NVIDIA GameGAN, an AI that can create new environments for the mythical Pac-Man without the need for a graphics engine

Today marks the 40th anniversary of the appearance of the classic game Pac-Man. To commemorate this date, NVIDIA researchers have recreated this legendary game based on Artificial Intelligence. These researchers have used GameCAN, the first neural network that can mimic a graphics engine of a computer game. This has been achieved through Generative Adversary Neural Networks (GAN).

This system is based on two competing neural networks, a generator and a discriminator. Models that are based on GAN have the ability to learn to develop new content in a convincing way, posing as the original system. Translated, the AI-based version of the game is fully functional.

No products found.

NVIDIA creates an AI capable of generating the Pac-Man

This is the first research that emulates a graphics engine that uses GAN-based neural networks. We wanted to see if the AI ​​could learn the rules of an environment just by looking at the script of an agent moving through the game. So it was.

Seung-Wook Kim, NVIDIA Researcher and Lead Author of the Project

An artificial entity plays in the system generated by GAN, GameCAN responds to the actions of the agent. This is how new frames are played within the game in a real-time environment. GameCAN has the ability to create designs for the game never seen before, generating different levels or versions of it.

The ability of this system could be used by game developers to create new levels in an automated way. In addition, the researchers of this AI can develop new simulation systems for the training of autonomous machines.

We were astonished when we saw the results, in disbelief that AI could recreate the iconic PAC-MAN experience without a game engine. This research presents exciting possibilities to help game developers accelerate the creative process of developing new level designs, characters, and even games.

Koichiro Tsutsumi of BANDAI NAMCO Research Inc