Glowforge has camera so you should just be able to drop your work in, align it once manually and then drop in another identical piece and have it automatically align to that, preserving the same offset from two edges. I don’t see why you need jigs with fiducials.
I do have a an L shaped jig on my CNC and it is screwed down into screw holes drilled by CNC, so I have run thousands of jobs from that. But my CNC doesn’t have cameras, it does have limit switches though.
If I cannot use a laser with the degree of accuracy it is capable of, its pretty much just a toy to me. I cant really justify spending that much money on a toy.
It IS capable of cutting things with incredible accuracy, they just need to find the best way to accomplish that. Weve dropped a few viable ideas on them that I dont think would be too hard to implement, especially considering the work will overlap with stuff already promised.
I have faith they can make something that will work for the not-so-picky crew, as well as the super picky crew that I was basically born into by way of two incredibly science-oriented parents hahaha
dan said the margin of error by using the camera is in the mm range. I cant work with that =\
Neither can pass through or double sided cutting, so they need to improve that. The head cam should be able to get an accurate fix on an edge.
Lid camera or head camera?
thats the one gotcha for a permanently attached L guide in the glowforge, is the passthrough. Its going to have to be removable so you can make full use of the bed. after removal youll want to know that you have got it back in place and perfectly aligned with the GFs 0,0 location. if it werent for the fact that all that stuff has to be removed sometimes I dont think we would have this predicament
Getting the hyper accurate alignment that people are asking for cannot physically be done manually, due to limitations both in our eyesight and the cameras eyesight. To get down to .001 inch accuracy absolutely requires a carefully placed jig or set of guides that can be used again and again to provide hard and repeatable physical stops for the objects to be engraved, and also depends on not moving the objects within the UI whatsoever after it’s loaded.
If we’re just going for visual accuracy, then fiducials would probably suffice, but once they get the cameras dialled in, fiducial marks wouldn’t really be necessary for that either.
That would also need the Glowforge to have limit switches accurate to 0.001" as well. I don’t think it even has them. It just uses the cameras for everything unless I am mistaken. The head camera has a laser line as well.
It would only need to use limit switches if it depended on aligning everything from the 0,0 point, which it doesn’t do because it doesn’t need to. It aligns the head using the lid camera above the centre of the bed. I think the disconnect is that people keep trying to equate this machine with machines that don’t have cameras.
I think we must be talking at cross purposes. You said you need jigs to align because the cameras aren’t accurate enough. If the head is zeroed with the cameras then it won’t be a repeatable offset to the jig, unless the cameras are accurate. In which case they can accurately find the edge of the work.
lid cam = inaccurate
head cam = supposedly way more accurate
they would have to leverage the head cam to allow us the accuracy we need in recognizing our jig location with the minimal margin of error
Hey All.
Whooosh, what a big thread.
I just did some registration-critical stuff yesterday with one of our older beta units. I’ll run thru what I did in case it’s useful or interesting. Note, all of the artwork in this post is me goofin’ in Photoshop.
A friend who is a founder of a smart-lighting startup here in Seattle wanted to do some user testing with some different iconography on the acrylic faceplates of their switches. He had a bunch of blank faceplates and asked if we could engrave them over lunch. I emailed a bit with his manufacturer to verify that it was acrylic, and here’s what we did.
Well, first, let’s look at an empty bed (artist’s representation!)
The non-whitened area is roughly what you’d see on the screen: some of the black plastic rails (raised on the sides, NOT in front/back), the head in the upper left, and lots of honeycomb.
The reddish area are mechanical limits of the head-- you can’t print here. Of course we want to make sure MATERIAL can go here so you can cut the maximum-sized rectangle out of a larger sheet of material (i.e. the raised side plastic is wider than the mechanical limits).
My friend wanted to make certain that his test faceplates were the exact same every time. While the camera is pretty darn accurate (and getting more accurate every day!), it’s a pain in the backside to drop a faceplate on the bed and fuss with moving/rotating/etc.
So, 5 minutes later, I had a (laser cut) jig and we were off to the races (well, we had to do a few tests engraves because it wasn’t our familiar Proofgrade acrylic).
Step 1, tape some scrap cardboard to the bed so it doesn’t move (shown: blue tape!).
Looks like this:
Step 2, Upload the faceplate design to Glowforge and drag it to the cardboard. You see something like this in the UI:
Step 3, Set the engrave part of the design to “Ignore” (i.e. it won’t print). Then cut out the outline and remove the rectangle. Now you have this:
That’s a jig. Now I set the outline CUT to “ignore” and un-ignore the engrave parts of the design. I open the lid, drop in the white faceplate into the hole I cut in the cardboard, hit print. When it’s done, put in new faceplate, hit print again. Etc. Looks like this:
As long as my source artwork has the shapes in the same place, I could load any number of variants of that faceplate design and have my jig work great.
You could also make an open jig (an “L” shape for more varied-size stuff).
In practice, we’re seeing the need for this level of precision is actually really rare… It pops up when you want to cut/engrave an existing object, but obviously doesn’t matter much when you throw a raw sheet of material in the bed to cut out a fresh design. It’s certainly common enough that we want to make sure that people can get precision when they need it.
(Warning: I’m out of my element here, so Dan might correct me if I’m mistaken!)
So, could I make a permanent jig of some kind that could be used over and over again? Depends on your need for precision, I think. We derive 0,0 based the camera and the head. And we calibrate this every time you turn on your Glowforge. Given that the head and lid camera are attached to things, and that those things can change, our sense of 0,0 could change very slightly. Example: there is a tiny bit of debris in the lip of our lid that causes it to be 0.1mm higher than it used to be. 0,0 just changed.
So, by example. Say you put a piece of blue tape in a spot on your bed and on the first of every month you opened the same svg file of a tiny square and hit print (without dragging artwork around at all). Would it always cut in the exact same place? In theory, yes. In practice, it could vary fractions of a millimeter because of computer vision stuff (that we can improve). It could also vary because your lid / head relationship changes somehow.
Fairly long-winded-- I hope it’s helpful for someone.
For doing these engraves all in one go, that will definitely work. The real trick is doing one, then having to do something else (meaning you have to remove the jig), then putting the jig back and doing another one and ending up with the same result.
Also… for me, a lot of things I want to engrave are existing items. Either crafted elsewhere by me, or being something that I own that I want to add detail to. In those cases, the cut a jig on the spot isnt so bad, but for the product kits that I create that are sold on a one-off basis, I need to have repeatability… and hopefully without cutting another duplicate jig that ive already cut
If there were some marker that we could put down on the bed, or object with markers on it to use as a homing point for the head AND physical guide, we know that we could get accurate, repeatable cuts every time.
btw @tony, thanks for taking the time to read this thread and converse with us regarding this
e.g.:
I know the horse is long dead, but:
What I get out of this is that if we can locate something consistently to the interior of the gf, we can use that as a reference that is accurate to the homing accuracy of the head. Which may or may not be good enough.
With MOST existing items, the notion is that you should be able to drop it in the bed, drag your artwork where you want it, and hit print with enough accuracy that you’ll be thrilled.
Another interesting way to look at it. Say you really want a sense of 0,0, so you taped a big piece of wood to your bed and cut out the largest rectangle you could. So you start with blue-taped wood like so:
And you cut out your largest rectangle like so:
Is the top-left of that slot 0,0? Yes. You can use your original rectangle artwork as a template to key off of this corner. Is it possible that 0,0 could change VERY slightly (sub-millimeter)? Certainly yes on reboot (and we might add quick calibration checks, so it could happen other times down the road). Could it vary by more than that? Unlikely, but possible if the relationship between the lid camera and the head change substantively.
(standard disclaimer: I am a software designer founder by background, so Dan or someone else might correct/clarify what I’m saying!)
If it can locate the jig accurately then why can’t it just locate the work piece edges instead and dispense with the jig? Yes three fiducials are easier to home in on but Glowforge needs very accurate edge detection for the pass through to be useful. That is unless we have to add fiducials to our long pieces.
Thanks @Tony!
Since @Tony and I have just said opposite things… 
I think Tony may be better at metaphor than I am… you can sort of think of it as 0,0 for purposes of print-to-print consistency; it does not appear that way in the UI.
That was what I measured for you above - registration of the camera is < 0.005 right now between a few reboots, and will get better when we add detection for the case of the lid being ajar.
The original example was a circle or arbitrary shape. The passthrough is designed to work with corners as reference points.
Note that the lid cam is accurate to ~0.001 when used to home the head, since the head is right underneath it (per my pictures above).
It just occurred to me that the two homing images (the two images of the 
logo) possibly aren’t as indicative of consistency in the head’s position as it may seem.
Since the head homes to the camera, images of the head after a homing operation really, really should be quite similar to each other.
If I homed to the camera and took an image (local01.jpg), turned the Glowforge off, physically moved the camera +5mm in X and + 5mm in Y, turned the machine back on and allowed it to home to the camera again, and then took a second image (local02.jpg), I would expect the two images (local01.jpg and local02.jpg) to be very similar to each other, even though the position of the head would be off by 5mm in both X and Y.
I feel like my hypothetical is confusing, so here’s something analogous…
Imagine two “speeding enforcement” cameras set up at two intersections of the same street. Let’s say this street is very uniform (level, no potholes, no turns, etc.). Let’s also assume the two cameras are set up in a very similar way (same height off the ground, same triggering mechanism, same angle, etc.). If an orange Corvette blows threw both intersections while going the same speed the photos taken by the two cameras would probably look pretty similar, even though the car was, in fact, in two completely different locations.
For the head to be in the same position under the camera, it must be in the same position in the bed, since the camera is immobile unless there are e.g. crumbs under the lip. In your analogy, they’d need to be the same camera at the same location.
