Connect with us


      Nvidia’s AI has created a version of Pac-Man just by watching games

Kim Diaz



To celebrate the 40th anniversary of the launch of Pac-ManToday, Nvidia has presented a very special version of this classic. It’s about the a fully functional version of the game that has been recreated by artificial intelligence whose only training has been to watch 50,000 games.

This powerful AI is called GameGAN and has been created by Nvidia Research. It is the first neural network model capable of imitating a video game without an underlying game engine, only from the observation of the games. And it has been achieved in just a few days.

But how is this possible? This tool is based on Adverse Generative Neural Networks or GAN, for its acronym in English (Generative Adversarial Networks).

GANs are made up of two competing neural networks, the generator and the discriminator. The generator is responsible for creating new content from the observation of original models, while the discriminator must be able to identify the real content of that created by the adversary neural network. At first, it is a simple task, but the discriminator’s work becomes increasingly complicated as the generator improves her skills.

Thanks to this system, GAN-based models learn to create new content that is compelling enough to go through the original. Nvidia has previously used this model with AI GauGAN, which converts schematic drawings to realistic images.

“This is the first research that emulates a game engine that uses GAN-based neural networks.” explains Seung-Wook Kim, Nvidia researcher and lead author of the project. “We wanted to see if AI could learn the rules of an environment just by looking at the screen of an agent moving through the game. And he did.”.

GameGAN can create versions of any game just by watching games

Missing stocking.

In order to generate the game environment, the AI ​​tracks the virtual world, remembering what has already been generated to maintain visual consistency. By observing what is happening on the screen and the keys the player presses, you can convincingly reproduce any environment.

In order for GameGAN to recreate a functional version of Pac-Man, Nvidia’s team has trained neural networks by showing them 50,000 games and the beats of an AI agent playing the game.

With this training, GameGAN is able to generate new frames in real time to respond to the agent’s actions, creating both static elements, such as mazes or points, as well as dynamic ones, for example ghosts or Pac-Man himself. Observation allows you to learn the basic rules of the game to follow in your recreations.

Artificial intelligence is already on everyone’s lips, but few people know what it is. How does it work? How far can it go? What are its limitations? We answer these questions.

You can even generate designs that you have not previously seen, a very useful feature both for developers, who can generate new levels of a game automatically, and for AI researchers, since they can more easily develop complex simulators for training autonomous robots.

“This research presents exciting possibilities to help game developers accelerate the creative process for new level designs, characters, and even games.”Koichiro Tsutsumi of Bandai Namco Research Inc. notes.

The version of Pac-Man created by GameGAN will be available later this year in the Nvidia AI Playground, where anyone can try it.

Kim recently joined the team, and she writes for the Headline column of the website. She has done major in English, and a having a diploma in Journalism. She has worked for more than 1.5 years in a media house. Now, she joined our team as a contributor for covering the latest US headlines. She is beautiful both by her looks and nature. She is very good with everyone in the team.