Interactive Evolution and Exploration Within Latent Level-Design Space of Generative Adversarial Networks

Generative Adversarial Networks (GANs) are an emerging form of indirect\nencoding. The GAN is trained to induce a latent space on training data, and a\nreal-valued evolutionary algorithm can search that latent space. Such Latent\nVariable Evolution (LVE) has recently been applied to game levels. However, it\nis hard for objective scores to capture level features that are appealing to\nplayers. Therefore, this paper introduces a tool for interactive LVE of\ntile-based levels for games. The tool also allows for direct exploration of the\nlatent dimensions, and allows users to play discovered levels. The tool works\nfor a variety of GAN models trained for both Super Mario Bros. and The Legend\nof Zelda, and is easily generalizable to other games. A user study shows that\nboth the evolution and latent space exploration features are appreciated, with\na slight preference for direct exploration, but combining these features allows\nusers to discover even better levels. User feedback also indicates how this\nsystem could eventually grow into a commercial design tool, with the addition\nof a few enhancements.\n