Using ABB's Absolute Accuracy (603-1) Within a URDF

[jbeck28]

ABB sells a feature, Absolute Accuracy, which involves some calibration routine they perform. I've found what I believe are the values that came with my robot (which has AbsAcc) inside of a MOC.cfg that appears inside of backups. Here's the relevant data:

PARALLEL_ARM_CALIB:
 
       -name "rob1_2" -error_length 0.000251949
 
       -name "rob1_3" -error_length 5.31822E-05
 
 ARM_CHECK_POINT:
 
       -name "rob1_3" -position_x -0.217 -position_y 0 -position_z 0.2154
 
 ARM_LOAD:
 
       -name "r1_load_1"
 
       -name "r1_load_2"
 
       -name "r1_load_3"
 
       -name "r1_load_4"
 
 ARM_CALIB:
 
       -name "rob1_1" -error_offset_x 2.76348E-05 -error_offset_y 9.85993E-05\
       -error_offset_z -0.000325266 -error_roll 8.72513E-05\
       -error_pitch 0.00078713 -error_jaw 0.000384148
 
       -name "rob1_2" -error_offset_x -0.000104688 -error_roll -8.16564E-05\
       -error_pitch -0.000354871
 
       -name "rob1_3" -error_offset_z 1.49512E-06 -error_roll 0.000220894\
       -error_pitch 0.00308007 -error_jaw 0.000179515
 
       -name "rob1_4" -error_offset_y -0.000165182 -error_offset_z 0.00011376\
       -error_roll -0.000466978 -error_jaw -0.000281756
 
       -name "rob1_5" -error_offset_x -0.000169532 -error_offset_z 0.00021558\
       -error_pitch 2.60882E-05 -error_jaw 6.11623E-05
 
       -name "rob1_6" -error_offset_y 4.64985E-05 -error_offset_z 5.62064E-05\
       -error_roll -0.00067455 -error_jaw -8.66063E-05
 
 JOINT_CALIB:
 
       -name "rob1_1"
 
       -name "rob1_2" -Compl 2.12623E-06
 
       -name "rob1_3" -Compl 9.58023E-06
 
       -name "rob1_4" -Compl 4.50031E-05
 
       -name "rob1_5" -Compl 5.08124E-05
 
       -name "rob1_6" -Compl 0.000143289

The question, if you will, is how properly to apply these? It seemed at first glance to be as easy as applying these to the joints within the URDF. However, this does not give me a match with EEF locations between robot controller and ROS - they differ by roughly a millimeter. I then stumbled upon this, and realized the roll pitch and yaw might not be what I think they are in the URDF. I manually rearranged RPY to match the frames given in that diagram, and got a very different result that a similar order of magnitude of wrongness.

I've got a ticket open with ABB to see if they're willing/able to help, I'll update this once I hear back either way.

Thanks again,
Josh

[gavanderhoorn]

I've got a ticket open with ABB to see if they're willing/able to help, I'll update this once I hear back either way.

@jbeck28: did ABB get back to you?

[radhen]

@gavanderhoorn @jbeck28 I am trying to figure out the same thing. How can I use the parameters found in MOC.cfg into my ROS URDF file for ABB. Please respond to this thread if you have an update.

[jbeck28]

Sorry I never replied to this, no idea how it slipped under my radar. ABB was somewhat helpful but not to the point that I could figure out how to apply the transformations, and I no longer have access to ABB robots.

What I ended up doing - and this proved to work reasonably well - was to use robot_calibration. Basically, use (via RWS) the pose returned by the robot controller (using absolute accuracy) as the sensor, and populate a plane of points at each pose as the captured data. This was a while ago so I'm probably goofing up the verbiage a bit. This got us <100 microns difference between what ROS returns for EEF pose and what the robot controller did. A caveat is that it won't handle payload deflection at all, you're probably best off setting the payload to zero (-1kg if I recall, in ABB) while collecting data. The same idea worked for Kuka, who were far less helpful than ABB.

[radhen]

@jbeck28 thank you so much from the prompt reply. I checked out robot_calibration package briefly. My understanding is the package allows for camera-to-robot calibration, but probably also kinematic calibration of the robot, like what you did. Is that correct?

[radhen]

Hi @jbeck28 and @gavanderhoorn,

I wanted to share some information I came across in an issue from the robot_calibration repository that might be helpful to you or others encountering a similar challenge.

I am trying to perform robot kinematic calibration to determine offsets in joint positions (XYZ) and orientations (RPY), using ground truth poses from a laser tracker and poses from an uncalibrated robot. The goal is to optimize the XYZ and RPY values and update the URDF file accordingly, minimizing errors in the tool center poses.

From my research, it seems the robot_calibration package doesn’t currently provide an example for this type of calibration. This is based on the discussion in the issue titled Robot URDF calibration using touch probes and reference geometry. Is it possible? #157.

If anyone else reading this knows of other GitHub packages that support this functionality, I’d greatly appreciate your input. Thank you for your time and any guidance you can offer!