Skip to content

Latest commit

 

History

History
150 lines (110 loc) · 4.82 KB

README-calibration.md

File metadata and controls

150 lines (110 loc) · 4.82 KB
Raw

Calibration

This file explains how to plan, execute motions that move the camera in front of a chessboard, and how to collect calibration data. We then propose two calibration procedure

  • hand eye based with VIsP to compute the camera pose in the end-effector frame,
  • full calibration based on VIsP and figaroh that besides the above computes the geometric parameters of the kinematic chain of the robot.

Before starting, place a chessboard of 10 by 7 squares of size 2.7cm on a vertical plane in front of the robot (1.3m ahead). The center of the chessboardat should be at a height of 1.41m.

Remove the wooden spike since the chessboard should not be occluded at all.

Generating and executing motions

Start the demo with the following differences:

  • no need to start react_inria node,
  • roslaunch demo part:=april-tag-plank,

Move the robot to the "calib" configuration.

In the terminal where script_hpp.py has been run, copy paste the following lines.

q_init[-7:] = [1.3, 0, 0, 0, 0, 0, 1]
from calibration import Calibration, generateDataForFigaroh, checkData

calibration = Calibration(ps, graph, factory)
calibration.addStateToConstraintGraph()
calibration.generateConfigurationsAndPaths(q_init, filename="./data/optimal-configs.csv")

If no path appears in the "Path Player" of gepetto-gui after refreshing, re-execute the last line of the later script after increasing "calibration.nbConfigs".

in a new terminal, cd into ur10/pointing and run

python -i play_path.py

In the same terminal, run

playAllPaths(0)

The robot should execute the paths and collect the required data at each end of path. After the last path has been executed,

mkdir measurements

and run

cc.save('./measurements')

The procedure can be run several times if not enough data has been acquired the first time. For that, it is recommended

  • to go back to the "calib" configuration,
  • to save the data in another directory to avoid overwriting the previous data.

Hand eye calibration

Follow the instructions here https://visp-doc.inria.fr/doxygen/visp-daily/tutorial-calibration-extrinsic.html.

Updating the model

The calibration procedure computes the pose of the optical frame "camera_color_optical_frame" in the end-effector link, here "ref_camera_link". The problem is that this transformation is hard-coded in the camera model. We therefore need to compute a new transform between "ur10e_d435_mount_link" and "ref_camera_link" in such a way that the pose of the "camera_color_optical_frame" in "ur10e_d435_mount_link" is correct. For that, we denote by

  • m ur10e_d435_mount_link,
  • c camera_color_optical_frame,
  • e ref_camera_link (end effector).

The desired value of mMc is given by:

mMc_desired = mMe * eMc_measured

where mMe is the pose of "ref_camera_link" in "ur10e_d435_mount_link". eMc_measured is the result of hand eye calibration. As explained above, eMc is provided by the camera urdf model that we do not want to modify. We therefore need to modify mMe into mMe_new to get the same mMc_desired:

mMc_desired = mMe_new * eMc_provided

Therefore

mMe_new = mMe * eMc_measured * eMc_provided.inverse()

Function computeCameraPose in calibration.py makes this calculation.

Full calibration

Install figaroh

cd $DEVEL_HPP_DIR/src
make figaroh.install

Compute the camera poses with respect to the chessboard:

$DEVEL_HPP_DIR/src/visp/build-rel/tutorial/calibration/tutorial-chessboard-pose --square_size 0.027 --input image-%d.png --intrinsic camera.xml --output pose_cPo_%d.yaml

See https://visp-doc.inria.fr/doxygen/visp-daily/tutorial-calibration-extrinsic.html for details.

In the python terminal where script_hpp.py has been executed, type

generateDataForFigaroh(robot, './measurements', output_file, 30)

where output_file is the full path to examples/ur10/data/calibration.csv in figaroh source directory.

Compute the kinematic parameters

cd $DEVEL_HPP_DIR/src/figaroh/examples/ur10
python -i calibration.py

Several windows appear that show the result of the optimization procedure. Close the windows by typing 'q'. Then in the python terminal, type

from update_model import update_parameters
update_parameters(f_input, f_output, LM_solve.x, param)

where

  • f_input is the full path to the current kinematic parameters of the robot: $DEVEL_HPP_DIR/src/agimus-demos/ur10/pointing/config/calibrated-params.yaml,
  • f_output is the name of a file in the same directory, for example updated-params.yaml. Compare the two files to check that the differences are relevant. If so, copy the later into the former, commit, push and reinstall agimus-demos.

Compute the camera pose

The last 6 components of LM_solve.x represent the pose of camera_color_optical_frame in wrist_3_link: (x,y,z,roll,pitch,yaw).