industrial_calibration
industrial_calibration copied to clipboard
ros-industrial
WIP: added ganty calibration
new calibration type to calibrate a robot to the gantry upon which its mounted
I haven't read (all of) the code yet. Could you provide a short description of what this calibrates exactly? Is the visual info used to calibrate precise robot mounting on the gantry, or is it doing something else?
This calibration now works, but I don't feel it is a useful thing to do. I'd hoped it would isolate kinematic alignment errors between the robot and the gantry. It certainly does not to the extent that I'd hoped. In fact I've convinced my self that it only correctly refines 2DOF. When I get time to re-visit this issue, I'll investigate. Is it correct policy to withdraw, or let the wip PR become stale?@gavanderhoorn
I would suggest to close the PR.
It will always be listed in the closed section of the PR listing, and even if you remove the branch from your fork of industrial_calibration, the PR will contain the commits (so you could always recover them if you'd want to later).
I'd hoped it would isolate kinematic alignment errors between the robot and the gantry. It certainly does not to the extent that I'd hoped. In fact I've convinced my self that it only correctly refines 2DOF.
Seeing as I have similar setups: what would work better in your opinion?
@gavanderhoorn We experienced significant improvement in calibration by first following the suggested Fanuc re-mastering before doing wrist/eye-hand calibration. Now the YAK/kinfu output using forward kinematics results in mostly smooth surfaces. Some error may be attributed to intrinsic calibration of the depth sensor. A single image of a flat plane has some visible curl. I'll leave this open until we do an intrinsic calibration and depth calibration on our sensor.
We experienced significant improvement in calibration by first following the suggested Fanuc re-mastering before doing wrist/eye-hand calibration.
hm, ok. So this is with a Fanuc-integrated gantry then?
Would still be nice to have a mfg-agnostic calibration routine/infrastructure for extended kinematic setups. Must be papers about that which could be worth investigating and implementing.