Dr. Zsolnai reviews a new physics simulator that handles thin cloth and sheets without penetration, making realistic tablecloth pulls finally possible. The method runs in minutes per frame on CPU and could speed up on GPU. Examples include stacked sheets, noodles, hair braids, and granular materials.
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today is a glorious day because we are going to witness the first tablecloth pull simulation I have ever seen. Well, I hoped that it would go a little more glorious than this. Maybe if we pull a bit quicker. Yes, there we go. And we're good. Loving it. Now, if you're a seasoned fellow scholar, you might remember from about 50 videos ago that we covered the predecessor of this paper called Incremental Potential Contact ipc in short. So what could it do? It could perform seriously impressive squishing experiments. And it also passed the tendril test where we threw a squishy ball at a glass wall and washed this process from the other side. A beautiful and rare sight indeed. Unless you have a cat and a glass table at home, of course. Outstanding. So I hear you asking, are you trying to say that this new paper tops all that? Yes, that is exactly what I'm saying. The tablecloth pulling is one thing, but it can do so much more. You can immediately start holding onto your papers and let's go. This new variant of IPC is capable of simulating super thin materials and all this in a penetration free manner. Now why is that so interesting or difficult? Well, that is quite a challenge. Remember this earlier paper with the Barbarian ship? Tons of penetration artifacts. And that is not even a thin object. Not nearly as thin as this stack would be. Let's see what a previous simulation method would do. If these are 10mm each, that looks reasonable. Now let's cut the thickness of the sheets in half. Yes. Some bumpy artifacts appear and at one mm, my goodness, it's only getting worse. And when we plug in the same thin sheets into the new simulator, all of them look good. And what's more, they can be simulated together with other elastic objects without any issues. And this was a low stress simulation. If we use the previous technique for a higher stress simulation, this starts out well, until, uh oh, the thickness of the cloth is seriously decreasing over time. That is not realistic. But if we plug the same scene into the new technique. Now that is realistic. So what is all this good for? Well, if we wish to simulate a bowl of noodles, tons of thick objects, let's see if we can hope for an intersection free simulation. Let's look under the hood. And there we go. All of the noodles are separated. But wait a second. I promised you thin objects. These are not thin. Yes, now these are thin. Still no intersections. That is absolutely incredible. Other practical applications include simulating hair braids, in particular granular materials against A thin sheet work too. And if you have been holding onto your paper so far, now squeeze that paper because the authors promise that we can even simulate this in hand shuffling technique in a virtual world. Well, I will believe it when I see it. Let's see. My goodness, look at that. Love the attention to detail where the authors color coded the left and right stack so we can better see how they mix and whether they intersect. Spoiler alert, they don't. What a time to be alive. It can also simulate this piece of cloth with a ton of detail, and not only that, with large time steps, which means that we can advance the time after each simulation step in bigger packets, thereby speeding up the execution time of the method. I also love how we get a better view of the geometry changes as the other side of the cloth has a different color. Once again, great attention to detail. Now we are still in the minutes per frame region and note that this runs on your processor. Therefore, if someone can implement this on the graphics card in a smart way, it could become close to real time in at most a couple papers down the line. And this is a research paper that the authors give away to all of us free of charge. How cool is that? Thank you so much for creating these miracles and just giving them away for free. What a noble endeavor research is. Perceptilabs is a Visual API for TensorFlow, carefully designed to make machine learning as intuitive as possible. This gives you a faster way to build out models with more transparency into how your model is architected, how it performs and how to debug it. And it even generates visualizations for all the model variables and gives you recommendations both during modeling and training, and does all this automatically. I only wish I had a tool like this when I was working on my neural networks during my PhD years. Visit perceptilabs.com papers and start using their system for free today. Our thanks to Perceptilabs for their support and for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
