Skip Navigation
Lego @lemm.ee
Blaze (he/him) @sopuli.xyz

LegoGPT: Generating Physically Stable and Buildable Lego

avalovelace1.github.io

Generating Physically Stable and Buildable LEGO Designs from Text

cross-posted from: https://lemmy.bestiver.se/post/372778

Comments

4 comments

  • This doesn’t seem all that innovative to me.

    It seems like the process for generation is:

    1. Parse the input command and generate it as a 3D object (not yet in Lego form). There is already software that can do this.
    2. Translate the 3D object into Lego form using strictly basic bricks. This technology has been around for a while. This video doesn’t show this process directly but it is used in jobs such as these.
    3. Conduct a stability check. This is something that Lego specific software such as Stud.io already does.

    I suppose it is the first case in which I’ve seen all of these steps come together?

    There are also a couple of flaws that I see right away:

    • The publication emphasizes how either a human or a robot arm can build the program’s output. However, they overlook the need for subassemblies in some assemblies. The render of the “high-backed chair” example cannot be built from the bottom up as shown in the article because there would be pieces floating in midair until another layer is added; you’d have to either build subassemblies or invert portions of the build, both of which the robot arms seem incapable of doing.
    • You don’t seem able to set the scale of the build? You might want a build that is specifically 1:20 scale or 10 bricks long or 5 inches tall, for example.
    • [EDIT: Because Lego bricks are not dimensioned as cubic units, the orientation of the bricks relative to the assembled model may impact how well the model “reads” as what it’s supposed to be. As in, having the topside of the individual bricks oriented 90° from the topside of the assembly (i.e., SNOT) may produce better looking results than if they were aligned. The program doesn’t seem to account for this.]
    • The application of painted decoration forces that level of detail to remain digital, or for whomever is artistically disinclined enough to have a machine make a rudimentary Lego build for them to somehow be artistically inclined enough to paint that level of detail.
    • Omitting the painted decorations, the program doesn’t seem to consider what parts actually exist in which colors; this is also something that Stud.io can do. Similarly, it would be helpful to curate the parts used based on their cost to acquire.
    • The use of only basic bricks is very limiting. This isn’t a major critique since those types of builds do have their uses.

    Lastly, it seems like this is just another case of using ”AI” to supplant where humans can find fun and expression in this world, instead of reducing labor. I say this since the program seems targeted toward small scale commercial use, in contrast to the large industrial usage in the video I linked to above.

    If there is something about this program that I missed or misunderstood, please clarify for me.

    [EDIT: I found a paper with more information. I’ll have to read through it later.]