# Simulating Dragons Under Cloth Sheets! 🐲

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=ooZ9rUYOFI4
- **Дата:** 31.10.2020
- **Длительность:** 5:52
- **Просмотры:** 386,776
- **Источник:** https://ekstraktznaniy.ru/video/14046

## Описание

❤️ Check out Weights & Biases and sign up for a free demo here: https://www.wandb.com/papers 
❤️ Their mentioned post is available here: https://app.wandb.ai/stacey/droughtwatch/reports/Drought-Watch-Benchmark-Progress--Vmlldzo3ODQ3OQ

📝 The paper "Local Optimization for Robust Signed Distance Field Collision" is available here:
https://mmacklin.com/sdfcontact.pdf

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Haro, Alex Paden, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bruno Mikuš, Bryan Learn, Christian Ahlin, Eric Haddad, Eric Lau, Eric Martel, Gordon Child, Haris Husic, Javier Bustamante, Joshua Goller, Lorin Atzberger, Lukas Biewald, Michael Albrecht, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Ramsey Elbasheer, Robin Graham, Steef, Sunil Kim, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Tybie Fitzhugh.
If you wish to support the series, click here: https://www.patreon.com/TwoMinutePape

## Транскрипт

### Introduction []

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today, we are going to see a lot of physics  simulations with many-many collisions. In   particular, you will see a lot of beautiful  footage which contains contact between thin   shells and rigid bodies. In this simulation  program, at least one of these objects will   always be represented as a signed distance field.   This representation is useful because it helps us   rapidly compute whether something  is inside or outside of this object. However, a collision here takes two objects, of  course, so the other object will be represented   as a triangle mesh, which is perhaps the  most common way of storing object geometry in   computer graphics. However, with this, we have  a problem. Signed distance fields work great,   triangle meshes also work great for a number  of applications, however, computing where they   overlap when they collide is still slow and  difficult. If that does not sound bad enough,   it gets even worse than that. How so? Let’s  have a look together. Experiment number one.

### Experiments [1:06]

Let’s try to intersect this cone with this  rubbery sheet using a traditional technique,   and there is only one rule - no poking through  is allowed. Well, guess what just happened. This   earlier technique is called point sampling,  and we either have to check too many points   in the two geometries against each other,  which takes way too long and still fails, or,   we skimp on some of them, but then, this happens.   Important contact points go missing. Not good. Let’s see how this new method does with  this case. Now that’s what I am talking   about! No poking through anywhere to  be seen. And, let’s have another look.    Wait a second. Are you seeing what I am seeing?   Look at this part…after the first collision,   the contact points are moving ever so slightly,  many-many times, and the new method is not missing   any of them. Then, things get a little out of  hand, and it still works perfectly. Amazing!    I can only imagine how many of these  interactions the previous point sampling   technique would miss, but we won’t know  because it has already failed long ago. Let’s do another one. Experiment number two.   Dragon versus cloth sheet. This is the previous   point sampling method. We now see that it can  find some of the interactions, but many others   go missing, and due to the anomalies, we can’t  continue the animation by pulling the cloth sheet   off of the dragon because it is stuck. Let’s see  how the new method fared in this case. Oh yeah!    Nothing pokes through, and therefore, now, we can  continue the animation by doing this. Excellent! Experiment number three! Rope curtain. Point  sampling. Oh no! This is a disaster. And now,   hold on to your papers, and marvel at the new  proposed method. Just look at how beautifully   we can pull the rope curtain through this fine  geometry. Loving it. So, of course, if you are   a seasoned fellow scholar, you probably want  to know, how much computation do we have to do   with the old and new methods. How much longer do  I have to wait for the new, improved technique?

### Results [3:38]

Let’s have a look together! Each rigid shell here  is a mesh that uses a 129 thousand triangles, and   the old point sampling method took 15 milliseconds  to compute the collisions, and this time, it has   done reasonably well. What about the new one?   How much more computation do we have to perform   to make sure that our simulations are more robust?   Please stop the video and make a guess. I’ll wait.    Alright, let’s see…and, the new one does it in  half a millisecond. It is   not slower at all, quite the opposite - thirty  times faster. My goodness! Huge congratulations   on yet another masterpiece from scientists  at NVIDIA and the University of Copenhagen. While we look at some more results, in case  you are wondering, the authors used NVIDIA’s   Omniverse platform to create these amazing  rendered worlds. And now, with this new method,   we can infuse our physics simulation programs with  a robust, and blazing fast collision detector, and   I truly can’t wait to see where talented artists  will take these tools. What a time to be alive!

### Outro [4:49]

Thanks for watching and for your generous  support, and I'll see you next time!