Gamasutra: The Art & Business of Making Gamesspacer
View All     RSS
February 16, 2019
arrowPress Releases

If you enjoy reading this site, you might also want to check out these UBM Tech sites:

Researchers teach AI the art of 'interesting' level design

May 14, 2018 | By Chris Kerr

Researchers over at the Polytechnic University of Milan have successfully taught an artificial intelligence how to design and generate levels for the original Doom

Notably, the research team claim they've trained their neural networks (a type of algorithmic AI that learns by analyzing large datasets) how to create levels that are actually interesting to play -- so these aren't just aimless computer generated labyrinths.

As reported by Motherboard, the neural networks haven't been built to replace game designers, but rather to help their human counterparts cook up more compelling levels with minimal effort.

"Our research is meant to be a step toward intelligent tools to help level designers in generating better levels with less effort, not to substitute them," explained Edoardo Giacomello, one of the authors behind the study, speaking to Motherboard.

"This allows the designer to focus on more creative and high-level aspects using the generated level as a starting point."

The research group taught their networks the art of level design by feeding them vital information from over 1000 Doom stages. The networks then used that data to create completely new levels, such as the one shown above.

It's a fascinating project, so be sure to wheel back around to Motherboard for the full story.

Related Jobs

Monomi Park
Monomi Park — San Mateo, California, United States

Senior Game Designer
Curriculum Associates
Curriculum Associates — San Francisco, California, United States

Senior Software Engineer - Learning Games (Unity)
Boston Dynamics
Boston Dynamics — Waltham, Massachusetts, United States

Software Engineer
Deep Silver Volition
Deep Silver Volition — Champaign, Illinois, United States

Effects (VFX) Artist

Loading Comments

loader image