STONE: Off-Road 3D Traversability Dataset from South Korea

7,000 keyframes of multi-modal data for off-road 3D traversability prediction, collected by autonomous ground vehicles across four outdoor environments in South Korea. The dataset provides surround-view imagery from 6 cameras, 128-channel LiDAR scans, and voxel-level traversability annotations. It was created by Voxel51 and last updated in June 2026.

Use Cases

• Training 3D traversability prediction models based on LiDAR point clouds and camera imagery.
• Developing multi-modal sensor fusion algorithms for off-road autonomous navigation.
• Benchmarking terrain classification systems for free, traversable, potentially traversable, and non-traversable regions.
• Simulating and validating autonomous vehicle behavior in complex outdoor environments.

Strengths

• 7,000 annotated keyframes provide a substantial basis for model training.
• Multi-modal data includes 6 surround-view cameras (1904×1200) and 128-channel LiDAR scans (~230K points per frame).
• Voxel-level annotations offer detailed 3D traversability classification across four distinct outdoor environments.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Data may reflect geographic bias inherent to its collection in South Korea.

Provenance

Source

Voxel51

Collection Method

Collected by autonomous ground vehicles.

Time Range

null

Freshness

Last updated 2026-06-09 19:25:38; freshness should be verified.

Geography

Four outdoor environments in South Korea.