Why is the lid camera being used to home the machine?



We’ve danced around this question a few times on the forum, but I don’t think a dedicated thread has been started about it yet. I just can’t shake my opinion that the camera-dependent homing routine is a solution looking for a problem. I figured starting a dedicated thread would be better than potentially derailing an established thread.

I have the opinion that limit switches are a tried and true method for homing a CNC machine. They are certainly the go-to method used by the vast majority of hobbyist to “prosumer”-level machines that I can think of. Sure, if you don’t know what you’re doing you might end up smashing them (probably not a real worry), or a switch might fail now and then (theoretically), but their ubiquity must surely be a testament to their viability in machine design.

Dan has said that the/a reason Glowforge isn’t using limit switches is because the cost of routing the wires would be too high. I don’t expect Glowforge to share their bill of materials or anything like that, but if anyone would like to help unpack the costs/complications involved with routing wires I am all ears! Seems to me that the Glowforge would basically only need two limit switches (one for X, one for Y) and both of them could be stationary. The head has a variety of wires running to it to serve the focus mechanism, sensors (material height, accelerometer, others?), fan, and camera; are limit switches more complicated than these other electronic doodads?

Using the lid camera for homing has apparently also introduced some complication to the design. The lens(es?) on the camera must be capable of taking an exceptionally wide angle photo while also being able to focus close-up on the head. Is this dual-purpose lens part of the reason the center of the camera’s field of view is apparently the sharpest and most accurate? Are the two lights next to the camera dedicated to homing duty?

The possibility of modifying our Glowforges appeals to many of us. Unfortunately, using the head as a reference point complicates any modifications that involve removing or repositioning the head. And using the Glowforge™ logo as a reference creates a hurdle for any companies or individuals that want to build and sell a head accessory (possibly to fill a niche too small for Glowforge to want to bother with).

Will the coding necessary to home the Glowforge using the lid camera and head logo be included with the planned basic open-source firmware?


I personally don’t understand using the camera for locating the laser head either, I had assumed it was for placement of your project on materials but was then translated to a position on the laser bed and the head position was controlled by a tachometer. I based my assumption on the way tram motors are used in mining equipment, they have a tach that reads a 172 tooth wheel on the backside of the motor and a drive determines The distance the shaft has turned to determine distance traveled. Even with a 100hp motor with 2.5 inch shaft we can send the tram to hundredths of an inch of our desired destination. With the scaled down size of the shaft that I assumed would be used in the project you could keep track of head position with astonishing accuracy.

That being said I have no idea how they are driving the movement of the laser, and I have never worked on a laser cutter before. I had just assumed it was a scaled down model of what is used in large scale operation.

I am not trying to cast any doubt on the glowforge team but I am extremely interested in hearing why they chose the path they did. I’m sure it was a process that was not taken lightly and they have a solid argument for going the direction they did.


this sort of sums up my feelings on it. i’m sure they have sufficient justification; hopefully it was the right choice, haha.


So Imo (I’m no expert)

They have to do this to define the relation between the the gantry and the camera. The tolerance stack up between the camera mounting and the limit switches would limit the overall accuracy of the machine.

On most normal machines you don’t really care where the datum is. But because this system is marrying a camera and the gantry. The datum is crucial and one system needs to drive the other.

Just my guess at least.


All the high precision optical systems I’ve used at work use optical “homing”. They measure a fiducial (in this case, the Glowforge logo is the fiducial) and sets that as its datum point.

If you think of the Glowforge as a traditional CNC tool, limit switches are ideal. They set your x zero and y zero very easily and at the corner of your bed.

If you think of it as an optically driven system, then homing the head with the camera makes sense. You home the head to the camera and then use the camera to define your cuts. Everything is in the same system and reduces some of the tolerance stack. Not to mention the fact that the optical alignment can take out some of the tolerance slop from the magnetically attached head.

Feature Request: Numerical positioning

I’ve had a limit switch fail on me, so it does happen. As far as the optical homing making it hard to modify the laser – I don’t get the feeling that home modification is one of the things the GF is being built or designed for,. so perhaps a different laser might better suit those needs


Ah, so the camera is basically “square 1” for the whole system and everything the machine does is based on the images it takes. That explains why there aren’t any stops or rulers on the bed and why the “is there a (0,0) point?” question was so hard to nail down… there IS a (0,0) point, it’s located at a certain pixel in the camera’s image. And if the camera was to come out of alignment (the ol’ crumb in the lid scenario) the (0,0) point would still be at that same pixel, but at a slightly different physical location within the machine. Interesting.


I should probably say that’s my guess at how they are doing it, lol. But it does explain a lot.


I think the demos they did way back in the September '15 World Maker Faire launch and all the subsequent demos drive home the centrality of the camera to scan and print as the foundational activity and everything else gets built around this. Using 2D or 3D design programs is a whole additional layer of computer literacy that narrows the target audience. We want people who have no familiarity with coordinate geometry to use this machine without having to read the directions. Stick a picture or a hand drawn object in the bed and start clicking. That may or may not be optimal in the end, but it is a plausible design strategy. Why are backup cameras on cars so popular? All you have to do is understand how to look at three different mirrors. one of which has a non-normal focal length and all of which reverse the image and then turn your head around a couple times to back up. Who needs a camera to do this?




Let’s not go there again. :wink:


Aye there is, we just can’t use it as such just yet. But it’s getting closer. I think it’s at .80625in, .00001in. You can use more decimal places potentially, but it doesn’t really matter beyond that

Feature request: Absolute positioning - CNC Mill style



Yeah, they’ve definitely talked about how the camera is central, I just never took it as literally as I apparently should have.

I thought the “real” precision camera would be the one on the head¹ and the camera on the lid² would basically be the “helper”. :worried:

¹ the camera that is extremely close to the material and is at an easily calculable distance from the laser beam

² the camera taking an image of a fairly large area


That software is a part of the cloud service, not the firmware.