Return to Dr. Jacob Schrum's Homepage

Zelda Dungeons: Generative Adversarial Network Rooms in Generative Graph Grammar Dungeons

This page presents videos associated with a human subject study focused on evaluating a new hybrid Procedural Content Generation (PCG) technique referred to as Graph+GAN: The use of a generative graph grammar to define a mission for a dungeon crawling game, whose rooms are then specified by sampling from the latent space of a generative adversarial network (GAN) trained on level data from the original Legend of Zelda video game (adapted from the Video Game Level Corpus). Participants in the study played a Graph+GAN dungeon using a simple ASCII-based Rogue-like game engine. For comparison, they also played two other dungeons. One was a dungeon generated using the same graph grammar, but with prefabricated rooms from the original Legend of Zelda. The other was Dungeon 4 from the original Legend of Zelda, adapted to be playable in the Rogue-like engine. This research was conducted by Jake Gutierrez as part of Southwestern University's Summer research program SCOPE. All code is part of the MM-NEAT repository, whose source code is available via GitHub here (get release 3.2 for binaries).

Supplementary Figure

This figure is included in the end of the extended version of the publication available on this page, but absent from the officially published IEEE version:
Supplemental Material

Video Presentations

Video presentation prepared by Jake Gutierrez for the Congress on Evolutionary Computation, which was part of the World Congress on Evolutionary Computation. The conference shifted to a virtual format due to the COVID-19 pandemic.

Video presentation prepared by Jake Gutierrez for the Southwestern University Research and Creative Works Symposium. The symposium shifted to a virtual format due to the COVID-19 pandemic.

Human Subject Sessions in Graph+GAN Dungeons

Videos of sessions from all 30 participants in the human subject study are included in the playlist below. These videos show sessions in Graph+GAN dungeons.


Human Subject Sessions in Graph Dungeons (Rooms From Original Game)

Videos of sessions from all 30 participants in the human subject study are included in the playlist below. These videos show sessions in pure Graph dungeons. The GAN was not used. Instead, rooms were randomly sampled from those present in the original game The Legend of Zelda.


Human Subject Sessions in Original Dungeon 4

Videos of sessions from all 30 participants in the human subject study are included in the playlist below. These videos show sessions in Dungeon 4 of the original Legend of Zelda, adapted for the Rogue-like engine. Note that Subject 6 actually failed to beat the dungeon on the first attempt, and thus was allowed to try again.


Associated Publications


Peer-Reviewed Journal Articles


Peer-Reviewed Conference Publications


Extended Abstracts


Technical Reports


Dagstuhl Reports


Undergraduate Poster Presentations Supervised


Associated Movies and Images



Miscellaneous Content

  • Summer 2021: Quality Diversity and Creative Divergent Search: SCOPE Research Presentation made by my SCOPE Summer research students to present to other SCOPE students
  • Fall 2020: Procedural Content Generation for Games with Generative Adversarial Networks: Presentation for the Games AI Research Group at Queen Mary University of London (video)
  • Fall 2020: Interactively Evolving Video Game Levels with Generative Adversarial Networks: 403 Lecture for Math and CS Department
  • Fall 2020: Evolving Mega Man Levels with Generative Adversarial Networks: Virtual SCOPE Open House Website
  • Fall 2020: Evolving Lode Runner Levels with Generative Adversarial Networks: Virtual SCOPE Open House Website
  • Summer 2020: Generating Video Game Levels Using AI: SCOPE Research Presentation made by my SCOPE Summer research students to present to other SCOPE students
  • Summer 2019: Playing and Creating Games With Deep Neural Networks: "Mad Science Monday" presentation made by my SCOPE Summer research students to present to other SCOPE students
  • Fall 2018: Evolutionary Computation Applied to Digital Entertainment and the Arts, poster presented at the President's Appreciation Celebration for Southwestern University donors.
  • Fall 2018: Evolving Mario Levels in the Latent Space of a Deep Convolutional Generative Adversarial Network: presented to Southwestern University students as a 107 Lecture.
  • Summer 2018: The machines have taught themselves to make Mario levels, article in Fast Company about my recent GECCO 2018 paper on generating levels for Super Mario Bros.
  • Spring 2018: Bored with your video game? Artificial intelligence could create new levels on the fly, article in Science about my recent GECCO 2018 paper on generating levels for Super Mario Bros.
  • Spring 2018: Doom and Super Mario could be a lot tougher now AI is building levels, article in The Register about my recent GECCO 2018 paper on generating levels for Super Mario Bros.

  • Last Updated: 6/28/2020