Tile Embedding: A General Representation for Procedural Level Generation via Machine Learning

In recent years, Procedural Level Generation via Machine Learning (PLGML)\ntechniques have been applied to generate game levels with machine learning.\nThese approaches rely on human-annotated representations of game levels.\nCreating annotated datasets for games requires domain knowledge and is\ntime-consuming. Hence, though a large number of video games exist, annotated\ndatasets are curated only for a small handful. Thus current PLGML techniques\nhave been explored in limited domains, with Super Mario Bros. as the most\ncommon example. To address this problem, we present tile embeddings, a unified,\naffordance-rich representation for tile-based 2D games. To learn this\nembedding, we employ autoencoders trained on the visual and semantic\ninformation of tiles from a set of existing, human-annotated games. We evaluate\nthis representation on its ability to predict affordances for unseen tiles, and\nto serve as a PLGML representation for annotated and unannotated games.\n