Nvidia and University of Copenhagen researchers built a deformation simulation that runs in real time. A car with tire deformation hits 55 fps, a rolling coin reaches 3000+ fps, and the method handles collisions and complex shapes without breaking.
Questions this video answers
18 milliseconds per frame (ā55 frames per second).
About 1/3 of a millisecond per frame (ā³3,000 frames per second).
Nvidia and the University of Copenhagen.
Chapters
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai today. With the power of computer graphics research, we can use our computers to run fluid simulations simulate immersing a selection of objects into jelly or tear meat in a way that much like in reality, it tears along the muscle fibers. If we look at the abstract of this amazing new paper, we see this quoting this allows us to trace high speed motion of objects colliding against curved geometry to reduce the number of constraints, to increase the robustness of the simulation and to simplify the formulation of the solver. What? This sounds impossible, but at the very least outrageously good. Let's look at three examples of what it can do and see for ourselves if it can live up to its promise. 1. It can simulate a steering mechanism full of joints and contact. Yep, an entire servo steering mechanism is simulated with the prescribed mass ratio. Loving it. I hereby declare that it passes inspection. And now we can take off for some off roading. All of the movement is simulated really well. And. Wait a minute, hold on to your papers. Are you seeing what I am seeing? Look, even the tire deformations are part of the simulation. Beautiful. And now let's do a stress test and race through a bunch of obstacles and see how well those tires can take it. At the end of the video I will tell you how much time it takes to simulate all this. And note that I had to look three times because I could not believe my eyes. 2. Restitution. Or in other words, we can smash an independent marble into a bunch of others and their combined velocity will be correctly computed. We know for a fact that the computations are correct because when I stop the video here, you can see that the marbles themselves are smiling. The joys of curve geometry and specular reflections. Of course this is not true, because if we attempt to do the same with a classical earlier technique by the name position based dynamics, this would happen. Yes, the velocities became erroneously large and the marbles jump off of the wire. And they still appear to be very happy about it. Of course, with the new technique, the simulation is much more stable and realistic. Talking about stability, is it stable only in a small scale simulation or can it take a huge scene with lots of interactions? Would it still work? Well, let's run a stress test and find out. Aha. This animation can run all day long and not one thing appears to behave incorrectly. Loving this. Three, it can also simulate these beautiful high frequency rolls that we often experience when we drop a coin on a table. This kind of interaction is very challenging to simulate correctly because of the high frequency nature of the motion and the curved geometry that interacts with the table. I would love to see a technique that algorithmically generates the sound for this. I could almost hear its sound in my head. Believe it or not, this should be possible and is subject to some research attention in computer graphics. The animation here was given, but the sounds were algorithmically generated. Listen, Let me know in the comments if you are one of our OG fellow scholars who were here when this episode was published hundreds of videos ago. So how much do we have to wait to simulate all of these crazy physical interactions? We mentioned that the tires are stiff and take a great deal of computation to simulate properly. So as always, all nighters, right? Nope. Look at that. Holy mother of papers. The CAR example takes only 18 milliseconds to compute per frame, which means 55 frames per second. Goodness. Not only do we not need an all nighter, we don't even need to leave for a coffee break. And the rolling marbles took even less. And woohoo. The high frequency coin example needs only 1/3 of a millisecond, which means that we can generate more than 3,000 frames with it per second. We not only don't need an all nighter or a coffee break, we don't even need to wait at all. Now at the start of the video I noted that the claim in the abstract sounds almost outrageous. It is because it promises to be able to do more than previous techniques and simplify the simulation algorithm itself, make it more robust, and do all this while being blazing fast. If someone told me that there is a work that does all this at the same time, I would say that give me that paper immediately because I do not believe a word of it. And yet it really lives up to its promise. Typically, as a research field matures, we see new techniques that can do more than previous methods, but the price to be paid for it is in the form of complexity. The algorithms get more and more involved over time, and with that they often get slower and less robust. The engineers in the industry have to decide how much complexity they are willing to shoulder to be able to simulate all of these beautiful interactions. Don't forget, these code bases have to be maintained and improved for many, many years, so choosing a simple base algorithm is of utmost importance. But here none of these factors need to be considered because there is nearly no trade off. Here it is simpler, more robust and better. At the same time it really feels like we are living in a science fiction world. What a time to be alive. Huge congratulations to scientists at Nvidia and the University of Copenhagen for this. Don't forget they could have kept the results for themselves, but they chose to share the details of this algorithm with everyone free of charge. Thank you so much for doing this. What you see here is a report for a previous paper that we covered in this series which was made by Weights and Biases.
Weights and Biases provides tools to track your experiments in your deep learning projects. Their system is designed to save you a ton of time and money and it is actively used in projects at prestigious labs such as OpenAI, Toyota Research, GitHub and more. And the best part is that if you have an open source, academic or personal project you you can use their tools for free. It really is as good as it gets. Make sure to Visit them through wnb.com papers or click the link in the video description to start tracking your experiments in five minutes. Our thanks to WeightsAndBiases for their long standing 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.
