Dataset assetOpen Source CommunityMulti‑Sensor FusionNavigation Algorithms

GREAT Dataset

The GREAT dataset is a novel multi‑sensor (stereo camera, LiDAR, IMU) lunar surface dataset collected from a vehicle‑mounted platform in complex urban environments. It includes high‑precision multi‑frequency GNSS observations, tactical‑grade IMU, MEMS IMU, two CMOS cameras, and LiDAR. All sensors are hardware‑synchronized and well‑calibrated. The dataset comprises eight sequences covering the Wuhan University campus and surrounding urban/suburban areas, intended for evaluating multi‑sensor fusion navigation algorithms.

Source

github

Created

Nov 7, 2024

Updated

Nov 8, 2024

Signals

840 views

Availability

Linked source ready

Overview

Dataset description and usage context

GREAT DATASET

Introduction

Abstract

GREAT Dataset is a new multi‑sensor raw observation dataset collected from a vehicle in complex urban environments, featuring high‑precision multi‑frequency GNSS receivers, tactical‑grade IMU, MEMS IMU, two CMOS cameras, and LiDAR. All sensors are hardware‑synchronized and well‑calibrated. The dataset includes eight sequences covering the Wuhan University campus and nearby urban/suburban areas, supporting evaluation of various multi‑sensor fusion navigation algorithms.

Main Contributions

Provides raw GNSS observations, IMU measurements, camera images, and LiDAR scans with timestamps unified to GPS time via hardware sync.
Captures diverse environments (campus, urban canyon, suburb) to thoroughly assess SLAM robustness and accuracy.
Supplies high‑precision raw GNSS and inertial navigation data for robotics, SLAM, and satellite navigation research.

License

The dataset is released under the MIT License for academic use only. Commercial use or collaborations require contacting xingkonggreat@163.com.

Sensor Setup

Platform

Details and diagrams are provided in the original documentation.

Sensors

Sensor	Model	Frequency	Details
GNSS Receiver	Septentrio PolaRx5	1 Hz	Multi‑frequency, multi‑system (GPS, GLONASS, GALILEO, BDS, IRNSS, QZSS)
Tactical‑grade IMU	StarNeto XW‑GI7660	200 Hz	Gyro bias 0.3 (°/h), accel bias 100 mGal, angular random walk 0.01 (°/√h)
MEMS‑IMU	ADIS‑16470	100 Hz	Gyro bias 8 (°/h), accel bias 1500 mGal, angular random walk 0.34 (°/√h), velocity random walk 0.037 (m/s/√h)
Camera	FLIR BFS‑PGE‑31S4C	20 Hz	Sony IMX265 sensor, global shutter, PoE GigE, max resolution 2048 × 1536
LiDAR	Velodyne VLP‑16	10 Hz	Vertical FOV 30° (+15° ~ ‑15°), vertical angular res 2°, 360° horizontal FOV, max range 100 m

Comparison with Other Datasets

Dataset	GNSS Raw	Multi‑frequency	MEMS IMU	Tactical IMU	LiDAR	Camera	Hardware Sync
WHU‑Helmet	×	√	√	×	√	√	√
SubT‑MRS	×	×	√	×	√	√	√
GEODE	×	√	√	×	√	√	√
KITTI	×	√	×	√	√	√	×
Hilti SLAM	×	×	√	×	√	√	√
RobotCar	×	×	√	×	√	√	×
M2DGR	√	×	√	×	√	√	×
Brno Urban	×	√	√	×	√	√	√
SmartPNT‑POS	√	√	×	√	×	×	√
OURS	√	√	√	√	√	√	√

Environment

Eight sequences are provided, including four day‑time campus runs, one night‑time campus run, two urban canyon runs, and two suburban runs. Sample images illustrate each environment.

Data Sequences

Sequence	Date	Buildings	Dense Trees	Overpasses/Tunnels	Dynamic Vehicles	Dynamic Pedestrians	GNSS Observations	Duration (s)	Difficulty
campus‑01	2020/10/27	√	√			√	√	606.85	Medium
campus‑02	2020/10/27	√	√			√	√	806.9	Medium
campus‑03	2020/10/27	√	√			√	√	1200.1	Hard
campus‑night	2020/10/29	√	√			√	√	545.7	Hard
suburb‑01	2020/10/29		√				√	1081.35	Easy
suburb‑02	2020/10/29		√		√		√	837	Easy
urban‑01	2020/10/27			√	√		√	767.3	Medium
urban‑02	2022/10/23			√	√		√	1622.2	Hard

Download

Sequence	IMU	Images	LiDAR	Reference Solution
campus‑01	IMU.zip	Image.zip	LiDAR.zip	Reference
campus‑02	...	...	...	...
...	...	...	...	...

Reference Solutions

Reference trajectories for each sequence have been plotted in Google Earth. Figures 7‑14 in the original document illustrate these trajectories.

Directory Structure

Raw GNSS observations and ephemerides are stored in the GNSS_RAW_DATA folder. Each sequence folder contains IMU data, stereo images, LiDAR scans, and reference solutions.

Development Toolkit

A script is provided to convert raw visual‑LiDAR observations to ROS bag files. The tool requires a ROS environment and necessary dependencies.

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