Repository navigation
Printbed clearing #35
Replies: 4 comments
|
Yeah, I asked Matt to share his approach about four minutes after he posted it. He has not replied lol. I actually got something similar working last night, and unlike the 16 attempts in the demo, it basically worked immediately. The MCP plus the printer knowledge/context it gives the model means the agent already knows an absurd amount about the machine, so getting it to reason through the process was pretty straightforward. When I replied to Matt I said “I really want to add this!!” I still do. But after actually implementing it, I’m much less convinced I want to add it this way to the official MCP. The reason is safety. This project has existed since March 2025, including the parent fork, and for most of that time it quietly accumulated users who generally already knew something about 3D printing. That is changing fast. Models are now genuinely good at CAD/Blender, printers are getting easier to use, and increasingly the person telling the agent “make this and print it” may know absolutely nothing about printers. Honestly, they shouldn’t have to. That has made me much more conservative about what primitives the MCP exposes to an agent. A fairly ridiculous amount of work has gone into making the existing print path fail closed. The MCP inspects the final G-code, checks fresh printer telemetry, verifies the printer/nozzle/material/plate configuration, imposes independent thermal ceilings, rejects unsupported heating commands, and refuses dispatch when required safety information is missing or contradictory. It also deliberately does not claim to know things it cannot actually know from telemetry. One of those things is whether the physical build plate is clear. Even with confirmation disabled for headless operation, a printer reporting Automatic bed clearing changes that assumption. The implementation I got working essentially requires deliberately using low-level printer behavior to manipulate the finished object off the plate. Once the MCP exposes something equivalent to “perform arbitrary clever motion because the model thinks this should be safe,” that is a very different capability from the constrained print path I have now. And agents increasingly being an attack surface matters here too. I do not want a prompt injection, bad model decision, or simply weird reasoning to turn a convenience feature into unrestricted motion commands, arbitrary heater setpoints, or some other sequence that can damage the machine or create a real physical hazard. So I am not closing this, because I absolutely want the feature. But I think the acceptable implementation would need to be much more boring than the demo: probably a narrow, deterministic, model-specific bed-clear operation implemented programmatically, with hard limits and state checks that the LLM cannot override. Ideally it would live in, or be based on, upstream printer/software behavior rather than teaching the agent that safety constraints are merely suggestions it can reason around. Basically: “clear the bed” can be a capability. “Here is enough low-level control to invent a way to clear the bed” probably should not be. Sorry for the enormous context dump, but I wanted to explain why something I was extremely enthusiastic about yesterday became a “yes, but we need to be very careful how we do this” after I actually built it. |
|
Oh and there's also the issue of the door, forgot that one. Would need a way to open the door and close it again. The door being closed during printing could be a safety issue, based on what your printing with; however, it's more just a print quality issue. The printer is so well calibrated to the temperatures it's expecting, and when it's an encosed printer, it expects to be closed... I have a few ideas for this but would love more! |
|
Thanks for the detailed account of your thinking on this, makes sense. Regarding the door, yeah I see your point, raises the question of whether there should be an external door opening/closing mechanism. It may make sense for this to be tackled upstream, and I wonder how far upstream the "and then clear the bed if these criteria are met" instruction can go, and if the printer will ever have enough information to for the criteria to be met safely... I imagine that auto bed clearing has been considered by Bambu and other companies making these things and there are probably good reasons for the functionality not coming out-of-the-box, safety-related reasons being high on the list. I imagine quite a few things could go wrong, e.g. forces on the printer head that it was not designed for, meaning calculations estimating likely forces would be needed as part of the safety check criteria etc. I found a related thread here which shows indeed a lot of people have thought about this: https://forum.bambulab.com/t/project-to-automatically-remove-parts-from-build-plate/17231/130 I'm not an expert and quite new to 3D printing so looking forward to learning more in this space and will keep an eye on this conversation with interest. |
|
I think ultimately the solution will come from bambu and be something like: de-magnetize plate (which is easy if you run a current through the bed, and flip it off/on); having the bed roll x-axis out of the main print space, whether that's out the door, or whatever, new plate slides in, continue. Something like that, where the bed isn't actually scraped, is probably the only safe way to do it for any print. It's going to be expensive AF, overengineered but perfect, and most of us will go with a robot arm someone starts selling on etsy or something. I've gotta think some print farm has a home made solution and I just haven't seen it, or they haven't shared it. |
Uh oh!
There was an error while loading. Please reload this page.
I've been wondering about this for some time and it seems that someone's actually implemented it. This is a feature request based on a demo online: https://x.com/mattshumer_/status/2104777202684301750
All reactions