Vibe Coding Robotics | Demo 12 Second time building Theo Jansen's Strandbeest simulator The first time, I couldn't get any model to build the crankshaft correctly This time, Fable managed to pull it off

This is the first demo I made a couple of months ago. Even though I got the linkage working, no model I tried could build the crankshaft correctly.

感觉像只猫

so cool

thanks

that cool！

thanks

Very nice!

thanks

Did it use machine learning to generate the exact right beam lengths, like Jansen himself did on an amiga?

No, this one uses the set of beam lengths that's already well known. I will do some tests to see if the models can use the same kind of evolutionary algorithms Jansen used to generate entirely new designs. From what I've seen while experimenting with other linkage systems, models like Fable can come up with novel mechanisms. The challenge is that they don't really know whether the design is mechanically correct, so they make a lot of mistakes that take quite a few follow-up prompts to fix.

hmm yeah I figured it doesn't automatically does the physics/spatial simulations right the first time... still would be interesting if it could come up with something better, since the originals were done on an old amiga computer if I remember correctly. no wait it was an atari!

Good inspiration.

No coding. No tech skills. Just your ideas. 🎁 Sign up now and get free credits.

Every failed crankshaft proves the same point: physics ignores your vibes. You can't iterate your way out of bad boundary conditions. Good to see this one walking.

두 번째 시도에서 완성된 게 오히려 핵심 같아요. 첫 시도에서 크랭크축 구현이 막혔던 지점을 그냥 넘긴 게 아니라, 그 실패를 AI 도구에 다시 먹여서 구조를 풀어낸 거잖아요. AI 활용은 한 번에 정답 받는 기술보다, 막힌 부분을 더 빨리 특정하고 다음 시도로 넘기는 루프에 가까운 것 같아요 ㅎㅎ

Sir I copied yours project 😜. I got so worse animation.. If possible i want your patent.