Thoughts on Glowforge Laser

[ajf]

So the Glowforge is now available for preorder.  From my perspective it's a bit pricey, but there are some nice ease-of-use features.  The camera based registration and tracing are quite cool.  As is the autofocus.  Might be fun to hack.

regards,

aj

[Arthur Wolf]

About camera registration and focus-based engraving, I believe Visicut has all that : https://hci.rwth-aachen.de/visicut and doesn't make you dependent on "the cloud".

About the glowforge itself, at it's real price of $4k, I'd much rather get a chinese laser cutter of similar size and put open-source electronics inside ...

I guess it's kind of like the Makerbot of laser cutters, so there is a market for that ...

I'm also really dubious about how well they cool the tube, and how well the extract the smoke. I wouldn't be surprised if it aged very badly.

[ajf]

Well, I don't think VisiCam is quite as sophisticated as what GlowForge is claiming their setup can do, but mostly that's due to hardware.  GlowForge apparently has two cameras, one on the lid and the other on the head.  They also seem to mention other sensors in a couple of the videos I've seen, though it's unclear what those may be...  But, on the software side, that's all pretty easy especially if you have a known/standardized sensor package. Making a dynamically focusable head is also pretty straight forward and cheap. Combine the two and you have a pretty neat retrofit package for any cutter.

I'm actually going to explore this a bit and would be interested in working with others to bring something to market.

The cloud stuff is just stupid, I'd bet they do a native app in short order.

As to the machine itself, I agree, there are a lot of questions re the design. Appears to be an ABS enclosure, that's not too bright methinks. And, having the tube on the carriage doesn't seem like a great idea either...

Think the MSRP is BS, even the "discounted" price is way out of line for that class of machine.  Though, you're right, folks who think MakerBot is worthwhile will go for it... $500 for a charcoal/HEPA filter add-on, really? Then another $1500 for a pass-through...

[Shane Middleton]

Its definitely an interesting product. Seems marketing guys are at work, so many of the mechanical details are hidden.

It does seem to use water cooling but, haven't seen a photo of the rig.  I guess that would get in the way of the 1 box press material they show, they might be setting them self up to overstate the consumer friendly side.

Laser on the carriage is great / scary.  Its great as you keep the optical path shorter = less dispersion and they close the 1st 2 mirrors in a sealed chamber so nothing to adjust (you hope) and wont get dirty so less to clean.  But 40,000 volt wires are moving on the Y carriage along with water recirculating tubing...either comes loose and wont be fun.

There vision solution is awesome (takes visicut and goes to next logical step)  and the raster engine results also look great, so for $2,000 USD you get a good laser (TEM00) with good mechanics & electronics plus a DSP etc.. so its good value for the package at the intro price. 

Their filter system also sounds well thought out. I am seriously tempted, currently running a puny 500mw laser diode engraver (17cm / 22cm bed with 1.8o steppers controlled by an arduino nano stepper board running GRBL), which is teaching me a lot about post processing gcode. 

[ajf]

Hi Shane,

Yeah, I'm not sure about the "no adjustment" thing, and you still need to do cleaning of the enclosing bits.  Also, aside from the cables and cooling, my concern with the moving laser tube is if a piece of scrap or whatnot gets in the way and breaks it...

Sounds to me like they have a pretty conventional radiant cooling system, with an active piezo cooler as an add-on for the pro version.  

The filtration upgrade seems very overpriced to me.  Active charcoal and HEPA for $500, that's pretty steep, you could implement a better setup with limited effort from stock parts.

Curious if you would find a ~$200 vision/adaptive-autofocus add-on to a K40/DYI laser to be an attractive alternative...? Say it was provided with a software package that could be implemented with an opensource solution like Smoothie or Marlin...

regards,

aj

[Shane Middleton]

Your right, standard is a closed loop cooling system. And pro has thermo electric cooling. That must kill the duty cycle between long cool downs, although probably not a real issue for a consumer type product for diy projects or prototyping work.

500 seems like a lot for fume filter, but the form factor is nice and worth some premium over a large external fan a nd duct.

If someone comes up with a visicut style plug and play solution with visual placement registration and z height tracking for curved engraving for $200 I would buy it.

[quillford]

I'd be willing to help with software on making something like glowforge that works with the open source firmwares (smoothie and marlin). I think going web based for the UI would be the best option, but have it run locally rather than through the cloud. It would be similar to OctoPrint. A lot of the features the glowforge software has can be really easily implemented like the tracing of hand drawn images. However, features such as double sided engraving/cutting would take more time and would be harder to configure across setups.

I think using something like replicape would be ideal because it would have the power to run an OctoPrint like thing on its own, but it could also be separate from the replicape firmware while being portable so that it could run on something like a raspberry pi to attach to RAMPS/smoothie. It would be cheaper than smoothie + a raspi but more expensive than a china ramps + raspi. Although it could do raster engraving better than ramps as far as I know.

[ajf]

My current thinking is to have the sensor package be standalone and just expose an API. Given that there are going to be multiple cameras and maybe other sensors, the DSP stuff is going to eat up a lot of resources.  Will know more once I meet with my DSP expert friend next week.  

[ajf]

Update:

Talked to my DSP expert friend and he confirmed the 2D stuff is pretty straightforward, but that going to 3D would be exponentially more complex.  Will leave 3D for another time.

Been playing around with Visicut a bit, but the version that supports Smoothie isn't playing nicely on my Mac.  I'm going to have to dig into the code a bit an see if I can port out the visualization stuff, I've got an aversion to Java.  

Have also spent a little time studying all the Glowforge videos I could find and, interestingly, it looks like they used RaspberryPi and the CSI camera module while prototyping.  Sneaking suspicion that they also started out with Visicut. I've tried the stock CSI camera in my K40, but the focal length is such that it will need to be mounted a couple of feet above the machine to capture the entire bed.  That's about where the Glowforge folks had it on their prototypes too.  Going to try and find a wide angle lens that doesn't have too much fisheye effect, would appreciate any recommendations! In the longer term, suspect that that would need to be a custom part.

Once I get a reasonable 'overview' camera setup working, will move on to the head mounted camera. Glowforge says their head camera is near-UV, other than scanning invisible QR codes, I'm not sure what that gets you...? Would appreciate any input on that also...

As you may be able to tell from my verbosity on this topic, I'm pretty excited about the project.  Going a bit Tony Stark and crudely renovating my shop to accommodate building a larger laser to prototype this more freely than I could on the K40.

regards,

aj

[timne0]

Very interesting aj!  Make sure you document the steps on here or a blog so we can follow!

[jegb]

I have the feeling that webcam used for CAM immediate conversion is at the core of Glowforge, you may be aware of this 

https://github.com/t-oster/VisiCam

[jegb]

I have started documenting a possible schematic with few paths for control loops:

+water temp, perhaps adding an exchanger too with a fan on same loop

+water flow sensor (analog) for interlock or alarm

looking for feedback, also working on a smoothie variant on same schematic.

jegb

[ajf]

My understanding is that VisiCam is essentially a image server, but I haven't dug too deeply into the perspective correction and marker detection aspects.  Haven't been able to run VisiCut in any usable fashion yet as I'm on a Mac and there are issues with the Smoothie compatible binary. However, I was able to get it to recognize the camera on a rPi running an octoPi image. One thing I noted when reading the VisiCut docs is that visual registration is accurate to "some millimeters", that's really not good enough...

Right now I'm trying to find a good wide angle lens that will work with the rPi CSI camera.  Just got a cheapo set of add-on lenses for smartphones, need to print an adapter but the wide angle looks promising.  

regards,

aj

[jegb]

Hello,

Pick and place machines need to be very accurate, we have a liteplacer, and I will investigate which cameras they use, there is a visual homing. 

Also there is a framework for pnp machines for open cv https://github.com/firepick1/FireSight/wiki not sure how smoothie could evolve to have visual homing as a start, in either octoprint or directly on smoothie firmware.

The current opto couplers or mechanical stops seems to be rather simple solution hard to beat. The optical setup would be great for :

1)  focus (need Z axis control)

2)  converting visual capture to gcode on the fly (Visicam, similar to Glowforge)

Perhaps a combination of ultrasound and camera can help compensate or complement each other.

jegb

[jegb]

Now after some more investigation, the up-facing camera system on PP4 (also opensource) pick and place looks like a standard camera part with focus system is 70$ from Microsoft, stripped the enclosure and kept electronics.

The interesting aspect is the red/green light and brightness control, performed by the system, so calibration is more accurate, my machine has a diffuser film and blooming and reflections are problematic.

In the case of the laser, we could have:

- down facing camera for opencv to raster/vector and gcode conversion

- down facing camera for focus control 

- side facing camera for z distance control or material thickness (which could be solved with other probing system, though

Since the code on this project is open, anybody is welcome to take a peek and see if its fit to smoothie project. I personally will try the ultrasound approach to see how accurate the ping is for material thickness sensing. There are other options as piezo sensors as the one used on cnc probing tips...

jegb

[ajf]

While watching the LitePlacer videos I saw the PP4 too, the LED calibration was pretty cool.  Looks like the LitePlacer uses cheap (~$10) USB endoscope cameras, which might be a good choice for the close up/head mounted camera on a laser.  I think the rPi CSI camera is going to be the best choice for the overview/bed camera.  The one I have came with an M12x0.5 threaded mount and there are a bunch of lenses available, just have to find the right wide angle one with minimal fisheye.

Would be interested in your reasoning for wanting to probe for material thickness.  GlowForge doesn't do that, and given that one would know the thickness of the stock they are using and the focal length of the laser lens, there really isn't a need to probe.  Glowforge is going to sell there own stock which will apparently have IR QR codes that the camera will read to automatically set focus, feedrate and intensity, but for off the shelf stock the user has to input the parameters.

regards,

aj

[jegb]

one reason to detect the thickness of the material to set focus at different depths of the material is to allow folding on acrylic, see this video 

I have an rPi camera module over here too, on the octopi setup. I did not know there are many camera lenses setups, sounds good idea to test.

jegb

[ajf]

The rPi camera I have came with an aluminum version of a mount like this, not sure if they all have that but it's a pretty cheap add-on.

Warming up acrylic to allow it to bend is just a matter of taking the lens out of focus, it doesn't rely on dynamically detecting the material thickness.  GlowForge promises to do that, though I don't think they've demonstrated it publicly yet. The linked video doesn't seem to indicate any probing, they are just changing Z on the table to defocus the laser for the bends.  Also, doing this sort of bending requires that the stock be aligned on a pin table such that the bends can be formed by gravity.

I do see a use case for doing a 3D scan of the bed, but that would be for etching/cutting on unique 3D stock.  And that's pretty much an edge case... For example, GlowForge says that you can put an iPhone or MacBook in and etch on it, they do that by detecting the object visually and matching it to their database to get the geometry. There's no need to scan the object in real time as it's a manufactured item and you know all the dimensions.

regards,

aj

[jegb]

Thanks for the link! what is the focal length you have for the throw 50-100mm ?

I have a z lift on the table, which is on my to do list to remove and have a fixed honeycomb grid, then having the laser lenses moving together with the air assist nozzle, but may create problems with air pressure loss, fiddling with o-rings on 3d printed surfaces may not work well, so only option may be alu machined on a lathe.

Anyway, long project good to start ultrasonic and see where it goes.

jegb

[ajf]

Sorry, I don't think I understand the question. No idea what you mean by "throw", and I'm not really sure what the focal length is of the stock rPi lens, think it's 3.6mm. Photography isn't my thing...  I got one of these add-on lens kits for phones and, with the wide angle lens attached to the stock rPi lens, I can get the whole bed in the shot at about 430mm.  Need to be able to do it at something less than 140mm and with less fisheye effect.

I'm going to try one of these lenses, seems about right.

regards,

aj