Researchers at the Polytechnic University of Milan in Italy have found it is possible to have a deep learning algorithm create compelling new game levels in Doom automatically, based on the data in existing levels.
The team started with 1,000 Doom levels from an existing repository; they processed each level to produce a set of images representing its most vital features, and created a vector that captured important level features in numerical form. They then applied a generative adversarial network to the dataset to learn how to generate new levels.
Researcher Edoardo Giacomello says after about 36,000 iterations, the deep learning networks could generate levels that capture the intrinsic structure of Doom levels, and could be used for level generation in other first-person-shooter games.
Such automation of the creation of new levels could have significant implications for game designers.
From Technology Review
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA