📷 Example Scenes from the STONE Dataset
- [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.
- [2026] The final dataset will be released. Download will be available via a Google Form.
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
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
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
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
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:
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}
}
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.