The problem we are discussing is that we have no reasonable acknowledgement that the home position of the glowforge is exactly the same every time. From what we’ve been told it is not, which would negate the purpose of making one of these… =\
2 Likes
Lets say I need to add lettering to this irregular part. If I burn the letters in the wrong spot, I lose $200 and have to make the part over again.
I know that the lettering needs to be placed 6" over in the x direction, and 2" up in the y direction.
If I could place my jig on the glow forge, have it pick up the two points on the jig, count that as (0,0) and place an x axis along that line, I could place my drawing for the lettering on (0,0) and engrave it in exactly the correct location.
1 Like
My approach would be to use the 
's superpower — place the part on the bed, and use the camera to make sure the text is in the proper place.
True — I wouldn’t try that on my first go out of the box, but once I’m comfortable with the unit, I will absolutely be trusting with one-off parts.
3 Likes
The variable here is actually the means with which they are homing the head of the laser. From what we know the lid camera looks to see if the logo on top of the laser head is in the correct spot. However, being at the edge of the cameras viewable area and it using a wide angle lens, the ability for it to be as accurate as we might need has come into question. And as the machine does not use limit switches there is no secondary means of verifying that the location is exactly the same as it was last time.
Problems that I can see arising from this approach are:
- camera can shift or become physically moved over time by accident. (Materials hitting it when removing them etc), and/or slight manufacturing differences in placement
- lid doesn’t close at the exact angle it did when purchased
- resolution of camera isn’t high enough to accurately determine location within an acceptable margin of error
- smoke or other contaminants causing visibility of lens to change
- discoloration or distortion of homing logo on laser head
I think you get the idea. Without some way of making sure the head starts at the exact same position every time, any thing being permanently attached to the laser bed as a reference will also be inaccurate.
5 Likes
In this case, you would need to create a hole shaped like the part, in some sort of design software, include the lettering exactly where you want it to be in the file, and proceed according to the second description in my last post.
Creating jigs is just locking certain design elements into place relative to the whole…if you are trying to get something exact by eyeballing it, it ain’t gonna work.
If it doesn’t have to be exact, you can use manual placement via camera - should be good for about 90% of most users needs.
To get exact, requires 3-Point Registration, and the machine is not coded to do that.
I’m gonna be lazy and link to my earlier comments on it:
2 Likes
I’m having flashbacks from last year… (the first thread I posted in multiple times)
3 Likes
The accuracy of those alignments, at the time of maker faire NYC, was quite different from what was shown on the screen. They have themselves a couple of mm of play to make sure the pieces they were cutting wouldn’t fall off the edge of the material. Barrel distortion can be a pain to deal with.
I’m sure they will be dialing this in even more in the future, but it could potentially be different for every laser if there is any variance in the mounting of the lid cams.
3 Likes
I think this is what they will be using for flip-over cutting; Edge-finding and corner detection. Same process for alignment. So it might not be yet, as that part isn’t done, but it will be! Maybe they can parlay that into a solution for this situation.
5 Likes
@dan, you’ve lost me here. Why would the ‘bottom’ be better than the top? Aren’t there covers on both sides?
What would we fix by using acrylic and adhesive?
‘Register large materials’?
We’re trying to find a way of placing a jig in the same place every time.
I know you must be swamped right now but when you do get a moment, a few of us here would love to better understand how to do this with our GF.
We know that the lid camera is doing the looking, but I’d be surprised if it did the alignment by looking at the logo when the head is over in the corner. (It might, but I’d be surprised.) You’d get the same homing effect with much less optical trouble by running the head directly under the camera, or in the center of its field of view or through the region where the camera has the best focus. That spot will have XY coordinates that are determined by the mechanical stability of the whole thing, so you can just offset.
1 Like
I’ve been meaning to ask - just how accurate is that camera?
I vaguely recall it being 5 megapixels. If that’s correct each pixel represents 0.007" of distance on the bed assuming that the camera exactly covers the 20x12 cutting area and there is no noise in the image at all, so that may be a best possible answer. I’m wondering how much worse the actual accuracy is, especially if there is any image distortion.
Didn’t see this from the Pre-release letter mentioned in this topic yet:
On one hand we have the “preview exactly where it’s going” and on the other, improvements to do for accuracy with thicker material.
1 Like
Just to be pedantic, with a high-contrast target like a logo, you could plausibly do sub-pixel processing by watching how the light/dark value changes as you step (slowly) across the camera’s field of view near your alignment position. Worst case would be several pixels of slop on a moving image (which for a 5mp camera these days seems to be 1080p).
1 Like
There’s one feature of Full Spectrum’s mysterious Muse that I thought was well thought out. Instead of capturing one image of the whole bed with a very expensive wide angle lens, yielding severe distortion and the need for significant image processing to mitigate it, the Muse apparently takes many images from distributed vantage points, and stitches them together. Makes a whole lot more sense to me.
4 Likes
True… but you could sure pitch the idea if you wanted to.
“stressing” was probably the wrong word… “spending mental energy upon” might have been more appropriate. And of course a CNC needs to be accurate and repeatable; I’m not at all saying that you are wrong, merely suggesting that sometimes a new tool means learning new ways of doing things, instead of adapting the old way to the new tool. This may be one of those times… or not.
I prefer real vinyl, but I do run Ms. Pinky’s Interdimensionally Wrecked System and Torq. There are areas of the digital vinyl control system which must and do act exactly the same as traditional vinyl. But there are options and functions and settings to play with that enable you do stuff that isn’t possible at all.
I can tell the software to ignore absolute relativity, and do a needle drop anywhere on the record, and it will start playing at my designated cue point. Or I can tell it to be relative, and dropping the needle in the middle of the record would play the middle of the song. Marking, crayons, little round stickers, tic-marks… no longer needed. Software do the work.
That is what I am hoping for with the GF. Between the two cameras and fancy machine vision, you should be able to rely on the GF to accurately find the 0,0 point on rectangular media, no matter where it is in the bed.
My plotter is, technically, a CNC-type machine. It uses an Automatic Registration-Mark System to account for mis-aligned media, so that I can accurately cut shapes that were printed on a different machine as long as you set the file up with reg marks. Since the glowforge can see the actual design, it should be able to use that instead of printed reg marks for re-positioning.
It will all be a matter of getting the positional accuracy dialed all the way in… and seeing how well the cameras hold up over time. If they can get two-sided engraving to work they will have gotten the machine vision dialed enough to accurately find the top corner of rectangular material and define that as 0,0. Right? Don’t we have machine vision that can guide a missile with remarkable accuracy at very high speeds? Would you have thought, ten years ago, that a mobile telephone could do the incredibly complex process of multiple image recognition while at the same time processing digital video and playing it back in real-time? (Snapchat filters)
Granted: I am assuming the positional accuracy has increased since maker faire, and will continue to do so. I am assuming that, eventually, “snap to object edge” , “align vertical/horizontal to top/center/bottom”, “set origin at (X,Y) relative to object”, and other similar options will come along in the interface.
…and I let this sit while a customer came in, and the thread has progressed right on through what I was trying to get at:[quote=“takitus, post:144, topic:3386”]
I think this is what they will be using for flip-over cutting; Edge-finding and corner detection. Same process for alignment. So it might not be yet, as that part isn’t done, but it will be! Maybe they can parlay that into a solution for this situation.
[/quote]
3 Likes
You’ve nailed the problem…
if you place the cardboard in the same relative location each time.
For it to work it’d have to be precisely the same location each time. Without a 0,0… a place I know to exist relative to everything else… how could I ever do that?
3 Likes
Up against the back walls. Big piece of cardboard.
Honestly, I’m not sure if it will work or not, I haven’t seen the machine yet. 
But there should be a way. It might not be the most convenient method for certain specific needs, but we’ll figure something out. 
2 Likes
Yeah… I think I’m going to bail on this topic. That is until a staff member provides some uber-clear answer. Otherwise we’re all just going around and around.
(NOTE: That is not specifically aimed at you, @Jules! )
3 Likes
No prob…I don’t have any answers either! Chuckle!
1 Like
This is exactly correct. It’s very repeatable between homing operations.
To see how repeatable, here’s a head image from a 'forge after homing:
Here’s another image after power cycling the machine, moving the head, and re-homing:
As you can see, we can still make things a little better… the second one’s about a thousandth of an inch off.
The ‘bottom’ (side towards you) is close to the printing area. The side away from you is farther away. So it’s easier to reference from the side towards you.
Someone in the thread wanted a permanent stop.
You could rig something up that registers against the front wall and side rail.
I love to explain this stuff but I’m obviously falling far short here, as I’ve tried several times and still failed.
Correct.
Also correct. 
It does.
13 Likes