FAST-LIO-LOCALIZATION

A simple localization framework that can re-localize in built maps based on FAST-LIO.

News

2021-08-11: Add Open3D 0.7 support.
2021-08-09: Migrate to Open3D for better performance.

1. Features

Realtime 3D global localization in a pre-built point cloud map. By fusing low-frequency global localization (about 0.5~0.2Hz), and high-frequency odometry from FAST-LIO, the entire system is computationally efficient.

Eliminate the accumulative error of the odometry.

The initial localization can be provided either by rough manual estimation from RVIZ, or pose from another sensor/algorithm.

2. Prerequisites

2.1 Dependencies for FAST-LIO

Technically, if you have built and run FAST-LIO before, you may skip section 2.1.

This part of dependency is consistent with FAST-LIO, please refer to the documentation https://github.com/hku-mars/FAST_LIO#1-prerequisites

2.2 Dependencies for localization module

python 2.7
ros_numpy

sudo apt install ros-$ROS_DISTRO-ros-numpy

Open3D

pip install open3d==0.9

Notice that, there may be issue when installing Open3D directly using pip in Python2.7:

ERROR: Package 'pyrsistent' requires a different Python: 2.7.18 not in '>=3.5'

you may firstly install pyrsistent:

pip install pyrsistent==0.15

Then

pip install open3d==0.9

3. Build

Clone the repository and catkin_make:

    cd ~/$A_ROS_DIR$/src
    git clone https://github.com/HViktorTsoi/FAST_LIO_LOCALIZATION.git
    cd FAST_LIO_LOCALIZATION
    git submodule update --init
    cd ../..
    catkin_make
    source devel/setup.bash

Remember to source the livox_ros_driver before build (follow livox_ros_driver)
If you want to use a custom build of PCL, add the following line to ~/.bashrc export PCL_ROOT={CUSTOM_PCL_PATH}

4. Run Localization

4.1 Sample Dataset

Demo rosbag in a large underground garage: Google Drive | Baidu Pan (Code: ne8d);

Corresponding map: Google Drive | Baidu Pan (Code: kw6f)

The map can be built using LIO-SAM or FAST-LIO-SLAM.

4.2 Run

First, please make sure you're using the Python 2.7 environment;
Run localization, here we take Livox AVIA as an example:

roslaunch fast_lio_localization localization_avia.launch map:=/path/to/your/map.pcd

Please modify /path/to/your/map.pcd to your own map point cloud file path.

Wait for 3~5 seconds until the map cloud shows up in RVIZ;

If you are testing with the sample rosbag data:

rosbag play localization_test_scene_1.bag

Or if you are running realtime

roslaunch livox_ros_driver livox_lidar_msg.launch

Please set the publish_freq in livox_lidar_rviz.launch to 10Hz, to ensure there are enough points for global localization in a single scan. Support for higher frequency is coming soon.

Provide initial pose

rosrun fast_lio_localization publish_initial_pose.py 14.5 -7.5 0 -0.25 0 0

The numerical value 14.5 -7.5 0 -0.25 0 0 denotes 6D pose x y z yaw pitch roll in map frame, which is a rough initial guess for localization_test_scene_1.bag.

The initial guess can also be provided by the '2D Pose Estimate' Tool in RVIZ.

Note that, during the initialization stage, it's better to keep the robot still. Or if you play bags, fistly play the bag for about 0.5s, and then pause the bag until the initialization succeed.

Related Works

FAST-LIO: A computationally efficient and robust LiDAR-inertial odometry (LIO) package
ikd-Tree: A state-of-art dynamic KD-Tree for 3D kNN search.
FAST-LIO-SLAM: The integration of FAST-LIO with Scan-Context loop closure module.
LIO-SAM_based_relocalization: A simple system that can relocalize a robot on a built map based on LIO-SAM.

Acknowledgments

Thanks for the authors of FAST-LIO and LIO-SAM_based_relocalization.

TODO

Go over the timestamp issue of the published odometry and tf;
Using integrated points for global localization;
Fuse global localization with the state estimation of FAST-LIO, and smooth the localization trajectory;
Updating...

wing0716/FAST_LIO_LOCALIZATION