EA’s New AI: Next-Level Games Are Coming!

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today you are going to witness how an AI learned   to perform these amazing juggling  skills. And not only with the ball. And I am very excited for this. Why is  that? Well, because I see that this paper   was co-authored by our friend, Sebastian  Starke, which means that an amazing paper   is almost guaranteed. He was behind this  work too, and many other amazing papers. Now, creating a little AI that can learn these  kinds of skills seamlessly is immensely difficult.    Why? Well, the two hallmarks of a great animation  system is that the transitions between movements   are great and that no foot sliding happens.   And, as you see, this previous technique is   unfortunately not it. Yet, you see this kind of  animation in a lot of computer games, even today. And, unfortunately, it gets even worse. Look!   The ball does not obey the laws of physics. So,   it seems that getting quality animations  for gaming is absolutely hopeless. So,   is that it? Well, don’t despair quite yet! Now, previous techniques required a lot of  manual labor in creating these animations,   and weaving these animations together. However,  today, we see more and more AI-based systems that   can have a look at a big soup of movement data  that shows how humans move. Why is that? Well,   this is a learning based system, so it has a bit  of an understanding of what is happening. So,   what does that give us? Well, hopefully exactly  what we are looking for! It promises a smooth   transition from any move into any other move.   Well, after seeing these previous results, I don’t   have very high hopes, so I will believe it when I  see it. This is how a previous technique was able   to animate this character. Not great. And now,  hold on to your papers, and let’s see the new one! Oh wow. Just look at that! These animations are  indeed fluid, and no matter how we combine these   movements together, the AI is able to weave  them together into one smooth and believable   motion. It makes up these beautiful transitions  on the spot without ever having seen them. Now,   this captured my imagination. So what do  experienced Fellow Scholars like us do here? Of course, we make it harder. Much  harder. How? Well, for instance,   let’s add wind to the mixture, constantly  blowing the ball away from the player. Wow,   that is unbelievable. Not a problem  at all for this new AI. It can adjust   to new situations, which is a hallmark  of an intelligent algorithm. Love it. But wait! So what is the extent of new  situations that it can generalize to? Let’s see. Oh, I love that. Turning motions  are also possible. It had witnessed people   turning around in the training data, and it  had seen people juggle. That is not a surprise,   but what is surprising is that it was  able to put the two together seamlessly,   so it can juggle while turning. That is  absolutely amazing. Now these results   are already very impressive, but let’s be a  little cheeky. So, can we ask for even more? Well, you know what? No one said that we are  only allowed to juggle a ball. Who said that   it has to be a ball? It could be a box.   And my goodness! Look! It still works. And it also works with an American football.   Or even, yes, you are seeing correctly. It even   works with toilet paper. So, who is better,  the AI or Lionel Messi? Well, let’s see.    Look at that! Haha! Joke’s on you, little  AI! Team Humanity vs Team Robot - 1:0. But,   unfortunately it gets even worse. I am not  sure if I should be showing this to you, so,   are you prepared? Are you sure? I think  you are not prepared at all for this,   so if you continue the video, it’s on you.   You’ve been warned. Okay…go. Ouch. Whoo! This reminds me of an earlier paper where they  wrote up a musculoskeletal simulation system,

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today you are going to witness how an AI learned   to perform these amazing juggling  skills. And not only with the ball. And I am very excited for this. Why is  that? Well, because I see that this paper   was co-authored by our friend, Sebastian  Starke, which means that an amazing paper   is almost guaranteed. He was behind this  work too, and many other amazing papers. Now, creating a little AI that can learn these  kinds of skills seamlessly is immensely difficult.    Why? Well, the two hallmarks of a great animation  system is that the transitions between movements   are great and that no foot sliding happens.   And, as you see, this previous technique is   unfortunately not it. Yet, you see this kind of  animation in a lot of computer games, even today. And, unfortunately, it gets even worse. Look!   The ball does not obey the laws of physics. So,   it seems that getting quality animations  for gaming is absolutely hopeless. So,   is that it? Well, don’t despair quite yet! Now, previous techniques required a lot of  manual labor in creating these animations,   and weaving these animations together. However,  today, we see more and more AI-based systems that   can have a look at a big soup of movement data  that shows how humans move. Why is that? Well,   this is a learning based system, so it has a bit  of an understanding of what is happening. So,   what does that give us? Well, hopefully exactly  what we are looking for! It promises a smooth   transition from any move into any other move.   Well, after seeing these previous results, I don’t   have very high hopes, so I will believe it when I  see it. This is how a previous technique was able   to animate this character. Not great. And now,  hold on to your papers, and let’s see the new one! Oh wow. Just look at that! These animations are  indeed fluid, and no matter how we combine these   movements together, the AI is able to weave  them together into one smooth and believable   motion. It makes up these beautiful transitions  on the spot without ever having seen them. Now,   this captured my imagination. So what do  experienced Fellow Scholars like us do here? Of course, we make it harder. Much  harder. How? Well, for instance,   let’s add wind to the mixture, constantly  blowing the ball away from the player. Wow,   that is unbelievable. Not a problem  at all for this new AI. It can adjust   to new situations, which is a hallmark  of an intelligent algorithm. Love it. But wait! So what is the extent of new  situations that it can generalize to? Let’s see. Oh, I love that. Turning motions  are also possible. It had witnessed people   turning around in the training data, and it  had seen people juggle. That is not a surprise,   but what is surprising is that it was  able to put the two together seamlessly,   so it can juggle while turning. That is  absolutely amazing. Now these results   are already very impressive, but let’s be a  little cheeky. So, can we ask for even more? Well, you know what? No one said that we are  only allowed to juggle a ball. Who said that   it has to be a ball? It could be a box.   And my goodness! Look! It still works. And it also works with an American football.   Or even, yes, you are seeing correctly. It even   works with toilet paper. So, who is better,  the AI or Lionel Messi? Well, let’s see.    Look at that! Haha! Joke’s on you, little  AI! Team Humanity vs Team Robot - 1:0. But,   unfortunately it gets even worse. I am not  sure if I should be showing this to you, so,   are you prepared? Are you sure? I think  you are not prepared at all for this,   so if you continue the video, it’s on you.   You’ve been warned. Okay…go. Ouch. Whoo! This reminds me of an earlier paper where they  wrote up a musculoskeletal simulation system,

and hamstrung it in a variety of different  ways. That is, these poor little folks got   injuries and they simulated how they  would move afterwards. But worry not,   after this, simulating surgery  and healing them was possible. So, after perhaps a similar  surgery, when this new AI recovered,   it was also able to do three amazing  things that I really didn’t expect. One, it can juggle the ball with its chest. Two,  oh yes, this. I love this because this showcases   the advantages of a physics-based system for the  ball movement. It can stop the ball with its foot. Or, three, it can also kick the ball up  to varying heights. And to think that it   can do all this from just a soup  of motion capture data. That is,   recorded movement data from real humans.   That is incredible. And remember, it had   never seen these objects before. Wow. I cannot  wait to try the video games of the future where   we will be able to control these characters  using this paper. What a time to be alive! So, what do you think? What would you use  this for? Let me know in the comments below! Thanks for watching and for your generous  support, and I'll see you next time!

and hamstrung it in a variety of different  ways. That is, these poor little folks got   injuries and they simulated how they  would move afterwards. But worry not,   after this, simulating surgery  and healing them was possible. So, after perhaps a similar  surgery, when this new AI recovered,   it was also able to do three amazing  things that I really didn’t expect. One, it can juggle the ball with its chest. Two,  oh yes, this. I love this because this showcases   the advantages of a physics-based system for the  ball movement. It can stop the ball with its foot. Or, three, it can also kick the ball up  to varying heights. And to think that it   can do all this from just a soup  of motion capture data. That is,   recorded movement data from real humans.   That is incredible. And remember, it had   never seen these objects before. Wow. I cannot  wait to try the video games of the future where   we will be able to control these characters  using this paper. What a time to be alive! So, what do you think? What would you use  this for? Let me know in the comments below! Thanks for watching and for your generous  support, and I'll see you next time!