UP Camera error - offset

[David Griffiths]

Hi,

What is this error telling me:

Is this offset value configurable?

Is it because the component is being picked well off centre?

The component being picked off centre is another related question...

This is what I see when I click the 'Move camera to next pick location' button:

and this is indeed where the pick occurs, even though I have 'Use vision' checked for this feeder and the Pipeline looks like this:

The index holes are being correctly detected even on the dreaded clear tape.

What am I missing?

Cheers,

DG

[mark maker]

> Is this offset value configurable?

Yes, it is a new setting (testing version) on the Nozzle tip:

The assumption is that tips for larger parts (like your elcos) have larger tolerance.

The idea is to increase the probability of detection of mis-picks, and to reduce computation cost of some stages, by reducing the search space.

https://github.com/openpnp/openpnp/wiki/DetectRectlinearSymmetry#nozzle-tip-configuration

_Mark

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/ae4c258b-ce82-486b-96e2-d5951cb16c11n%40googlegroups.com.

[mark maker]

Btw. (as an extension of my own answer): This is the testing version and one of the ideas of testing is to find reasonable default values for new properties like the mentioned Max. Pick Tolerance.

So if you think 1mm is too little for the average machine, all you people's opinion is valuable! 😎

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/0bdacca7-8117-bd0d-76cd-e0837b959d34%40makr.zone.

[David Griffiths]

Thanks Mark.  I will play with the numbers, but I think the most important issue for me at present is to get the pick more accurate in the first place.

DG

[David Griffiths]

A question re vision on strip feeders...  this could probably be a new post but the pictures I posted above illustrate the problem.

I had assumed that when using vision to look at the index holes on a strip feeder, OpenPNP would adjust the pick point based on where it saw the index holes, eg if the tape had moved when peeling off the cover tape.

This does not seem to be happening. In the pictures above, the expected position of the index hole is displaced from where the index hole actually is, so the component is picked by this displaced amount and then has to be compensated using the UP vision - hence the offset tolerance error mentioned above.

So exactly what is the DOWN vision used for?  tape out detection?

Cheers,

DG

[mark maker]

> ...  OpenPNP would adjust the pick point based on where it saw the index holes, eg if the tape had moved...

> So exactly what is the DOWN vision used for?

Unfortunately the ReferenceStripFeeder does not adjust the feeder position per se, more specifically it does not adjust the first sprocket hole location. It only adjusts the advancement along the strip, i.e. it can cope with an angle/curve along the strip. If you move the whole strip after the position was nailed, it will detect a crazy angle and likely even amplify the error. The system really assumes you tape them down.

If you want a more advanced system, that can recalibrate all locations and moved feeders or whole arrays of feeders, look at the BlindsFeeder. I made it precisely as an improvement over the Strip Feeder (in more than this way). The latest version can even use OCR (not shown in the video, but in the Wiki):

https://makr.zone/new-openpnp-blindsfeeder/353/

_Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/aa28fd7f-af81-452d-8f12-2a5d5f16ea28n%40googlegroups.com.

[David Griffiths]

Hi Mark,

That is disappointing news. It seems like a major under use of the vision to not adjust the pick point each time the index holes are checked. Having to actually tape down the component tape is way too much trouble!

I am not sure I understand what you are saying about adjusting the advancement.  Is it looking at both index holes when it takes the imagecapture? and is it then calculating where it will look next time?

I did build some blinds feeders but I didn't have much success with them :-(  partly because I redrew the design in Fusion360 and didn't realise some of the interdependancies of various positions. They ended up being too flimsy and the components were always getting bumped and jumping around.

I think a lot of people are using strip feeders, so it would be worthwhile improving them.

Cheers,

DG