Minimal Hand-Eye Calibration Node for ROS 2

The minimal_handeye_ros2 package offers a minimal hand-eye calibration node for ROS 2, featuring:

• Capture of transformation data using TF.
• Solutions to both eye-in-hand and eye-to-hand calibration problems.
• Output of calibration results in multiple formats including homogeneous transformation matrix, rotation matrix, quaternion, roll-pitch-yaw.
• (Optional) Broadcasting of calibration results to TF.

Comparison of this package with the popular easy_handeye:

Feature	easy_handeye	minimal_handeye_ros2
ROS Version	Originally developed for ROS 1, there is a ROS 2 port but development is slow	ROS 2
Solver	OpenCV's implementation of the Tasi-Lenz algorithm	Self-implemented solver based on the paper by Faugeras and Hebert, 1986
GUI	Yes	No. Uses ROS 2 commandline interfaces for interaction
Autonomous Robot Movement	Yes	No

For details on the self-implemented solver, please refer to the PDF documentation.

Hand-Eye Calibration

This package supports two types of hand-eye calibration problems: eye-in-hand and eye-to-hand.

Eye-in-hand calibration	Eye-to-hand calibration
Goal: solve the transformation from robot end-effector frame to twist-mounted camera frame	Goal: solve the transformation from robot base frame to external camera frame

Prerequisites

This package assumes that the TF between the robot base frame and the robot end-effector frame and the TF between the camera frame and marker frame are being published. If the TFs are not available, the node will crash when you attempt to capture transformation data.

Getting Started

1. Clone this package in your workspace. Taking ~/handeye_ws as an example:

   mkdir -p ~/handeye_ws/src
   cd ~/handeye_ws/src
   git clone https://github.com/zixingjiang/minimal_handeye_ros2.git

2. Install dependencies.

   cd ~/handeye_ws
   rosdep update --rosdistro=$ROS_DISTRO
   sudo apt update
   rosdep install --from-paths ./ -i -y --rosdistro ${ROS_DISTRO}

3. Build the package.

   colcon build

4. Run the calibration node. You can specify the calibration type, TF frames, calibration tolerances as parameters. Please check this table for more details.

   source install/setup.bash
   
   # Fill the parameters with your setup
   ros2 run minimal_handeye_ros2 handeye_calibration_node --ros-args \
     -p calibration_type:=eye-to-hand \
     -p robot_base_frame:=<robot_base_frame> \
     -p robot_ee_frame:=<robot_ee_frame> \
     -p camera_frame:=<camera_frame> \
     -p marker_frame:=<marker_frame> \
     -p broadcast_tf:=true \
     -p rotation_tolerance:=0.005 \
     -p translation_tolerance:=0.005

5. Capture transformation data. Move the robot to different poses and capture transformation data using the following ros2service call:

   ros2 service call /calibrate std_srvs/srv/Trigger

   For each service call, the node will query the TF tree for transformation data and check if the data is valid. When more than two valid data points are collected, the node will automatically calculate the hand-eye calibration at every service call.
   Note: Please ensure the TF data is stable and robot is still when triggering each capture.
   Note: Please make sure that two adjacent robot poses have different rotations.

6. Check the calibration results. When a converged calibration result is obtained (usually 6 valid data points are required), the node will output the calibration result in the server log and the client service response message. If you have set broadcast_tf:=true, the calibration result will also be broadcasted to TF.

Parameters

Parameter	Description	Default Value
calibration_type	Type of hand-eye calibration, options: eye-to-hand, eye-in-hand	eye-to-hand
robot_base_frame	TF frame name for the robot base	robot_base_frame
robot_ee_frame	TF frame name of the robot end-effector	robot_ee_frame
camera_frame	TF frame name of the camera	camera_frame
marker_frame	TF frame name of the marker	marker_frame
broadcast_tf	Whether to broadcast the calibration result to TF, options: true, false	false
rotation_tolerance	Tolerance for the convergence of the rotation part of the calibration (quaternion coefficients)	0.005
translation_tolerance	Tolerance for the convergence of the translation part of the calibration (unit: m)	0.005

Gimbal Lock

For the roll-pitch-yaw output, when the pitch angle is ±90 degrees (approximately ±1.5707 radians), the roll and yaw angles are not unique due to the gimbal lock. In this case, the roll and yaw angles are set to zero to avoid ambiguity.