STONE: Multi-Modal Off-Road Traversability Dataset from South Korea

7,000 keyframes of multi-modal data for off-road 3D traversability prediction, collected by autonomous ground vehicles. The dataset includes surround-view imagery from six cameras, 128-channel LiDAR scans, and voxel-level terrain annotations. It was created by harpreetsahota and last updated on 2026-05-29.

Use Cases

• Train 3D traversability prediction models based on synchronized LiDAR and camera data.
• Develop multi-modal sensor fusion algorithms for off-road terrain classification.
• Benchmark autonomous navigation systems in diverse outdoor environments.
• Research voxel-level semantic segmentation of terrain into traversability categories.

Strengths

• Contains 7,000 annotated keyframes, providing a substantial base for model training.
• Includes high-resolution (1904×1200) surround-view imagery from 6 cameras for full scene context.
• Provides dense 128-channel LiDAR point clouds with approximately 230,000 points per scan.
• Offers voxel-level annotations classifying terrain into four distinct traversability categories.

Limitations

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

Provenance

Source

huggingface

Collection Method

Collected by autonomous ground vehicles.

Freshness

Last updated 2026-05-29 17:40:45; freshness should be verified.

Geography

Four outdoor environments in South Korea.