# Roblox Solved The Physics Problem That Stumped Everyone! ## Метаданные - **Канал:** Two Minute Papers - **YouTube:** https://www.youtube.com/watch?v=TzIKbjuSy2A - **Дата:** 13.07.2025 - **Длительность:** 6:19 - **Просмотры:** 352,873 - **Источник:** https://ekstraktznaniy.ru/video/12258 ## Описание ❤️ Check out Vast.ai and run DeepSeek or any AI project: https://vast.ai/papers 📝 The paper is available here: https://graphics.cs.utah.edu/research/projects/avbd/ Play with it! https://graphics.cs.utah.edu/research/projects/avbd/avbd_demo2d.html 📝 My paper on simulations that look almost like reality is available for free here: https://rdcu.be/cWPfD Or this is the orig. Nature Physics link with clickable citations: https://www.nature.com/articles/s41567-022-01788-5 🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible: Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Sven Pfiffner, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi If you wish to appear here or pick up other perks, click here: https://www.patreon.com/TwoMinutePapers My research: https://cg.tuwien.ac.at/~zsolnai/ X/Twitter: https://twi ## Транскрипт ### Segment 1 (00:00 - 05:00) [] This is an amazing new free and open simulation method that solves an almost impossible problem. And it's a bit like a virtual playground for grown-up people. And somehow we even get paid to run experiments like this. And I'll give you a link to it so you can try a mini version of it right now yourself at home for free. It is super mega fun. Now hold on to your papers, fellow scholars, and check this out. This collaboration between Roblox and the University of Utah can run this amazing scene easily faster than real time. Yep. At 100 frames per second on one consumer graphics card. Goodness. Now, a previous version of it was called Vertex Block Descent, VBD, that could simulate these incredible scenes that are modeled as an interaction of millions and millions of points. This would be incredible to have in all kinds of games and movies. So, are we done here? Why write this newer paper? Do we even need more research here? Well, the previous VBD technique was not perfect. In fact, you'll see a bunch of previous methods fail on cases that seem totally baffling. Just to make sure whenever we talk about number of iterations that means how much computational effort we spend on each frame to get things right. Of course higher is better but we also have to wait for longer. In return the simulation gets more accurate. However the question is how much better? For instance when using the previous method to roll a few balls at a house of cards it doesn't matter how many iterations you use here. This is flatout incorrect. Why? Well, suddenly collisions are not working properly because it generates an excessive amount of friction. Wow, that is unexpected. However, what I found to be even more unexpected. What? Are you kidding me? You'd think simulating a simple pendulum would be a solved problem, right? We've been doing this for decades. Well, so far so good. What could possibly go wrong? Well, this oof the old methods completely break down. The chain stretches like a piece of gum. How could they fail on something so simple? And then the new technique, augmented vertex block descent, EVBD, rock solid, is the moment you realize even the simple things weren't truly solved until now. Bravo. So, what is the problem? The problem is that the mass of the ball is 50,000 times higher than the chain holding it, which is given by 50 connected bodies. Imagine having a Tesla Model S hanging off of a few Lego bricks. Of course, it results in an excessive amount of stretching with previous methods, but not with the new one, which is the correct solution. Amazing. And when we throw a ball through a chain mail, once again, previous methods do something. Well, unfortunately, this is incorrect. All of them are incorrect. Given the physical properties of the scene, the collision constraints cannot overcome the momentum of the ball. However, the chain mail is supposed to hold up the ball. So, does it with the new method? Wow, once again, so much better. loving this. But now it gets even more crazy. Dear fellow scholars, this is two minute papers with Dr. Ko Jaer. Now hold on to your papers, fellow scholars, because even a few blocks connected by springs can give the previous method a headache. So just run it for longer, right? More iterations. Well, not so much. Look. And this is the part where I fell off the chair when reading this paper. The true solution would be a perfectly straight arrangement. And the old method is still sagging even with 100 iterations. And the new one, wow, even at one iteration, it is close to perfect. Much better than the previous one at 100. This is absolutely incredible. And once again, my heart is aching because almost no one is talking about this paper. You can likely only hear about it here on Two Minute Papers. Subscribe and hit the bell icon if you wish to see more like this and support us on this journey because it is not easy. YouTube is not recommending this kind of content that much if you don't help out. And to think that all this improvement happened in just one year and it is all handcrafted powered solely by human ingenuity and we get all of this for free. What a time to be alive. So how does it do all this black magic? You see it is called ### Segment 2 (05:00 - 06:00) [5:00] augmented vertex block descent. And the augmented part means that instead of being too rigid from the start, it gradually adjusts how strict it is about enforcing rules during physics simulation. It's a bit like a bouncer at the club. A bouncer's job is to keep people behind the velvet rope. And if one person leans on the rope, the bouncer might just put a hand up. But if a group of people starts pushing, the bouncer plants their feet and pushes back harder. Their effort adapts to how much the rule is being broken. And that is what this technique does. Genius. And yes, I left a link in the video description so you can try a mini version of it yourself for free right now. And it is super fun. The source code is also available for it. A great contribution to humanity. Thank you so much. And if you wish to run Deep Seek yourself, Vest AI helps you rent Nvidia GPUs to do that for an incredible price. I love it because my costs come down to a fraction of a dollar for a session. It is incredible. Make sure to try it out today at vast. ai/papers AI/papers or click the link in the description.