Back to datasets
Dataset assetOpen Source CommunityLunar ExplorationPerception and Navigation

LuSNAR

The LuSNAR dataset, created by the Institute of Space Application Engineering and Technology and the Key Laboratory of Space Utilization of the Chinese Academy of Sciences, is a multi‑task, multi‑scenario lunar surface dataset. It contains nine Unreal Engine‑based simulated lunar scenes, each divided by terrain roughness and object density, and provides high‑resolution stereo image pairs, panoramic semantic labels, depth maps, and point clouds. The dataset is intended to support development and validation of perception and navigation algorithms, suitable for 2‑D/3‑D semantic segmentation, visual SLAM, LiDAR SLAM, stereo matching, and 3‑D reconstruction, focusing on autonomous perception and navigation for lunar exploration.

Source
arXiv
Created
Jul 9, 2024
Updated
Jul 9, 2024
Signals
1,035 views
Availability
Linked source ready
Overview

Dataset description and usage context

LuSNAR Dataset

Introduction

LuSNAR is a multi‑sensor (stereo camera, LiDAR, IMU) lunar segmentation, navigation, and reconstruction dataset for autonomous exploration. It comprises 9 Unreal Engine‑based simulated lunar scenes, each partitioned by terrain undulation and object density.

Data Content

The dataset includes:

  • High‑resolution stereo image pairs
  • Panoramic semantic labels
  • Dense depth maps
  • LiDAR point clouds
  • IMU data
  • Rover pose data

Application Scenarios

Applicable for comprehensive evaluation of autonomous perception and navigation systems, including:

  • 2D/3D semantic segmentation
  • Visual/LiDAR SLAM
  • 3D reconstruction

Availability

Total size 108 GB, consisting of:

  • 42 GB stereo image pairs
  • 50 GB depth maps
  • 356 MB semantic segmentation labels
  • 14 GB single‑frame point clouds (with semantics)

Dataset structure:

├── image1
│   ├── RGB
│   │   ├── timestamp1.png
│   │   ├── timestamp2.png
│   │   └── ...
│   ├── Depth
│   │   ├── timestamp1.png
│   │   ├── timestamp2.png
│   │   └── ...
│   └── Label
│       ├── timestamp1.png
│       ├── timestamp2.png
│       └── ...
├── image2
│   ...
├── LiDAR
│   ├── timestamp1.txt
│   ├── timestamp2.txt
│   └── ...
├── Rover_pose.txt
└── IMU.txt

Semantic image color‑to‑class mapping:

Class IDClassColor
0Lunar RegolithBB469C
1Crater7800C8
2RockE8FA50
3MountainAD451F
4Sky22C9F8

LiDAR point‑cloud class mapping:

Class IDClass
-1Lunar Regolith
0Crater
174Rock

File Formats

LiDAR/timestamp.txt

| x [m] | y [m] | z [m] | Class ID |

Rover_pose.txt

| Timestamp [ns] | p_RS_R_x [m] | p_RS_R_y [m] | p_RS_R_z [m] | q_RS_w [] | q_RS_x [] | q_RS_y [] | q_RS_z [] | v_RS_R_x [m s^-1] | v_RS_R_y [m s^-1] | v_RS_R_z [m s^-1] | b_w_RS_S_x [rad s^-1] | b_w_RS_S_y [rad s^-1] | b_w_RS_S_z [rad s^-1] | b_a_RS_S_x [m s^-2] | b_a_RS_S_y [m s^-2] | b_a_RS_S_z [m s^-2] |

IMU.txt

| Timestamp [ns] | w_RS_S_x [rad s^-1] | w_RS_S_y [rad s^-1] | w_RS_S_z [rad s^-1] | a_RS_S_x [m s^-2] | a_RS_S_y [m s^-2] | a_RS_S_z [m s^-2] |
Need downstream help?

Pair the dataset with AI analysis and content workflows.

Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.

Explore AI studio