# The Simulator That Could Supercharge Robotics! ## Метаданные - **Канал:** Two Minute Papers - **YouTube:** https://www.youtube.com/watch?v=WZCOtnUT0b0 - **Дата:** 04.01.2025 - **Длительность:** 5:51 - **Просмотры:** 48,091 - **Источник:** https://ekstraktznaniy.ru/video/17203 ## Описание ❤️ Check out Weights & Biases and sign up for a free demo here: https://wandb.me/papers 📝 The paper is available here: https://github.com/Genesis-Embodied-AI/DiffTactile https://difftactile.github.io/ 📝 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: Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, 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://twitter.com/twominut ## Транскрипт ### Segment 1 (00:00 - 05:00) [] It’s crazy that no one is talking about this.   This work was done as a collaboration between   several research labs across the world and  is about creating physics simulations on   a computer to teach AIs about the world.   And what are the simulations about? Well,   to help robots noodle around with things. Yes, I hear you saying Károly, what  does that mean? Let me try to explain. When we try to teach a robot to  navigate safely in the real world,   first, we don’t just drop them on the street and  let them learn the basics there, not at all. What  we do is that they put it in a simulated  environment where we can give it crazy hard tasks,   and when it has demonstrated that it has  learned everything and is safe, then we can   consider testing them in reality. This is called  sim to real. Going from simulation to reality. Self-driving cars and robots are doing  better and better in this regard,   we have these little computer games where  they can train. But for everyone else,   there are two huge problems. Problem number one.   For robots that are not driving but moving objects   around, they keep on touching things, they  don’t have a really good game to play yet. But wait, is grasping that hard? Well, for  us, it is not hard at all. By the time we   are small children, this is more or less  child’s play for us. But, for robots? Yes,   it is incredibly hard. Look. If it grasps too  lightly, this happens, and if it grasps too hard,   the results are quite predictable. It has  to do everything with just the right amount   of force. And that needs tons and tons  of learning, and tons of training data. So, for these robots that have tactile sensors,  this fantastic new paper contains a video game   for virtual robots to play in, but it gets  better. In this game, they are meant to touch   things and based on how they  feel, follow their surfaces,   open a box, move an object and so  on. But unfortunately we are still   not ready to solve this grasping task  because we have one more huge problem. Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. So what is problem number two? It is the  sim to real gap. A simulation is great,   however, reality is not like a video game.   You can try to simulate everything down   to the tiniest detail, but when entering the  real world, plans go wrong all the time. So   does it even make sense to have a video  game that does not teach reality, but it   teaches useful knowledge in a made-up world? Of  course, it doesn’t. So what can we do with that? Oh my, look at that. This is a differentiable  system, which means that when we go from   simulation to reality, and we see that there  is a gap, we can account for the differences,   and automatically reprogram it to match  reality much more closely. In short,   we can close the gap between simulation and  reality. That is absolutely incredible and   is one of the key elements to make useful  robots to fold our laundry in the future. So, can it finally do the grasping  problem? We will see in just a moment. Now, there are many previous systems that can  do parts of what this is doing. Rigid bodies,   soft bodies, differentiability, optical  simulations, they can do parts of it,   but this new system seems to be the only one  that has it all. And, whoa, look at that. It   uses taichi as a backend. Yes, Fellow Scholars,  this is something that we have talked about   approximately 600 videos ago. Yes, I think this  might be the only corner of the internet where   you see the history of these research works unfold  right before your eyes. Amazing. I love my work. Now, grasping. You remember that too  soft was not great, grasping too hard,   also not great, but after learning  in this simulated environment,   let’s see…yup. That is just the right amount  of force! This concept really works! So cool! So I would like to send a huge thank  you for these brilliant scientists,   for giving this away for all of us for free.   And of course, a big thank you to this cat too   who surely is the mastermind behind all  this. It is probably thinking, finally,   I am inventing robots that play pat-a-cake  cake with us! What a time to be alive! I mean,   when cats do it among themselves, it always  ends up in a brutal fight. Look at that. And I hear so few people talking  about this, it’s crazy, and this   is why Two Minute Papers exists. Not to  create videos that are two minutes long, ### Segment 2 (05:00 - 05:00) [5:00] I have clearly failed mightily at that. It  exists to talk about amazing research papers,   and amazing human achievements that very  few people, if anyone is talking about. Subscribe and hit the bell icon  if you wish to hear more. And   what would you Fellow Scholars use this  for? Let me know in the comments below.