This content was created by LARRUZ. This content was uploaded by LARRUZ. This is a YouTube video titled "DeepMind-Like Gaming AI: Incredible Driving Skills!" created on August 27, 2023, featuring Dr. Zsolnai discussing advancements in AI agents learning to play video games, including driving and laser tag, and a new computer graphics simulation technique addressing volumetric dissipation.
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to look at a new AI agent that learned to play video games really well and learned some really interesting things. I'll leave it at that for now and show it to you in a moment. Now, first things first. The goal was to create a large number of procedurally generated levels. These are created by your computer, if you will. Here you see their AI driving on the Hungarian Formula 1 track. The goal is that it would start to play increasingly difficult levels and see if it has proper zero shot performance. That is, can it play a level that it hasn't seen yet. So let's switch to a previously unseen track. Great job, little AI. But why is this so interesting? It is interesting because of two things. One, it learned some really interesting skills in laser tag, a little shooter game. Look, it learned to hide behind walls to evade these shots. So what happens if a fight breaks out? Well, not a problem. It can also dodge these bullets. Or it gets better. It can even run away from them. Loving it. And here comes the interesting part. Second, this little AI learned to be a criminal. And I hear you asking, doctor, how? Well, let me show you these little shenanigans. It can not only drive on new levels, but it has also learned to force an opponent off of the road. Ouch. It can also overtake by cutting corners. Very creative. But the creative elimination of the opponent does not stop there. Oh my. Are you seeing what I am seeing? Dear fellow scholars, that is a hit and run. I can almost hear it say bye bye, see you next lap. And if you happen to meet it again, this little criminal can also block your cornering. And all this was trained on a single graphics card. Not a cheap one, but one graphics card nonetheless. Really cool experiment. So, ok, but what is all this good for? Well, this technique could be extended to games that require more than two players. And we could see what kind of collaboration these AI agents can learn together. Or if you ask me perhaps what kind of criminal rings they create. Kidding, of course. And if you allow me, I would like to show you an incredible computer graphics simulation. Paper. This one addresses something that we call the volumetric dissipation problem. What is that? Well, look, we start the simulation with a given number of particles and over time, due to the inaccuracies of our calculations, they slowly disappear. Like a magic trick. Don't believe it. Well, look, we started with this many particles and now we only have this many. And this new technique is able to create these magical simulations where the amount of volume over time remains the same. So does it apply to messier scenes than this? You bet your papers it does. Check this out. The scene can be an absolute mess and it still works. And it not only works, but it can be easily plugged into already existing systems and it will work up to two and a half times faster than previous methods. And all this with volume preservation. Loving it. Now one more interesting thing. Look, only 500 people have seen this paper yet, so I am really worried that if we don't talk about these incredible computer graphics research works here on two minute papers, perhaps no one will talk about them. If you like this, please share it with your friends. If you're looking for inexpensive cloud GPUs for AI, Lambda now offers the best prices in the world for GPU cloud compute. No commitments or negotiation required. Just sign up and launch an instance and hold on to your papers. Because with the Lambda GPU cloud you can now get on demand H100 instances for just 1.99per hour. Yes, 1.99and they are one of the first cloud providers to offer publicly available on demand H100XS. Did I mention they also offer persistent storage? So join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances, workstations or servers. Make sure to go to lambdalabs.compapers to sign up for one of their amazing GPU instances today. Thanks for watching and for your generous support and I'll see you next time.
