Small Point-LIO

Small Point-LIO is an advanced implementation of the Point-LIO algorithm, delivering a 2-3x speed improvement over the original.

The default branch is for ROS2. If you want to run it without ros, please checkout the main branch.

If you want to know why it so fast, please read this.

Contact

QQ group: 1070252119

Email: 1709185482@qq.com

Param

here are some parameters you can set in config file:

small_point_lio:
    ros__parameters:
        lidar_topic: /livox/lidar # LiDAR topic name
        imu_topic: /livox/imu # IMU topic name
        lidar_type: livox # Lidar type
        lidar_frame: livox_frame # Lidar frame
        save_pcd: false # Whether to save point cloud

        # Point Cloud Filtering
        point_filter_num: 1 # keep one point every N points
        min_distance: 0.5 # Minimum point cloud radius; points closer than this will be filtered
        max_distance: 1000 # Maximum point cloud radius; points farther than this will be filtered
        space_downsample: true # Whether to enable point cloud downsampling
        space_downsample_leaf_size: 0.5 # Voxel size used for point cloud downsampling

        # IMU Processing
        gravity: [0.0, 0.0, -9.810] # Gravity vector
        fix_gravity_direction: true # Whether to use the first 200 IMU data points to correct gravity direction (magnitude still from gravity parameter)
        check_satu: true # Whether to enable IMU data saturation check
        satu_acc: 3.0 # IMU acceleration saturation threshold
        satu_gyro: 35.0 # IMU angular velocity saturation threshold
        acc_norm: 1.0 # IMU acceleration norm

        # Map
        map_resolution: 0.5 # Map resolution
        init_map_size: 10 # Number of points required to initialize the map

        # LiDAR-IMU Extrinsic Calibration
        extrinsic_est_en: false # Whether to estimate LiDAR-IMU extrinsic transformation online
        extrinsic_T: [-0.011, -0.02329, 0.04412]
        extrinsic_R: [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0]

        # Kalman Filter Parameters
        # R
        laser_point_cov: 0.01 # Laser point covariance
        imu_meas_acc_cov: 0.01 # IMU measured acceleration covariance
        imu_meas_omg_cov: 0.01 # IMU measured angular velocity covariance
        # Q
        velocity_cov: 20.0 # Velocity covariance
        acceleration_cov: 500.0 # Acceleration covariance
        omg_cov: 1000.0 # Angular velocity covariance
        ba_cov: 0.0001 # Acceleration bias covariance
        bg_cov: 0.0001 # Gyroscope bias covariance
        plane_thr: 0.1 # Plane matching threshold (smaller value = stricter)
        match_s: 81.0 # Point-to-plane association threshold (smaller value = stricter)

        # Data Publishing
        publish_odometry_without_downsample: false # Whether to publish high-frequency odometry. Note that this does not enhance the real-time nature of the odometry and but degrades performance. It is recommended to increase the point cloud publishing rate to achieve highly real-time odometry.

Save map

Step 1: set save_pcd to true in config file.

Step 2: run small point lio until the map is finished.

Step 3: save map by calling service:

ros2 service call /map_save std_srvs/srv/Trigger

Note: Please make sure you have enough memery to save map. Don't forget to set save_pcd to false after saving.

Third-party

Small Point-LIO is built in on and with the aid of the following open source projects. Credits are given to these projects.

project	description	license
Eigen	A C++ template library for linear algebra	Mozilla Public License Version 2.0
unordered_dense	A fast & densely stored hashmap and hashset	MIT License
small_gicp	Efficient and parallel algorithms for point cloud registration	MIT License
PCL	Point Cloud Library	BSD License

License

Copyright (C) 2025 Yingjie Huang

Licensed under the MIT License. See License.txt in the project root for license information.