This week three researchers might’ve changed the course of history in the video game industry. They’ve used a pair of Generative Adversarial Networks (GANs) to study DOOM levels – DOOM, the original DOOM, the first-person shooter. With artificial intelligence, these researchers went about creating new DOOM levels without the assistance of humans.
While we were out here focusing on how AI can drive cars and buy movie tickets, researchers at Cornell had a better idea. Three researchers did a study on DOOM. They trained Generative Adversarial Networks (GANs) with DOOM map topographical features. They sent a set of images “identifying the occupied area, the height map, the walls, and the position of game objects of already-made maps.”
One GAN was trained with plain level images. Another GAN was trained with images and features extracted during their preliminary analysis. Once they’d taught their artificial intelligence programs all about what a whole bunch of DOOM maps looked like, they set these monsters to create their own levels. And what do you know, they did it just fine.
“Our results show that GANs can capture intrinsic structure of DOOM levels and appears to be a promising approach to level generation in first person shooter games,” said the research paper authored by Edoardo Giacomello, Pier Luca Lanzi, and Daniele Loiacono.
To learn more about this subject, seek out the paper by the name “DOOM Level Generation using Generative Adversarial Networks.” This paper was published by the Cornell University Library and can be found online at Arxiv with code arXiv:1804.09154 submitted by Pier Luca Lanzi.
ALSO NOTE: This team of 3 didn’t necessarily do this experiment to make a bunch of DOOM levels for the DOOM-loving community. They did it for the future of video games – or something like that. If you’re looking for new DOOM levels, we recommend you check out one of the repositories these Cornell researchers used – have a peek at the Github-hosted “Video Game Level Corpus.”