Multi LiDAR Calibrator

This package estimates the extrinsic calibration between two LiDAR point clouds with the Normal Distributions Transform (NDT) algorithm.

The multi_lidar_calibrator node synchronizes parent and child sensor_msgs/msg/PointCloud2 messages, downsamples the child cloud for registration, and publishes the unfiltered child cloud transformed into the parent frame on /points_calibrated.

Supported branches

• ros2 — ROS 2 Jazzy on Ubuntu 24.04; the default branch.
• ros1 — ROS 1 Noetic on Ubuntu 20.04; the maintained legacy branch.

ROS 2 Jazzy

The ROS 2 port targets ROS 2 Jazzy on Ubuntu 24.04 and uses ament_cmake, C++17, PCL, and message_filters approximate-time synchronization.

Build

Create a ROS 2 workspace, install dependencies, build, and source it:

mkdir -p ~/ros2_ws/src
cd ~/ros2_ws/src
git clone https://github.com/hdh7485/multi_lidar_calibrator.git
cd ..
source /opt/ros/jazzy/setup.bash
rosdep update
rosdep install --from-paths src --ignore-src -r -y
colcon build --symlink-install --packages-select multi_lidar_calibrator
source install/setup.bash

Run

The node can be started directly with ROS 2 parameter overrides:

ros2 run multi_lidar_calibrator multi_lidar_calibrator --ros-args \
  -p points_parent_src:=/lidar_parent/points_raw \
  -p points_child_src:=/lidar_child/points_raw \
  -p x:=0.0 -p y:=0.0 -p z:=0.0 \
  -p roll:=0.0 -p pitch:=0.0 -p yaw:=0.0

The converted example launch files preserve the defaults from the ROS 1 XML launch files:

ros2 launch multi_lidar_calibrator multi_lidar_calibrator.launch.py
ros2 launch multi_lidar_calibrator multi_lidar_calibrator_flying_car.launch.py
ros2 launch multi_lidar_calibrator multi_lidar_calibrator_i30.launch.py

Every launch file accepts the same arguments as the original XML files, including points_parent_src, points_child_src, voxel_size, ndt_epsilon, ndt_step_size, ndt_resolution, ndt_iterations, x, y, z, roll, pitch, and yaw. The launch files start the executable with the historical lidar_calibrator launch name; direct execution uses the node name multi_lidar_calibrator.

Provide a useful initial guess for the sensor pair before starting a recording or rosbag. The node reports the estimated transform and prints a ROS 2 static transform command. The command arguments are translation followed by yaw, pitch, roll, parent frame, and child frame:

ros2 run tf2_ros static_transform_publisher \
  <x> <y> <z> <yaw> <pitch> <roll> <parent_frame> <child_frame>

Parameters and topics

Parameter	Type	Default	Description
points_parent_src	string	points_raw	Parent point-cloud topic.
points_child_src	string	points_raw	Child point-cloud topic.
voxel_size	double	0.1	Voxel side length used to downsample only the child cloud before NDT.
ndt_epsilon	double	0.01	NDT transformation epsilon.
ndt_step_size	double	0.1	NDT line-search step size.
ndt_resolution	double	1.0	NDT target resolution.
ndt_iterations	integer	400	Maximum NDT iterations.
x, y, z	double	0.0	Initial translation guess in metres.
roll, pitch, yaw	double	0.0	Initial rotation guess in radians.

The node subscribes to the two configured PointCloud2 topics using approximate time synchronization. It converts messages to PCL PointXYZI, registers the downsampled child cloud against the parent cloud, transforms the original child cloud, and publishes /points_calibrated. The output frame is the parent frame and the output timestamp is the child cloud timestamp. Point count and intensity values are retained from the unfiltered child cloud.

Regression tests

The launch-testing suite verifies both message preservation and registration accuracy. The identity test publishes identical synthetic parent and child clouds and verifies:

• output frame is the parent frame;
• output timestamp is the child timestamp;
• point count and intensity values are preserved; and
• output XYZ values remain near identity.

The known-transform test applies a fixed translation and yaw to an asymmetric synthetic child cloud, starts NDT from a nearby estimate, and verifies that the calibrated output recovers the expected parent cloud within the configured tolerance.

Run it with:

source /opt/ros/jazzy/setup.bash
colcon test --packages-select multi_lidar_calibrator --event-handlers console_direct+
colcon test-result --verbose

Apple container verification

The following commands were used with Apple container on the host. The bind mount is the repository root, and the resource limit avoids the memory pressure seen with an unrestricted ROS 1 build.

ROS 1 baseline (ros1; run this command while that branch is checked out):

container run --rm --memory 6G --cpus 4 \
  --volume "$PWD:/ws/src/multi_lidar_calibrator" \
  --workdir /ws ros:noetic-ros-base-focal bash -lc '
    set -euxo pipefail
    apt-get update
    DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
      build-essential cmake libpcl-dev ros-noetic-pcl-ros \
      ros-noetic-pcl-conversions ros-noetic-message-filters \
      ros-noetic-rostest python3-rosunit
    catkin_make -j2
    source devel/setup.bash
    catkin_make run_tests_multi_lidar_calibrator -j2
    catkin_test_results
  '

ROS 2 Jazzy (ros2):

container run --rm --memory 6G --cpus 4 \
  --volume "$PWD:/ws/src/multi_lidar_calibrator" \
  --workdir /ws ros:jazzy-ros-base bash -lc '
    set -eo pipefail
    apt-get update
    DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
      build-essential libpcl-dev ros-jazzy-pcl-conversions \
      ros-jazzy-message-filters ros-jazzy-launch-testing-ament-cmake \
      ros-jazzy-sensor-msgs-py
    source /opt/ros/jazzy/setup.bash
    colcon build --symlink-install --packages-select multi_lidar_calibrator
    source install/setup.bash
    colcon test --packages-select multi_lidar_calibrator \
      --event-handlers console_direct+
    colcon test-result --verbose
  '

The ROS 1 baseline and ROS 2 port both pass their build and integration-test gates on Linux ARM64 with a 6 GB memory limit.

License

Apache-2.0