Generating and Blending Game Levels via Quality-Diversity in the Latent Space of a Variational Autoencoder

Several works have demonstrated the use of variational autoencoders (VAEs)\nfor generating levels in the style of existing games and blending levels across\ndifferent games. Further, quality-diversity (QD) algorithms have also become\npopular for generating varied game content by using evolution to explore a\nsearch space while focusing on both variety and quality. To reap the benefits\nof both these approaches, we present a level generation and game blending\napproach that combines the use of VAEs and QD algorithms. Specifically, we\ntrain VAEs on game levels and run the MAP-Elites QD algorithm using the learned\nlatent space of the VAE as the search space. The latent space captures the\nproperties of the games whose levels we want to generate and blend, while\nMAP-Elites searches this latent space to find a diverse set of levels\noptimizing a given objective such as playability. We test our method using\nmodels for 5 different platformer games as well as a blended domain spanning 3\nof these games. We refer to using MAP-Elites for blending as Blend-Elites. Our\nresults show that MAP-Elites in conjunction with VAEs enables the generation of\na diverse set of playable levels not just for each individual game but also for\nthe blended domain while illuminating game-specific regions of the blended\nlatent space.\n