Strike 2 Glowforge!


Well, unfortunately, Support was not able to fix the issue of the perpetual calibration loop, even after downloading the logs and accessing our original unit directly. They therefore graciously sent out a replacement unit, which we set-up immediately… ONLY TO RUN INTO THE EXACT SAME ISSUE!!
The video feed from the camera can be seen and updated, so there is a good connection. It simply WILL NOT CALIBRATE or allow any file to be uploaded and/or cut/etched… even a trace (which does show in the portal, again good connection).



The good news it is highly unlikely that you received two machines with a H/W calibration loop problem. Other than your report, have not seen one for many months.

Unfortunately these are the type of issues that are very difficult to find remotely. It would probably be easier if this was your first time. But human nature will be to skim through steps when a person has done something 12 times and to self diagnose. Add to that the problem that nobody is looking over our shoulders in real time to suggest a different approach. Don’t envy Support.


Sorry, but not sorry.
This is $5k purchased piece of hardware that is utilized in a business/profession setting. Support is paid to… give support.
Yes, I am disgruntled.


Long ago I remember some people having a similar problem that was due to room lighting. Put a large towel over the top of the glass and see what happens after you power on. Can’t hurt to try.


Whether you are disgruntled or not is irrelevant to getting you up and running. Support will help as they always have.

Just pointing out that probability is against the problem being two machines. I’m thinking the only way you will be up and running quickly is if one person works with you step by step via direct messaging in real time. It’s too confusing to do it from the forum. And would need to start from scratch. They will need to have pictures of your user interface for each step, understand what devices you are using, which browsers, lighting, position of the head, etc. etc. One click at a time. Not an easy process for you or the person helping.

Or you can wait for support.


The fact that they have free support is amazing, we all paid for a device with a warranty, we didn’t pay for premium support with an SLA. This usually costs much, much more. That would actually be a great upsell opportunity for them with all the people here trying to run businesses with their GlowForge’s :smiley:


I opened my garage door about four more feet then usual once, which changed the lighting and the GF went into a calibration loop, so give that a try too @paul.chabala !

Closed the door and all was ok. Don’t know if that caused the problem, but it could have been the lighting, kind of strange!


I still await the rational explanation of why they opted to go with software defined calibration instead of physical endstop limit switches. $5k for machine that freaks out when it’s too bright because (I’m sure there’s reason; Anyone want some Kool-Aid?; that the calibration camera doesn’t adjust auto-exposure levels to a point where this is no longer a continuing problem for users.


At this point in time, I’m ready to try just about anything.
Gave the “cover the unit to shield from lighting” trick a go… no luck.
Losing faith, fast.


Dan claimed it was to save the cost of switches and wiring. Limit switches are only about $1 and wiring not much when there are already wires running all over the machine.

On the other hand, the cost of developing and maintaining a cloud base neural net solution and all the support questions it creates must dwarf the cost saved over 10000 machines. I doubt they will ever break even if they ever get it to work reliably and accuately because they have to pay for the cloud server usage forever. And it is slow because it needs to send several images to the cloud for analysis. Homing with switches only takes a few seconds when the home position is near the switches.


Do you really think there is a neural net behind the scenes? I’m cannot fathom what benefit that would actually provide. I haven’t seen a single feature that would benefit from a neural net yet.

The cost of hiring a single engineer would dwarf the amount of money saved by not adding proper homing to every machine made.


I heard a rumour that is was a neural net, otherwise I would not have imagined such a mad solution. It would explain why they can’t fix simple bugs like head banging easily. All you can do is retrain it to not be so stupid. I don’t think you can build logic into it easily.

It makes me wonder about the business model. You pay a one off price for the GF but every time you switch it on it costs GF money and every job you run costs them more and that is for the lifetime of the machine. Unless you buy PG or catalogue items they don’t get any more revenue. And you get free technical support, which must cost. And it is considerable because it is sold to non-technical users and also it is so opaque that even highly technical users struggle to diagnose problems.


There definitely needs to be some consideration by Glowforge for the business use case scenario. When the machine is down (as it is now, for a considerable time) it hurts our business. Selling a “business support” package would absolutely be justified, and worth every penny in my opinion. We do not have the time to baby sit the unit. This is not a hobbyist venture that we can come and go from at a whim.
We need it up and running, and we need it now.


I don’t think we’ve come close to seeing what offerings they plan on. Different heads for who knows what (material extruder for 3D printing? inkjet? router?). Rotary tool? Some sort of subscription services possible. (Not to existing services as they’ve already said they won’t charge for those.) I think they have plans that haven’t even been hinted at.


The specifics of how it is processed have not been disclosed (as per the norm), so whether or not it is an actual neural net is unknown.

What has been mentioned is that they use machine learning to identify/locate the head, and that is trained with a large collection of images.


Fabulous discussion… and yet my $5k Doorstop still sits… not working.

Not a peep from Support yet, after 48 hours!!
Emailed, direct messaging, and through this forum… crickets.


As with any support system, all that did was create 3 tickets in the queue.


You aren’t really providing any specifics as to what is going on. Does it attempt to calibrate? Does the head just sit there? Does the button flash at any point?


IIRC I saw a post by dan where he mentioned they used machine learning for one piece of the software. I don’t recall what part though.


Being that SVG, PDF, etc are known standards it would be far more likely that they would have their own internal compiler to that takes an uploaded file and outputs movement information. On top of that machine learning doesn’t do transpiling.

Even this doesn’t imply neural networks. It’s a good assumption as they are once again the cool kid on the block after being blacklisted for the whole 2000’s. Using a machine-learned algorithm would explain why we don’t have consistent results, it would be very difficult to machine learn an algorithm to consistently align the head for small adjustments in all of our varying environments.