Illuminating the Space of Beatable Lode Runner Levels Produced By Various Generative Adversarial Networks

TL;DR

This paper investigates how different GANs trained on varying data subsets generate diverse, beatable Lode Runner levels, revealing that less training data can sometimes produce more diverse and playable levels.

Contribution

It demonstrates that training GANs on smaller datasets can yield more diverse and playable game levels, challenging the assumption that more data always improves generative quality.

Findings

A GAN trained on 20 levels produced the most diverse beatable levels.

A GAN trained on 150 levels produced the least diverse beatable levels.

Training data size impacts the diversity and quality of generated game levels.

Abstract

Generative Adversarial Networks (GANs) are capable of generating convincing imitations of elements from a training set, but the distribution of elements in the training set affects to difficulty of properly training the GAN and the quality of the outputs it produces. This paper looks at six different GANs trained on different subsets of data from the game Lode Runner. The quality diversity algorithm MAP-Elites was used to explore the set of quality levels that could be produced by each GAN, where quality was defined as being beatable and having the longest solution path possible. Interestingly, a GAN trained on only 20 levels generated the largest set of diverse beatable levels while a GAN trained on 150 levels generated the smallest set of diverse beatable levels, thus challenging the notion that more is always better when training GANs.