elementrobotics/lunarsim-shoemaker-traverse-100m-v1

LunarSim Rover Traverse Synthetic Dataset V1

Overview

Synthetic lunar surface dataset generated using LunarSim, an NVIDIA Isaac Sim-based lunar environment simulator. Images simulate a forward-facing rover camera traversing terrain near the Shoemaker crater at the lunar south pole. The dataset includes RGB images, depth maps, semantic/instance segmentation masks, shadow masks, and COCO-format rock detection annotations.

Generation

• Simulator: LunarSim (NVIDIA Isaac Sim 4.5.0)
• Terrain: Shoemaker crater DEM (5m/pixel native, upscaled to 0.1m)
• Location: Lunar south pole (-87.18° lat, 62.84° lon)
• Terrain features: Upscaled from LRO DEM, sub-5m craters added based on DSNE and Lunar Sourcebook distributions, rock distribution using Apollo rock models and DSNE data
• Traverse distance: 100.0 m
• Rover speed: 0.25 m/s
• Camera frame rate: 5.0 Hz
• Number of frames: 2000
• Frame spacing: 0.05 m
• Duration: 400.0 s

Camera Configuration

• Resolution: 1920 x 1080 px
• Height above ground: 0.3 m
• Pitch offset: 15.0° below ground-parallel
• Focal length: 24.0 mm
• F-stop: 0.0 (pinhole)
• Clipping range: 0.01–15.0 m
• Ground plane estimation: Least-squares plane fit over 0.5m x 0.5m rover footprint
• Orientation: Forward along direction of travel, pitched down from local ground plane

Directory Structure

├── images/            RGB frames (PNG, uint8)
├── depth/             Depth maps (PNG, uint16, values in millimetres)
├── segmentation/      Semantic masks (PNG, uint8: 0=void, 1=rock, 2=terrain, 3=shadow)
├── instance/          Instance ID maps (PNG, uint16)
├── shadow/            Binary shadow masks (PNG, 0=lit, 255=shadow)
├── annotations.json   COCO-format annotations
├── cam_poses.csv      Per-frame camera poses (x,y,z,qw,qx,qy,qz)
├── frame_poses.csv    Per-frame metadata (frame_id, x, y, theta_deg, lat, lon)
├── dataset_info.json  Full generation configuration and parameters
└── README.md          This file

Annotation Format

• Format: COCO object detection with instance segmentation
• Categories: rock (id 1), terrain (id 2)
• Bounding boxes: [x, y, width, height] in pixel coordinates
• Instance segmentation: RLE-encoded binary masks
• Per-image metadata: camera_position [x,y,z] and camera_orientation [qw,qx,qy,qz]
• Filtering: Annotations filtered by minimum area (0.005%), maximum occlusion (0.95), and minimum brightness (0.005)

Coordinate Systems

Reference	Local (m)	Projected (m)	Lunar Lat/Lon
SW corner	(0, 0)	(75950, 38950)	-87.186°, 62.850°
Centre	(50, 50)	(76000, 39000)	-87.184°, 62.835°
NE corner	(100, 100)	(76050, 39050)	-87.181°, 62.821°

Projection: South-pole polar stereographic (Moon sphere R=1,737,400 m)

Use Cases

• Rock detection and segmentation model training
• Depth estimation from monocular lunar images
• Shadow-aware terrain classification
• Visual odometry and navigation algorithm development
• Sim-to-real transfer learning for lunar rover perception

Citation

If you use this dataset, please cite:

@misc{lunarsim_sdg_2025,
  title={LunarSim Rover Traverse Synthetic Dataset V1},
  author={Element Robotics},
  year={2025},
  url={https://huggingface.co/datasets/elementrobotics/lunarsim-shoemaker-traverse-100m-v1}
}