LE-VINS: INS-Centric Visual-Inertial Navigation System With LiDAR Enhancement

This repository provides a modified version of IC-GVINS that supports pure VINS. Besides, we also implement the LiDAR-enhanced method in LE-VINS into this repository.

Authors: Hailiang Tang, Xiaoji Niu, and Tisheng Zhang from the Integrated and Intelligent Navigation (i2Nav) Group, Wuhan University.

Related Paper:

• Hailiang Tang, Tisheng Zhang, Xiaoji Niu, Liqiang Wang, and Jingnan Liu, "LE-VINS: A Robust Solid-State-LiDAR-Enhanced Visual-Inertial Navigation System for Low-Speed Robots," IEEE Transactions on Instrumentation and Measurement, 2023.
• Xiaoji Niu, Hailiang Tang, Tisheng Zhang, Jing Fan, and Jingnan Liu, “IC-GVINS: A Robust, Real-time, INS-Centric GNSS-Visual-Inertial Navigation System,” IEEE Robotics and Automation Letters, 2023.
• Hailiang Tang, Tisheng Zhang, Xiaoji Niu, Jing Fan, and Jingnan Liu, “Impact of the Earth Rotation Compensation on MEMS-IMU Preintegration of Factor Graph Optimization,” IEEE Sensors Journal, 2022.

Contacts:

• For any technique problem, you can send an email to Dr. Hailiang Tang (thl@whu.edu.cn).
• For Chinese users, we also provide a QQ group (481173293) for discussion. You are required to provide your organization and name.

1 Prerequisites

1.1 System and compiler

We recommend you use Ubuntu 18.04 or Ubuntu 20.04 with the newest compiler (gcc>=8.0 or clang>=6.0).

# gcc-8
sudo apt install gcc-8 g++-8

# Clang
# sudo apt install clang

1.2 Robot Operating System (ROS)

Follow ROS Melodic installation instructions for Ubuntu 18.04 and ROS Noetic installation instructions for Ubuntu 20.04.

1.3 OpenCV

The supported version is OpenCV (>=3.2.0). You can install OpenCV from your system repository or build from the source code. OpenCV 4 is also supported.

1.4 oneTBB

Threading Building Blocks (TBB) are used for parallel processing. We recommend you use oneTBB, and install the latest released version. You should install oneTBB before Ceres Solver.

1.5 Ceres Solver with its Dependencies

We use Ceres Solver (>=2.1.0) to solve the non-linear least squares problem. Please follow Ceres installation instructions.

The dependencies Eigen (>=3.3.7), glog (>=0.4.0) are also used. You can install them as follows:

sudo apt install libeigen3-dev libgoogle-glog-dev

1.6 yaml-cpp

The yaml-cpp is employed for reading configurations. It can be installed as:

sudo apt install libyaml-cpp-dev

2 Build and run LE-VINS

2.1 Build the source code

# Make workspace directory
mkdir ~/workspace && cd ~/workspace
mkdir src && cd src

# Clone the repository into src directory
git clone https://github.com/i2Nav-WHU/LE-VINS.git

# To workspace directory
cd ..

# Build the source code using catkin_make
catkin_make -j8 -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=gcc-8 -DCMAKE_CXX_COMPILER=g++-8

2.2 Run demo dataset

If you have already downloaded the open-sourced dataset, run the following commands.

# Open a terminal and source the workspace environments
# For bash
source ~/workspace/devel/setup.bash
# For zsh
# source ~/workspace/devel/setup.zsh

# Run LE-VINS node
# 1. Download the dataset.
# 2. Modify the configurations in le_vins.yaml.
# 3. Change the path in the follwoing command.
# 4. Run the follwoing command.
roslaunch le_vins le_vins_read.launch configfile:=path/le_vins.yaml bagfile:=path/rosbag.bag

3 Datasets

3.1 Format

Please be careful the following notes:

• The IMU should be in the front-right-down format in LE-VINS.
• We support Livox, Velodyne, and Ouster LiDARs, but they should in correct formats.
• CompressedImage messages is also support, and you can enable it by modifying the configuration file.

3.2 IC-GVINS Robot Dataset

You can run LE-VINS in visual-inertial mode using the open-sourced IC-GVINS dataset.

3.3 R3LIVE Handheld Dataset

The R3LIVE dataset is supported. However, you should be careful about the following issues:

• You should change the IMU data into the front-right-down format.
• The image message is with CompressedImage format.
• The camera and IMU are not well synchronized, and the initial time-delay parameter is every large in some sequences. We have provided the initial time-delay parameters for each sequence in the configuration file.

3.4 FusionPortableV2 Vehicle Dataset

The FusionPortableV2 dataset is supported, and the vehicle sequences have been fully tested. The STIM-300 IMU, the left camera, and the Ouster LiDAR are employed.

• You should change the IMU data into the front-right-down format.
• The sensors are not well synchronized. The time-delay parameter for camera-IMU can be estimated online in LE-VINS, while the time-delay parameter for LiDAR-IMU are not estimated. Hence, you should modify the time-delay parameter of LiDAR-IMU for each sequences in the configuration file.

3.5 Your own dataset

You can run LE-VINS with your self-collected dataset. Keep in mind the following notes:

1. You should prepare dataset in a ROS bag
2. The IMU data should be in the front-right-down format
3. Modify the topic names in the configuration file
4. Modify the parameters in the configuration file

3.6 Evaluation

We use evo to evaluate the TUM trajectory files. We also provide some useful scripts (evaluate_odometry) for evaluation.

4 Acknowledgements

We thanks the following projects for the helps in developing and evaluating the LE-VINS:

• IC-GVINS: A Robust, Real-time, INS-Centric GNSS-Visual-Inertial Navigation System
• VINS-Fusion: An optimization-based multi-sensor state estimator
• OB_GINS: An Optimization-Based GNSS/INS Integrated Navigation System
• evo: Python package for the evaluation of odometry and SLAM

5 License

The source code is released under GPLv3 license.

We are still working on improving the codes. For any technical issues, please contact Dr. Hailiang Tang (thl@whu.edu.cn) or open an issue at this repository.

For commercial usage, please contact Prof. Xiaoji Niu (xjniu@whu.edu.cn).