ArcosWiki

WARNING! This tutorial assumes that you have an advanced knowledge of: git, linux, ROS1, YARP, Python, KUKA arm, FRI, Robot Kinematics

This tutorial was created by Daniel García Vaglio. If you have any questions about this tutorial or about the calibration process ask him.

• You need to install ROS1 and YARP.
• Clone repository oms-cylinder and follow the README for installation instructions
• Install the Humanoid Robot Simulator
• Clone the keyboard-cart-control repository in ~/local/src
• Install and setup FRI
• Attach the calibration board to the robot.

TODO: Add an explanation of the process and why we do things the way we do them

• Turn on the KUKA arm and set it up with FRI for being used from a Linux Computer.
• Execute a YARP server
• Continue until the bridge and vfclik are running and talking to the robot using FRI.
• Go to ~/local/src/keyboard_cart_control, and execute the following command:

python keyboard_cart_control.py -n /arcosbot-real

• Execute the realsense viewer with the following command:

realsense-viewer

• Move the arm so that the calibration board is inside the field of view of the camera.
• Press i to store this first pose.
• press y to activate automatic pose storage.
• Move the arm in such a way that the calibration board is always visible to the camera. Be careful not to reach singularities.
• Press u to see how many poses have been stored. We want around 2500 poses.
• When you are done, press o to save the poses to a csv file.
• Close keyboard_cart_control.py
• Move the generated file to the following directory:

mv <poses-filename>.csv  <catkin_workspace>/src/oms-cylinder/camera2png/data

• Unplug and re-plug the realsense from the computer.
• open realsense-viewer
• Disable the following options:
  1. Enable auto Exposure
  2. Enable auto white Balance
  3. Auto Exposure Priority
• Close realsense-viewer
• Remember to source tour workspace before executing the following commands.
• Now we need to launch the camera node:

cd <catkin_workspace>/src/oms-cylinder/oms_launcher/launch/
roslaunch rs_camera.launch

• In a separate terminal execute the script that will create the photos for calibration

cd <catkin_workspace>/src/oms-cylinder/camera2png/data
mkdri outpout
cd <catkin_workspace>/src/oms-cylinder/camera2png/scripts/
python camera2png.py ../data/example.py ../data/<poses-file-name>.csv ../data/output

Sometimes the KUKA robot stops working (we are not sure why). When that happens, you will have to:

• Kill the camera2png script
• Re-enable the arm
• check in <catkin_workspace>/src/oms-cylinder/camera2png/data which was the last generated photo. Lets say it was N.
• Restart the camera2png script like this:

python camera2png.py ../data/example.py ../data/<poses-file-name>.csv ../data/output --start <N+1>

• First we have to measure the transformations between the kinematic arm base to the camera and from the wrist to the calibration board. This will be used as the initialization guess for the calibration program to use. The guess has to be given in translation vector(x, y, z) and orientation quaternion(x, y, z, w). This configuration has to be stored in YAML format in the following file: <catkin_ws>/src/oms-cylinder/oms_launcher/cal/configs/static_camera_guesses.yml. Here you can find an example (Do NOT use this as your guess).

base_to_camera_guess:
  x: 1.04
  y: 1.1
  z: 0.0
  qx: 0.3
  qy: 0.4
  qz: 0.5
  qw: 0.01
  
wrist_to_target_guess:
  x: 0.109
  y: 0.2
  z: -0.05
  qx: -0.-4
  qy: 0.5
  qz: 0.5
  qw: -0.4

• You also need to get the camera intrinsics and store them in another YAML. Here you can find and example:

intrinsics:
  fx: 1387.6160888671875
  fy: 1387.5479736328125
  cx: 943.9945678710938
  cy: 561.1880493164062