An analysis of DOOM level generation using Generative Adversarial Networks

Edoardo Giacomello, Pier Luca Lanzi, Daniele Loiacono

Abstract: Highlights•Generative Adversarial Networks (GANs) learn models of data distributions.•GANs can extract regularities from human-designed Doom levels.•GANs generate Doom levels that are structurally similar to the human-designed ones.•Some input features can be used to guide the generation process.•Software available at https://github.com/edoardogiacomello/DoomPCGML.