AI driven 3D bin packer written in F#

3dpack.ing

3 points·reverseblade2··5 comments

So the AI part is because the customers state their problem in plain English? In your experience, do they prefer that rather than a GUI?

Incidentally, what method do you use to solve the problem once the LLM gives you a formulation? Or is the LLM itself tasked with solving the problem directly?

Hi, Yes AI merely acts as an interface, not part of the algorithm. It's because trying to add everything as a UI, I personally find it challenging. Also the options can exponentially grow. In the future I plan to add more sophisticated features as in

"- Do not add Clothes over Shoes, while Shoes and Buttons can be stacked over each other but do not exceed 140 cm height."

This in turn can be translated into a json level config but it is difficult to acquire user input with plain forms.

There are 2 different algorithms. One is adding each box partitions space into 3 and those spaces are recursively filled. Eventually the problem is reduced to finding the correct permutation of boxes along with their correct orientation.

That is primarily derived by simulated annealing of which I had good experience with combinatory problems. Also some magic heuristics which I also don't fully understand takes into play :)

Hello,

I have written a 3D bin packer fully in F# https://3dpack.ing

I plan to open source the frontend. And a fork of older version of the algorithm is available here (albeit that version is very slow)

Sorry forgot the repo link

https://github.com/goswinr/BinDrake

PS. I am convinced "bin packing" problems are related (in AI) to an AI's "proprioception" (or, something that might lead to it), and that this - in turn - a "sense" of "body" (or/and a body) ...

... is somewhat of a prereq for AGI.-

Reply on news.ycombinator.com