EIKF-VIO-LIO

(Currently, only the necessary steps for using the code are observable. Full open access will be enabled after the review.)

Table of Contents

1. Prerequisites
2. Installation
3. Run on datasets
4. Evaluation
5. Acknowledgement
6. Citation

Prerequisites

1. Livox_ros_driver

   follow https://github.com/Livox-SDK/livox_ros_driver

Installation

1. Create ROS workspace

   mkdir -p ~/EIKF_LIO_VIO/src
   cd EIKF_LIO_VIO/src

2. Clone the repository and build

   git clone git@github.com:LIAS-CUHKSZ/EIKF-VIO-LIO.git
   cd ..
   catkin_make

Run

We have tested the following datasets, which config file can be directly used. We only support the dataset type: rosbag containing lidar, IMU and camera measurements.

Datasets	Dataset link	Launch file	data preprocess	evaluation method	evaluation script
URCA	https://advdataset2019.wixsite.com/urbanloco/california	URCA.launch	Rosbag with all sensors is provided, you can directly use.	evo	UR_evaluation.sh
URHK	https://advdataset2019.wixsite.com/urbanloco/hong-kong	urbanloco.launch	Rosbag with all sensors is provided, you can directly use.	evo	UR_evaluation.sh
NTU Viral	https://ntu-aris.github.io/ntu_viral_dataset/	NTU_VIRAL.launch	Rosbag with all sensors is provided, you can directly use.	https://ntu-aris.github.io/ntu_viral_dataset/evaluation_tutorial.html	https://ntu-aris.github.io/ntu_viral_dataset/evaluation_tutorial.html
NTU MCD	https://mcdviral.github.io/	NTU_Viral2.launch	Need to merge rosbag with different topics. You can refer to : https://github.com/LIAS-CUHKSZ/DataProcessTools4SLAM/blob/main/merge_rosbag.py	evo	evaluate_ntu2.sh
private dataset	Our hand-held device	ilive_mid.launch	Rosbag with all sensors is provided, you can directly use.	evo	lab_evaluation.sh

• You can run the evaluation script directly. Evo is already included in the script.
• evo: https://michaelgrupp.github.io/evo/doc/performance.html
• Evaluation metric we used in evo: root-mean-square-error (rmse) of absolute pose error (ape) with translation part (evo_ape), with unit (m).

1.1 Download datasets and remember its path and the name of rosbag .

1.2 Modify the corresponding launch file.

<!-- what ros bag to play -->
<arg name="dataset"   default="<Your BagName>" /> 
<arg name="bag_path"   default="<Your BagPath>" />  
<!-- where to save the recorded poses -->
<arg name="path_save"   default="<Your SavePath>" /> 

1.3 Run the code

source devel/setup.bash
roslaunch ilive <Your launch file>

Evaluation

If you want to get evaluation result, please run the corresponding evaluation script.

Acknowledgement

We thank the authors of the following repositories for their open-source code:

OpenVINS: https://docs.openvins.com/

Fast-LIO: https://github.com/hku-mars/FAST_LIO

R3live: https://github.com/hku-mars/r3live

Citation

@misc{li2024efficient,
      title={Efficient Invariant Kalman Filter for Inertial-based Odometry with Large-sample Environmental Measurements}, 
      author={Xinghan Li and Haoying Li and Guangyang Zeng and Qingcheng Zeng and Xiaoqiang Ren and Chao Yang and Junfeng Wu},
      year={2024},
      eprint={2402.05003},
      archivePrefix={arXiv},
      primaryClass={cs.RO}
}