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

---
license: cc0-1.0
task_categories:
- object-detection
- image-segmentation
- depth-estimation
tags:
- lunar
- synthetic
- robotics
- COCO
- Isaac-Sim
- rover
pretty_name: LunarSim Rover Traverse Synthetic Dataset V1
---

# LunarSim Rover Traverse Synthetic Dataset V1

## Overview

Synthetic lunar surface dataset generated using
[LunarSim](https://elementrobotics.space), 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:

```bibtex
@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}
}
```