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

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