--- title: Someone built a 200k rigidbody CUDA simulator and the... description: A developer known as Yacine built a custom rigid body physics simulator on top of box3d, running entirely on CUDA, capable of handling up to 200,000 rigid... canonical: https://daily.dev/posts/someone-built-a-200k-rigidbody-cuda-simulator-and-the-microduck-walks-zxlgb1tc9 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Someone built a 200k rigidbody CUDA simulator and the microduck walks | daily.dev og:description: A developer known as Yacine built a custom rigid body physics simulator on top of box3d, running entirely on CUDA, capable of handling up to 200,000 rigid... og:url: https://daily.dev/posts/someone-built-a-200k-rigidbody-cuda-simulator-and-the-microduck-walks-zxlgb1tc9 og:image: https://api.daily.dev/og/posts/Zxlgb1tc9.png og:image:alt: Someone built a 200k rigidbody CUDA simulator and the microduck walks og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Someone built a 200k rigidbody CUDA simulator and the microduck walks **[Trends](https://daily.dev/sources/trends)** · 2 min read · 2 upvotes · 0 comments ## Summary A developer known as Yacine built a custom rigid body physics simulator on top of box3d, running entirely on CUDA, capable of handling up to 200,000 rigid bodies with accurate enough contact dynamics to make a small simulated duck walk. The project is notable for deliberately avoiding MuJoCo, the dominant physics engine in robotics and sim-to-real research since Google acquired it in 2022. Legged locomotion is a demanding test of contact physics accuracy, so a walking duck suggests the simulator's collision and friction modeling holds up. The demo spread after Hugging Face's Clement Delangue and Thomas Wolf retweeted it, though no technical writeup, benchmarks against MuJoCo, or open-source release have been shared yet. ## Content Yacine (@yacineMTB) just dropped a flex that's getting passed around ML circles: a custom rigid body simulator, built on box3d, running entirely on CUDA, accurate enough to make a tiny duck walk. No MuJoCo. That's the point he's making explicitly. MuJoCo is the default physics engine everyone reaches for in robotics and sim-to-real research — it's fast, well-tested, and has Google's backing since the 2022 acquisition. Building your own simulator from scratch that can handle up to 200,000 rigid bodies and still get the contact dynamics right enough for locomotion is a genuinely hard problem. The microduck walking is the proof-of-concept here. Legged locomotion is a classic stress test for physics simulators because it requires accurate contact modeling — get the friction or collision response slightly wrong and the thing just falls over or skates around uselessly. If the duck walks, the contact physics is probably solid. Hugging Face's Clement Delangue and Thomas Wolf both retweeted it, which is how it's spreading. No detailed technical writeup yet — just the claim, the demo, and the CUDA flex. The obvious next questions (how does it compare to MuJoCo on speed per body, what's the accuracy tradeoff at scale, is this going open source) are unanswered so far. But "200k rigidbodies on CUDA, no MuJoCo" is the kind of thing that makes simulation researchers pay attention. ## Community take How the wider developer community reacted, aggregated from 3 discussions and 59 comments across x (as of 2026-08-30). **TL;DR:** Reactions are mostly playful hype and duck-torture jokes, with a running bit about demanding the code get released on GitHub and some pushback questioning how original/novel the work actually is. **Sentiment:** 40% positive · 45% mixed · 15% skeptical **The case for** - Several call the 200k rigidbody count and the walking duck genuinely impressive. - Some see it as a fun, well-executed side project worth exploring further. **The pushback** - One commenter points out it's not a from-scratch build but a CUDA translation of Erin Catto's existing box3d, done with help from AI tools. - Repeated demands that the project actually be open-sourced on GitHub, suggesting skepticism it will ever be released. - A commenter questions whether a solo effort can realistically beat existing SOTA paid/OSS solutions. **By community** - x (mixed): Mostly jokey engagement about the duck's ordeal and calls for the code to be released, with a bit of genuine skepticism about originality and whether it beats existing tools. **Hottest debate:** Whether this is an impressive original achievement or just a CUDA port of someone else's existing box3d work. **Open questions** - Will the simulator actually be released on GitHub as promised? - How does it perform against MuJoCo or other established physics engines in practice? **Highlights** > @AlpinistXBT @yacineMTB No, there isn't. And he didn't build it himself, he translated Erin Catto's box3d to cuda. So it was made by 2 people (plus AI). > — [cacomoneta on x · 1 points, 1 comments](https://x.com/cacomoneta/status/2093837775019942138) > @yacineMTB Im out of the loop, what’s the benefit of this? Surely there’s SOTA paid or OSS solution better than what 1 man (you) can build himself? > — [AlpinistXBT on x · 2 comments](https://x.com/AlpinistXBT/status/2093794594580304142) > @yacineMTB yes you can pay me any time you like, but I will accept you just putting this up on github so I don't have to sue over broken obligatory promises. we can do this the easy way or the hard way. > — [Sentdex on x · 2 points](https://x.com/Sentdex/status/2093797576193593588) > @alirezaafshan My robotics training sim is on sale for 6 million rn > — [yacineMTB on x · 7 points](https://x.com/yacineMTB/status/2093856822612611479) > @yacineMTB Bro built a rigid body simulator that laughs at 200k objects and still lets a tiny duck walk around without glitching. No Mujoco. Just pure CUDA pain and skill. Respect. > — [AnnaFadar2000 on x](https://x.com/AnnaFadar2000/status/2093783377342001557) **Source threads** - [x](https://x.com/Thom_Wolf/status/2093787445867298825) · 0 points · 0 comments - [x](https://x.com/ClementDelangue/status/2093776641197510835) · 0 points · 0 comments - [x](https://x.com/yacineMTB/status/2093769774786384058) · 0 points · 59 comments --- Tags: [#gpu](https://daily.dev/tags/gpu), [#robotics](https://daily.dev/tags/robotics), [#physics](https://daily.dev/tags/physics), [#cuda](https://daily.dev/tags/cuda) [View this post on daily.dev](https://daily.dev/posts/someone-built-a-200k-rigidbody-cuda-simulator-and-the-microduck-walks-zxlgb1tc9) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Someone built a 200k rigidbody CUDA simulator and the microduck walks","url":"https://daily.dev/posts/someone-built-a-200k-rigidbody-cuda-simulator-and-the-microduck-walks-zxlgb1tc9","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/someone-built-a-200k-rigidbody-cuda-simulator-and-the-microduck-walks-zxlgb1tc9"},"datePublished":"2026-08-29T19:47:05.512Z","dateModified":"2026-08-30T01:48:01.391Z","description":"A developer known as Yacine built a custom rigid body physics simulator on top of box3d, running entirely on CUDA, capable of handling up to 200,000 rigid...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/someone-built-a-200k-rigidbody-cuda-simulator-and-the-microduck-walks-zxlgb1tc9","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"gpu,robotics,physics,cuda","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Someone built a 200k rigidbody CUDA simulator and the microduck walks"}]} ```