Procedural Content Generation of Puzzle Games using Conditional Generative Adversarial Networks

Andreas Hald; Jens Stuckermann Hansen; Jeppe Theiss Kristensen; Paolo Burelli

doi:10.1145/3402942.3409601

Procedural Content Generation of Puzzle Games using Conditional Generative Adversarial Networks

Andreas Hald, Jens Stuckermann Hansen, Jeppe Theiss Kristensen, Paolo Burelli

Digital Design

Research output: Conference Article in Proceeding or Book/Report chapter › Article in proceedings › Research › peer-review

Abstract

In this article, we present an experimental approach to using parameterized Generative Adversarial Networks (GANs) to produce levels for the puzzle game Lily’s Garden1. We extract two condition-vectors from the real levels in an effort to control the details of the GAN’s outputs. While the GANs performs well in approximating the first condition (map-shape), they struggle to approximate the second condition (piece distribution). We hypothesize that this might be improved by trying out alternative architectures for both the Generator and Discriminator of the GANs.

Original language	English
Title of host publication	FDG '20: International Conference on the Foundations of Digital Games
Publisher	Association for Computing Machinery
Publication date	2020
Article number	99
ISBN (Electronic)	9781450388078
DOIs	https://doi.org/10.1145/3402942.3409601
Publication status	Published - 2020
Event	Foundations of Digital Games - Malta Duration: 15 Sep 2020 → 18 Sep 2020 http://fdg2020.org

Conference

Conference	Foundations of Digital Games
Location	Malta
Period	15/09/2020 → 18/09/2020
Internet address	http://fdg2020.org

Access to Document

10.1145/3402942.3409601

PCG_Parametric_GANSAccepted author manuscript, 9.48 MB

Cite this

@inproceedings{390fd2a89fc1463098c7e4691cf744ef,

title = "Procedural Content Generation of Puzzle Games using Conditional Generative Adversarial Networks",

abstract = "In this article, we present an experimental approach to using parameterized Generative Adversarial Networks (GANs) to produce levels for the puzzle game Lily{\textquoteright}s Garden1. We extract two condition-vectors from the real levels in an effort to control the details of the GAN{\textquoteright}s outputs. While the GANs performs well in approximating the first condition (map-shape), they struggle to approximate the second condition (piece distribution). We hypothesize that this might be improved by trying out alternative architectures for both the Generator and Discriminator of the GANs.",

author = "Andreas Hald and {Stuckermann Hansen}, Jens and Kristensen, {Jeppe Theiss} and Paolo Burelli",

year = "2020",

doi = "10.1145/3402942.3409601",

language = "English",

booktitle = "FDG '20: International Conference on the Foundations of Digital Games",

publisher = "Association for Computing Machinery",

address = "United States",

note = "Foundations of Digital Games, FDG ; Conference date: 15-09-2020 Through 18-09-2020",

url = "http://fdg2020.org",

}

Procedural Content Generation of Puzzle Games using Conditional Generative Adversarial Networks. / Hald, Andreas; Stuckermann Hansen, Jens; Kristensen, Jeppe Theiss et al.

FDG '20: International Conference on the Foundations of Digital Games. Association for Computing Machinery, 2020. 99.

Research output: Conference Article in Proceeding or Book/Report chapter › Article in proceedings › Research › peer-review

TY - GEN

T1 - Procedural Content Generation of Puzzle Games using Conditional Generative Adversarial Networks

AU - Hald, Andreas

AU - Stuckermann Hansen, Jens

AU - Kristensen, Jeppe Theiss

AU - Burelli, Paolo

PY - 2020

Y1 - 2020

N2 - In this article, we present an experimental approach to using parameterized Generative Adversarial Networks (GANs) to produce levels for the puzzle game Lily’s Garden1. We extract two condition-vectors from the real levels in an effort to control the details of the GAN’s outputs. While the GANs performs well in approximating the first condition (map-shape), they struggle to approximate the second condition (piece distribution). We hypothesize that this might be improved by trying out alternative architectures for both the Generator and Discriminator of the GANs.

AB - In this article, we present an experimental approach to using parameterized Generative Adversarial Networks (GANs) to produce levels for the puzzle game Lily’s Garden1. We extract two condition-vectors from the real levels in an effort to control the details of the GAN’s outputs. While the GANs performs well in approximating the first condition (map-shape), they struggle to approximate the second condition (piece distribution). We hypothesize that this might be improved by trying out alternative architectures for both the Generator and Discriminator of the GANs.

U2 - 10.1145/3402942.3409601

DO - 10.1145/3402942.3409601

M3 - Article in proceedings

BT - FDG '20: International Conference on the Foundations of Digital Games

PB - Association for Computing Machinery

T2 - Foundations of Digital Games

Y2 - 15 September 2020 through 18 September 2020

ER -

Procedural Content Generation of Puzzle Games using Conditional Generative Adversarial Networks

Abstract

Conference

Access to Document

Fingerprint

Cite this