I have been trying to work out how a hand eye calibration should be validated and I have not found any complete answer yet
There are some tools I looked at (MoveIt Calibration, easy_handeye, Kalibr) that just print reprojection error or transform, but I could not find a clear way to decide from that whether the result can be trusted or not. I read some of their GitHub issues and many people seem to end up in the same place “result looks off” and the only way to find out why is to ask someone more experienced.
Checking by eye on one robot and looking at each result in RViz is one thing and it’s fine. But that shouldn’t be the proper way to do it, especially at scale right?
So I would really like to know how people handle it in practice. What do you run after a calibration to decide it is good? Validation motion, touching a known point, comparing against a measured pose and does any of it work without someone experienced doing the judging?