hard wired approximation settings

In your case I would create a KRCDiag for each robot, unzip them than compare using e.g. Windiff or kdiff3. If you could tell us these differences we probably are able to give better advice. So far we only know you have two robots that might be the same. No info on KRC versions, software versions, messages in case of the different behaviour, ...

Otherwise there could be hundreds of reasons (different tool and base, different load, different settings, ...). Usually KUKA does not change stuff incompatible.

what does that mean 'robot does what it wants'?

is robot following programmed path but fails to do approximation (stops at points)?

or does it mean that robot goes off path and runs toward some far point, starts reorienting wrist etc?

post being of the program up to first few points. are you sure $ADVANCE is set to non-zero? how about $APO?

Quote

But when run on new robot it will sometimes (not all the time) approximate a sharp programmed corner with a 50mm radius.

And this only happens when you execute for the first time and on the second run through your program behaves as expected or after transferring to a new robot the problem is persistent?

Sorry I still do not totally understand your problem description. Does the robot not approximate or the robot still approximates but run a different blending path. Any messages in case of the bad behaviour like blending not possible? Did you check logbook what's happening.

I bet on backlash.

backlash causing 50mm radius on a path?

Quote

programmed with Power Mill robot plug in.

you mentioned that several times. it provides no info. KRC does not care what editor was used to generate code. it only cares what the code actually looks like.

Quote

Ptp velocity 20%

Accel 20%

i don't know what that means.... are your points PTP? what Accell? post the actual code

Quote

Smoothing .1mm

i don't know what that means. is this some PowerMill setting or $APO? show the code

Quote

Advance 3 points.

show the code

Quote

We program the robot and tool paths and simulation do as tool path generated by powermill is. But when run on new robot it will sometimes (not all the time) approximate a sharp programmed corner with a 50mm radius. Very noticeable on leadin and rapid moves but also within the actual surface milling as well. But run the same file on our older robot it does not approximate anything at all. Surface milling is as it should be.

need to see the code

Quote from panic mode

backlash causing 50mm radius on a path?

OK, that's a bit too much, but the infos he gives are a bit vague, so may be 50mm also are a bit vague.

But many years ago we had a real big path error on a brand new robot, we could see it just by looking to the robot tool while he moved in free air. That was not far from 50 mm (nobody measured it), it was KRC1 with bad gear in axis 1.

I agree: without the code and a description where the issue appears, it will be impossible to have a clue.

Can you check the value of $APO when the actual error happens under Display > Variable > Single. Also it would be interesting not to see your program header but also the actual faulty program code segment. Maybe posting the two KRCdiags you took earlier might also be helpful with a short description in which program at what program line of code the error happens.

So far all information we have does not allow any conclusions and we probably talked a lot about stuff that does not relate to the faulty behavior.

try to monitor you apo during the program run. im guessing the this might change after running home sub prg and tool change sub prg. if thats the case then try to change the location of you apo after the tool change.

It never made sense to post a screenshot instead of the whole program.

On the screenshot there is not one faulty movement.