Do you have data on “most units”? I’d love to see that.
I, of course, can only speak anecdotally (as I don’t have access to data on “most units”), but in my case the unit is most definitely off by a fixed amount consistently.
Your incident of error is the same for say… top left, bottom left, top right, bottom right quadrants? And then the center is the same as all of those?
Wow. This is something… I would have added switches
Many tried. They chose to not listen.
Yours might be an unusual case.
Ignoring the probable tone… You have read less than 2K posts. Some of us have read 285K and been here from the first deliveries. Many hundreds of bed images showing offsets. Very few if any with a simple fixed offset across the bed. I’ve only had three machines, all with variable offsets. If you don’t want information or a reasonable discussion then that’s OK with me.
As an engineer, I’m pretty sure this is not the case. Related to the camera and warping issues there does need to be mapping transform, but for a consistent offset issue there is no transform that needs to be done.
That said, I’m curious how they calibrate the system in production. I’m guessing the wide angle lens / camera system is not highly consistent across machines which causes all of the variation we see.
I would lay down a known square grid with sufficient line spacing and image it thru the camera, then the data transform can be easily calculated. Since the error might not be consistent at different focus points, this might be repeated at different heights and then you can calculate and add in the height variable to correct for this distortion.
This is similar to how automotive HUDs are calibrated as the image moves up and down the windshield with varying degrees of distortion.
I think the point is that no machine (of the many, many, screenshots posted here) has exhibited a consistent offset across the bed. That’s just not the nature of distortion exhibited by wide-angle lenses (barrel distortion).
You couldn’t simply say the offset is -1.4mm on the X and +2.4 on the Y. It might be that for the upper-left quadrant, but you’ll need a different offset for quadrant 2, 3, 4 and then the center.
Didn’t say his unit needed anything other than an offset correction. My point was that his unit is the exception and most of the units seen here seem to have an error that varies across the bed space. If a unit has a fixed offset then a fixed adjustment would be possible. If not, some re-mapping is likely to be required. And that type of correction would address both the fixed and variable errors. A straight error correction might make him happy, but few others.
The rest of your comment agrees with what I believe to be required. BTW: I am a Computer Engineer and an Electrical Engineer. Considerable hands-on experience with S/W image distortion correction for military aircraft training simulators and Head up Displays. Doesn’t mean anyone should listen to me, but smiled a little when you started “As an engineer”.
No worries @rpegg – we’re in agreement actually. I agree a fixed offset is only useful in limited situations, as described.
I am also an electrical and software engineer (and biomedical in previous life), currently specializing in embedded graphics.
Same. Funny part of it is I don’t really care whether I’m right or not. Only way to learn is for me to be wrong. Just providing the most relevant info where I can.
This is all interesting discussion about accuracy of the camera but I have a slightly different take on it.
I simply don’t want the camera to claim sub-1/4" accuracy.
[Some speculation follows, but I feel pretty good about it]
Aside from the economic argument against it, anything that accurate will require a lot of work to maintain both on the GF side and on ours as end users. I think the “click and go” aspect of Glowforge is its strongest asset, and we’d lose a good bit of that if we chase down tighter placement accuracies that have diminishing returns (see economic argument above, and that’s only about the user’s economics, not the equation on GF’s side). I don’t want to get my camera to pixel accuracy only to have it go out of whack because humidity changes caused my table to shift somewhat, and causes some torsion in my GF frame, thus negating all that perfect accuracy. I don’t want to put in a material piece that I didn’t even notice was very slightly warped, throwing off my accuracy to the point where I ruin a cut.
“…but these things don’t happen that often!” I can hear people saying. Not to you, but multiply that across all the people who use the GF in all the ways. People who keep the GF on a rolling cart. People who installed it in a non-climate-controlled garage. People who haven’t cleaned their lid camera lenses lately. People who installed a GF on a non-level surface. People who store their materials vertically vs horizontally. People who live in Minnesota and people who live in Puerto Rico and people who live in Austria, climate zones that are completely different from each other.
There are just a tremendous number of variables in terms of physical installation; variables that don’t move the needle enough to upset the 1/4" limit, but will likely throw off predictions in the sub-mm range. It’s a daunting task to do what you’re calling for in anything but a highly controlled environment. I just don’t see the benefits outweighing the costs to try to account for all that.
From GF’s more practical persective, if they claimed more than 1/4" accuracy it would no doubt lead to a sharp uptick in support calls. We already see enough “my alignment is off” posts as it is, that would skyrocket. I think the worst case would be for GF to roll out a system that is highly susceptible to environmental changes like this, giving us false expectations is absolutely worse than giving us a reasonable margin of error.
Essentially, having a reasonable and true expectation is what I think is important here. If you need higher accuracy of placement (anything less than 1/4" officially, probably less in most machines), I say optimize your workflow for jigs. You’ll save time, effort, materials, and frustration if you do.
[As an aside, and not to get too far off track, that’s where I think GF should be investing its development efforts: give us a dang pause action so jigging gets simpler and faster. They’re killing me over here.]
I enjoyed reading your post, but I’ll have to strongly disagree.
Calibration of a system via the methods I described should easily remove MUCH of the error and make it accurate to far better than 1/4" for all users. I’d be happy w 1 mm accuracy at this point. There won’t be much variation for temperature / environmental concerns, it just doesn’t change the system enough for that to be a concern.
All of the calibration I’m suggesting is factory calibration and not something for the end-user.
It would still be plug n play. Obviously camera replacement would require recalibration at the factory.
Warped material would change the accuracy of the cutting start location, but I see that as a materials problem not software.
You’re argument about the “pixel perfect” accuracy is mute - even if we had that accuracy in a tightly controlled environment, you don’t put that in the specification. You put in the number that is valid across all other operating specifications.
It’s well known that non-level installation will warp the frame of the gf, throwing the factory camera alignment out of whack. Shipping subjects the GF to a lot of abuse, likely also throwing things out of whack. Factory calibration just wouldn’t be enough.
And I’ll make one final logical argument for you here: you say it it should “easily” remove the error. GF worked on this machine for years with people who live and breathe the ins and outs of the machine. I think it’s safe to say that “easy” doesn’t describe what you’re talking about, or we’d have it by now. I stand by my entire post.
I would love all of these things to be here now.
I don’t expect the bed camera is going to get appreciably better for alignment at this point. They have had enough time to work out the major kinks in it, I just don’t see it getting much better. Material height plays a major factor in the dewarp, especially on non-proofgrade, which is 90% of my Glowforges diet.
I’m holding out for head camera alignment to make things better, But there has been no chatter about it lately. I’m not even sure its a thing anymore.
So I think we will not hear from them.
They won’t comment on any timeline info.
In contrast, I got an update on my Maslow shipment, It was due to some items not arriving, and some simple packaging items. It does make me feel better to know why things are delayed. At this time some roadmap would be very much appreciated, but I don’t expect to see one. Like ever…
I agree completely. Their lack of communication is not fun at all. It’s actuslly holding me back from buying 3 more systems for our training centers.