Tutorial: Calibrating Fetch

Calibrating your Fetch Research Edition Robot is essential to getting the best performance. Fetch Robotics recommends re-calibrating the robot after every shipping or travel excursion your robot undertakes.

Understanding Calibration

The fetch_calibration package provides a method of making sure that the hand-eye coordination of the robot is well calibrated. Calibration involves:

  • Moving the arm to a series of poses
  • At each pose, recording the joint_angles reported by the drivers. Calibration then blinks the gripper LEDs several times to find their pose in the camera frame and records it.
  • Performing a non-linear optimization which adjusts joint offsets and the head camera pose to minimize the error between the expected pose of the the LEDs based on projection of the kinematics of the arm and the actual pose seen by the camera.
  • Updating the URDF and robot launch files with newly determined offsets.

The Upstart Services that start the robot will use the launch file in /etc/ros/indigo/robot.launch. Therefore, the last step in calibration is to update that launch file and restart the drivers. Currently, the following aspects are updated:

  • The URDF file is copied to /etc/ros/indigo, and the name of the file is put into robot.launch. The calibration offsets are updated in the URDF file.
  • The camera calibration YAML files are copied to /etc/ros/indigo, and the name of the files are put into robot.launch.
  • The head_camera driver has two parameters, z_offset_mm and z_scale which are calibrated. Their updated values are stored in robot.launch.

Calibrate Robot Tool

The fetch_calibration package provides a tool called calibrate_robot which fully automates tasks related to calibration. Once you have sourced the ROS setup.bash, running calibrate_robot without any arguments will show the list of valid arguments:

>$ calibrate_robot
usage: calibrate_robot [-h] [--arm] [--base] [--install] [--reset] [--restore]
                       [--date] [--directory DIRECTORY]

Calibrate the robot, update files in /etc/ros

optional arguments:
  -h, --help            show this help message and exit
  --arm                 Capture arm/head calibration data
  --base                Capture base calibration data
  --install             Install new calibration to /etc/ros (restarts drivers)
  --reset               Reset the calibration to factory defaults (restarts
                        drivers)
  --restore             Restore the previous calibration
  --date                Get the timestamp of the current calibration
  --directory DIRECTORY
                        Directory to load calibration from or backup to

A useful command to note is the –date option, which will return the date that the current calibration was generated (or ‘uncalibrated’ if the robot has not yet been calibrated):

>$ calibrate_robot --date
2015-06-06 23:20:18

Running Calibration

Warning

Calibration will cause the arm to move through the environment. Raise the torso to full extension and be sure that the robot has at least 1 meter of free space all around it.

The first step to calibrate the robot is to reset the calibration to factory defaults:

>$ calibrate_robot --reset

This command might ask for your password, as it requires sudo to update the files in /etc/ros/. This will also restart the drivers to make the changes take effect.

Warning

When calibrating, make sure that no other applications are subscribed to the head_camera topics. Other applications, even RVIZ, connecting to the head_camera driver between the restart of drivers and the start of calibration may adversely affect the auto_exposure settings of the camera.

The robot is now ready to calibrate. The following command will move the arm through a series of poses and upon completion of the calibration will update the robot configuration in /etc/ros/indigo:

>$ calibrate_robot --arm --install

Finally, after the new calibration has been installed into /etc/ros/indigo, you can view the calibration by opening RVIZ and seeing that the robot model overlaps with the point cloud from the head camera. If for some reason, the calibration is not good, you can restart from the –reset command or roll back to the previous calibration with:

>$ calibrate_robot --restore

Calibrating Fetch Torso

When the torso controller board is first powered, it uses measurements from two different position sensors to determine the absolute starting position of the torso. Once the absolute starting position of the torso is determined, the position measurement will retain millimeter precision as long torso remains powered.

To work properly, the two torso sensors must be calibrated together. If the sensors are not properly calibrated, the calculation of the initial torso position could be incorrect in some situations. This problem is more likely to occur if the torso is first powered when in the “up” position. A bad torso position should be easy to detect when using RVIZ since the torso position shown in RVIZ will always be more than +/-10cm different than the true torso position.

The torso sensors are calibrated in production, so they will not usually need to be recalibrated. If there seems to be a torso positioning problem, the torso calibration tool should be first used to verify the calibration of the torso.

In release 0.7.4 of fetch_drivers package there is a tool to verify or calibrate the torso sensors. The tool has two options: verify and calibrate. The verify option will only verify that the calibration is good, it will not change any stored calibration parameters. The calibrate option will calibrate the sensors and update the parameters stored on the torso controller.

For both options, the torso will travel through its entire range of motion while sensor data is collected. While the tool is being run, the robot drivers will be stopped and the robot arm will not hold its position. Because of this, the arm should be tucked or soft fabric or cardboard should be placed between arm and base to avoid scratching any covers.

Torso Calibration Procedure

Warning

During torso calibration the arm will not hold position. Place cardboard or soft fabric between arm and base to avoid scratching covers during data collection.

Follow these steps in order to verify or calibrate the torso position sensors:

  1. Move torso to lowest position, and tuck the arm.
  2. Place a protective barrier between the arm and top base cover.
  3. Disable robot drivers by running : sudo service robot stop
  4. Run torso calibration tool:
  • To run calibration : rosrun fetch_drivers torso_calibrate calibrate
  • To verify calibration : rosrun fetch_drivers torso_calibrate verify
  1. Wait for torso to collect sensor data. The torso will move upwards in small increments through the entire range of motion. A clicking sound will be produced by the torso while moving, and is normal.
  2. Cycle Run-stop (optional). Sometimes tool will request that run-stop be cycled after it completes. Cycling run-stop will cycle power to the torso controller board, and is required in some situations.
  3. On tool has completed, restart robot drivers with sudo service robot start

Verify Output

Verification will produce output stating whether sensors are well calibrated. If the sensor calibration is good the tool will output something similar to:

VERIFY PASSED : max sensor error of 0.0109411 is within acceptable limit

Otherwise it will produce output like:

VERIFY FAILED : max sensor error of 0.0501323 is larger than acceptable limit of 0.04

When verification fails, run calibration produce.

Note

The value for max sensor error is the mismatch between the two torso sensors. The accuracy of the torso position measurement is unrelated to this value.

Calibrate Output

Once the tool has completed the calibration procedure it will check the expected results of calibration. If everything checks out, the tool will output something similar to:

VERIFY PASSED : max sensor error of 0.0116824 is within acceptable limit

If there was a problem calculating good calibration parameters, the output might look like:

VERIFY FAILED : max sensor error of 0.0501323 is larger than acceptable limit of 0.04

In case of failure, the torso sensor may be malfunctioning or damaged and a support ticket should be created.