONE dataset provides voxel-level 3D traversability annotations.

Traversability classes

The dataset contains 4 classes:

| Class ID | Label | |--------|--------| | 0 | Free | | 1 | Traversable | | 2 | Potentially Traversable | | 3 | Non-Traversable |

Ground-truth labels are provided as labels.npz files.

Voxel configuration

• Voxel size: [0.2 m, 0.2 m, 0.2 m]
• Range: [-25.6 m, -25.6 m, -2.0 m, 25.6 m, 25.6 m, 4.4 m]
• Volume size: [256, 256, 32]

Dataset Structure

• The dataset structure follows the conventions of nuScenes and Occ3D-nuScenes.
• The 4D radar data are provided in ROS bag format (.bag).
• The hierarchy of folder is described below:

STONE_Dataset
│
├── gts
│   └── [scene_name]
│       └── [frame_token]
│           └── labels.npz
│
├── samples
│   ├── CAM_BACK
│   │   └── n001-2025-08-22-07-14-16+0900__CAM_BACK__1755846856289490.jpg
│   ├── CAM_BACK_LEFT
│   │   └── ...
│   ├── CAM_BACK_RIGHT
│   │   └── ...
│   ├── CAM_FRONT
│   │   └── ...
│   ├── CAM_FRONT_LEFT
│   │   └── ...
│   ├── CAM_FRONT_RIGHT
│   │   └── ...
│   └── LIDAR_TOP
│       └── n001-2025-08-22-07-14-16+0900__LIDAR_TOP__1755846856289490.pcd.bin
│
└── v1.0-trainval
    ├── attribute.json
    ├── calibrated_sensor.json
    ├── category.json
    ├── ego_pose.json
    ├── instance.json
    ├── lidarseg.json
    ├── log.json
    ├── map.json
    ├── sample.json
    ├── sample_annotation.json
    ├── sample_data.json
    ├── scene.json
    ├── sensor.json
    └── visibility.json

<br>

📥 Download

The STONE dataset will be released through a Google Form.

Please fill out the form to request access.
The download link will be provided after approval.

• Dataset: Coming Soon
• ROS bags: Coming Soon

<br>

📜 License

<p> <img src="https://img.shields.io/badge/Dataset%20License-CC%20BY--NC--ND%204.0-green?style=for-the-badge"> <img src="https://img.shields.io/badge/Code%20License-Apache%202.0-blue?style=for-the-badge"> </p>

The STONE dataset is published under the Creative Commons Attribution–NonCommercial–NoDerivatives 4.0 International License (CC BY-NC-ND 4.0).

All codes in this repository are released under the Apache License 2.0.

Under the CC BY-NC-ND 4.0 license, the dataset may be used for non-commercial research purposes only.
Users must give appropriate credit to the original authors when using the dataset.

For more details, please refer to:

• https://creativecommons.org/licenses/by-nc-nd/4.0/
• https://www.apache.org/licenses/LICENSE-2.0 <br>

📑 Citation

If you find the STONE dataset useful in your research, please consider citing our paper:

@inproceedings{park2026stone,
  title={STONE: A Scalable Multi-Modal Surround-View 3D Traversability Dataset for Off-Road Robot Navigation},
  author={Park, Konyul and Kim, Daehun and Oh, Jiyong and Yu, Seunghoon and Park, Junseo and Park, Jaehyun and Shin, Hongjae and Cho, Hyungchan and Kim, Jungho and Choi, Jun Won},
  booktitle={Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)},
  year={2026}
}

<br>

🙏 Acknowledgement

The STONE dataset is contributed by Konyul Park, Daehun Kim, Jiyong Oh, Seunghoon Yu, Junseo Park, Jaehyun Park, Hongjae Shin , Hyungchan Cho, Jungho Kim, and Jun Won Choi.

We sincerely thank the maintainers of the following open-source projects that enabled the development of the STONE dataset:
MMDetection3D by OpenMMLab.