(embed) https://www.youtube.com/watch?v=udPY5rQVoW0 (/ embed)
Artificial intelligence stands out prominently in a wide range of games, especially in the way NPCs react to situations. But artificial intelligence is so much more than having a duck of characters behind a corner during a shootout, and in a completely different direction, a couple of artificial intelligence researchers have taken advantage of Nvidia's GameGAN neural network to create GAN Theft Auto, a fully AI-generated version. of Grand Theft Auto V. The result is quite remarkable.
The demonstration that can be played is to go down a short stretch of road to GTA V. From a modern graphic point of view, it's not what you might expect: the scene is very pixelated and even after trying again the sample, there is still a fog, because if you play GTA V in a dream state.
Still, there are some impressive details visible in the demo, such as generating a shadow under the car and precise reflections of sunlight in the rear window that change as the vehicle moves.
So how did this become possible? GameGAN is a controversial special generation network (or GAN) created by Nvidia, hence the name GAN Theft Auto. As its name suggests, GANs are made up of two neural networks that, in a sense, are adversaries. One of them is that it is a generator. A generator enters certain data, such as photos of humans, and then takes what it learns to create fake content that can be transmitted as reality.
The second part is a discriminator. This part has the task of keeping the generator honest, so to speak, by discerning what is real from what is false.
“GANs can create images that look like photographs of human faces, even though the faces don’t belong to any real person,” Nvidia explains. "GANs achieve this level of realism by pairing a generator, which learns to produce the target output, with a discriminator, which learns to distinguish the actual data from the generator output. The generator tries to trick the discriminator and the discriminator tries not to. 39; enganyin ".
It was at this time last year that Nvidia introduced GameGAN, the first neural network model that mimics a computer engine that uses a GAN and showed it off. cloning a fully functional version of Pac-Man. It has now been used to recreate GTA V.
At GAN Theft Auto, the neural network is the real environment and you can play with it, says youtuber Harrison Kinsley (via Engadget). Kinsley and collaborator Daniel Kukiela borrowed an Nvidia DGX Station A100 for this project, which was equipped with four A100 Ampere GPUs and a 64-core AMD Epyc processor.
They trained the GAN with a dozen simultaneous information from the highway scene and, from that data, learned how the car moves and how it responds to controls. . At first, he found a way to handle the limits, but eventually found out what to do if the car, for example, collided with a roadside barrier.
It is important to note that none of this is encoded by humans; the resulting demonstration is entirely generated by GameGAN, from how the world moves when the car moves to the controls that manipulate the vehicle.
Not everything is perfect. Beyond the low-resolution graphics, the GAN struggled with what to do with collisions against other vehicles. Kinsley describes a situation where the GAN simply split a police car in two when the main vehicle collided head-on with him. But it got better over time.
This is a possible vision of the future of games. It’s not hard to imagine an entire game invented by a GAN, or maybe parts of the game. That’s a bit far, though. Meanwhile, GAN Theft Auto is available on GitHub.