We taught robots to carve marble sculptures

(eleanorwiesler.com)

52 points | by ewiesler 21 hours ago

14 comments

  • mauvehaus 59 minutes ago
    Two things:

    1) A great podcast about why we're bad at statues in the US:

    https://slate.com/podcasts/decoder-ring/2026/07/how-taste-po...

    2) This isn't a particularly interesting model to sculpt. The parent model is a scan of a ceramic rabbit, and it doesn't have crisp details like really good marble sculptures do.

    I have seen Venus Victrix[0] in person, and the detail in the mattress the figure is lying on is absolutely transfixing. You would swear the marble would softly yield if you put your finger on it. There's a detail rendered around the edge of the mattress that likely represents embroidery and it's unbelievably crisp, even as it follows the folds and creases in the mattress.

    If we're going to have robots sculpt marble, that is the standard we should be reaching for. If we're reaching for that, I think we're going to find that modeling the finished product in 3d is going to be a significant fraction of the work of actually carving it in stone. I think we're also going to find that the mastery also lies in the tiny decisions the sculptor makes as they work, informed by how light plays off the material in different directions in real time throughout the day.

    Can the process in the article achieve results suitable for mounting at the top of a building and viewing from ground level? Probably, and if not now, certainly with time. Can it sculpt something that will command your attention in reverent, slack-jawed awe for hours? I don't see it happening.

    [0] https://en.wikipedia.org/wiki/Venus_Victrix_(Canova) Seriously, this and Bernini's David, also at the Galleria Borghese, are worth the trip to Rome to see. If there were nothing else in Rome worth seeing and the food was terrible, they would justify the pilgrimage on their own.

  • numpad0 16 minutes ago
    I don't get it. This is what everyone do when they need a giant mockups of things like new model of a cars or boats or giant pharaoh statues, except it's usually done with styrofoam.

    Is there some technical novelty to it? The hero image on the company website is an Arc de Triomphe that doesn't seem to be the one in Paris, it is like they are a startup specializing in quickly producing artistic stone decor for projects like the arc?

    https://www.youtube.com/watch?v=qJgdijvbJWg

    https://www.youtube.com/watch?v=kyHz90gA21s

    https://www.youtube.com/watch?v=ztGFtKJCN-M

    https://www.youtube.com/watch?v=N0T6Po0TIK0

  • runxel 52 minutes ago
    Reminds me of the link to the Carmelites[0] that was on HN a few weeks ago.

    But they do it in stone (they don't seem to mention what kind of stone, tho). I think marmble is a different experience, and I wonder what the implications are. Often enough marble requires an experienced master to "read" it. Since it's a natural stone there will be veins, imperfections etc. Sometimes you'll have to work around it and alter your design slightly while carving - which is not possibble if a robot just does its lusterless job.

    While I admire this task, and I hope that we will see more carved stones in our buildings and surroundings again, I also hope that we can use technology as our aid, not as the end in itself.

    There is a manufacturer in Germany who does exactly this, btw (rough cuts by robot, fine tune by human masters). So there is a lot of prior art out there - I wonder if there was anything new at all.

    [0] https://carmelitegothic.com/cnc-stone-carving/

  • voidUpdate 2 hours ago
    What part is being "taught"? It looks like a fairly standard CAM pipeline to me...
    • GrinningFool 1 hour ago
      In the before times, we might say we "taught" a system to do something and mean that we designed/implemented the feature.
      • suddenlybananas 2 minutes ago
        Did we? "We taught a computer to send emails"? Sounds bizarre to me.
    • amelius 1 hour ago
      Maybe they used an LLM?
      • Overdrivr 36 minutes ago
        Honestly for CAM you absolutely need deterministic output (Speaking from experience, I built automated CAD to CAM pipelines in the past using Python+Siemens NX). IMO the true innovation is the algorithm that they must have built that turns a target voxel into a 5+-axis machining toopath, but that's good old fashioned algorithms, and also getting to a 100% automated CAM generation. Achieving 95% automation is easy, 100% is very hard
        • voidUpdate 17 minutes ago
          There are a couple of mentions of "machine learning", in the parts about generating the toolpaths and connecting the toolpath segments into a continuous path, but I'm still not sure I'd call it teaching robots
  • zenapollo 29 minutes ago
    I watched this in a museum in Prague on a big screen, it's incredibly and enjoyably artful. That being said I couldn't help notice the final texture felt off. I suspect the techniques are already improving. I watched it Nov 26 but apparently according to yt it's several years old already. I also don't think this one is AI based at all.

    https://youtu.be/fVga1ay2DZk

  • pwmglenn 1 hour ago
    It's cool, but it kind of diminishes the value. I could imagine a future where marble sculptures are seen as trashy, because of their ubiquity.
    • fusslo 1 hour ago
      Huh, I remember discourse with the CEO (I think) of Monumental on twitter a while ago. There was a frenzy around anti-ai art at the time. If I remember right, he replied to many comments that very skilled human artisans still did the final touches on the sculptures that were cnc carved rough. It really seemed like he was taking that criticism seriously

      But now, "... our system turns a 3D model and a block of marble into a finished sculpture, with no human in the loop."

      I'm not judging one way or the other. I have 0 skin in this debate. I just find situations like that interesting; like trying to douse a publicity fire while secretly making wildfire

    • BeFlatXIII 12 minutes ago
      Arguably, this is why modern architecture caught on. Mass-production of ornament became cheap in the early 20th century.
      • suddenlybananas 1 minute ago
        But what I've heard from architects is that ornamentation is too expensive for modern buildings because of the cost of labour.
    • jackp96 1 hour ago
      Same; it's a fascinating tech exploration, and I enjoyed reading about how it works.

      But part of the reason I find carved statues to be impressive and monumental is the technical human feat and the craftsmanship. Looking at them, you can sort of feel the work they must have required.

      Mass produced goods aren't inherently bad; but there is something lost in translation for pieces like this, IMO.

    • olalonde 1 hour ago
    • martythemaniak 1 hour ago
      Yep, people think they have timeless taste, but mostly it's a reaction to their environment. If you grew up in a works surrounded by endless over the top ornamentation, clean simple lines look fresh, new and appealing. If you grew up in a world of glass boxes, you're desperate for ornamentation.
    • amelius 1 hour ago
      Maybe, but let's find out.
    • LtWorf 1 hour ago
      The material itself is not that easy to come by.v
    • deadbabe 1 hour ago
      Trashy marble sculptures already exist.
  • classified 9 minutes ago
    So this is a more flexible kind of CNC milling?
  • caidan 42 minutes ago
    “de-extinction, not displacement”

    The only voice left is Claude’s.

    It was cheaper to spend a trillion dollars on LLM’s and robotics and then “teach” a robot to carve out this first year bunny assignment rather than pay any artists or stonemasons. Historically their work can last for thousands of years and bring beauty to millions yet unborn who will marvel at the magnificence of this creation. But you know this little stone rabbit is cool too. I’m sure it will make a useful doorstop. Now we can have stoneslop!

  • cloudbonsai 1 hour ago
    I recommended seeking the first clip to 00:30 and play it from there. The perspective is somewhat confusing, but the robotic arm in use (KUKA KR240) is actually huge.
  • the__alchemist 9 minutes ago
    Bun alert!
  • ewiesler 21 hours ago
    We taught KUKA KR240 robots to carve carve marble sculptures end to end with autonomous toolpathing and simulation, starting with the Stanford bunny as a test case. Here’s how it goes from 3D model to finished sculpture using voxel representations and geodesic path planning.
    • le-mark 1 hour ago
      > We taught

      What does this mean exactly? Is the robot not being fed gcode?

    • moeffju 2 hours ago
      Cool work! Could you maybe say a few sentences about where/how specifically this improves on the current state of the art? I was under the impression that whole monasteries in Italy are running fleets of CNC robots to create marble sculptures - is your process easier to set up, does it go further, does the old process need human finishing etc.? Thank you!
    • GrinningFool 1 hour ago
      The words in the post look LLM generated. For some portion of us, that's offputting.

      Aside from that, this stood out:

      > our mission is to continue to work with artists and designers to decide on the final desired sculpture, and hand carvers to finalize the surface quality of many of our pieces.

      This sounds good on the surface (hah) but it seems like it's only valid until the technology allows that work to be replaced too.

  • curtisblaine 1 hour ago
    > Stone carving never disappeared, but rising costs pushed it out of ordinary construction

    Why "rising costs" though? Are the carving tools significantly more expensive? Or did we just find a way to do shittier stuff with less expense and we called carving too expensive?

    • voidUpdate 1 hour ago
      People like being paid in line with inflation etc, and theres likely a lot less stonemasons around today. It was probably a lot cheaper to get a marble carving made in ye olden days than now
    • iamacyborg 32 minutes ago
      An article on Works in Progress has a good breakdown of the situation.

      https://worksinprogress.co/issue/the-beauty-of-concrete/

    • bluGill 47 minutes ago
      Stone carving was always expensive. What changed is other parts of building went down and so not stone carving would be a much larger share.

      Most of us would live in closer to mud huts than a mansion 1000 years ago. The people who live in mansions still do stone carving if they feel like it, but the rest of us have moved up to much nicer houses. (mansions are much nicer too)

    • sokka_h2otribe 51 minutes ago
      From wikipedia: "Baumol’s cost disease (or the Baumol effect) is an economic theory explaining why wages and prices in labor-intensive services—like healthcare, education, and the performing arts—rise even when their productivity does not."
    • gauntr 1 hour ago
      The amount of people that actually know and do that kind of work has probably become very small while the demand might still be there. High demand, low supply -> usually increasing prices. There's generally a big decline in skilled trades and with that the same problem is coming up in other sectors as well.
    • LtWorf 1 hour ago
      I think estate developers just managed to somehow convince our regulators that copy pasting the same building across the world is unique and modern and we should do it.
      • bluGill 36 minutes ago
        For good reason. There are things that work in building layouts and things that don't. Everything looks the same, but the inside that you work with is often very functional. Even in the most work of art buildings, the building itself will fad into the background of all the things you do inside - unless the building is hostile to doing the things you are inside to do.
  • pentakkusu 29 minutes ago
    Marbleslop
  • sandworm101 1 hour ago
    The profit chain will be how quickly they can replicate a block from a scan of a real block in need of restoration. I have seen countless documentaries on the restoration of old buildings, with the inevitable interviews with stonemasons about how many hours go into each block. The ability to auto-chisel a replica from a scan of an original would radically reduce the cost of restoring old structures.
    • Arodex 1 hour ago
      Redoing the blocks by hand is the whole point. It is experimental archaeology, on top of preserving (and rediscovering lost) old techniques. Your suggestion is useless, counterproductive (where will we find the subject matter experts in old techniques once you kill their jobs?) and ridiculous ("profit chain"... Do you even know how little money there is in restoration?).

      >the inevitable interviews with stonemasons about how many hours go into each block

      Yeah, it is a point a pride, not a problem to be solved. And they will put their signature on the blocks they sculpted, like their predecessors and the predecessors before them, and someone in a century or a millenium will notice it. That's a form of eternity the tech bros will never attain.

      • Whitespace 59 minutes ago
        > Redoing the blocks by hand is the whole point.

        I agree. I also think that only 0.1% of existing marble will be redone by hand. The rest will either turn to dust or be replaced by machines.