Hybrid PCGRL-WFC Method for Game Content Generation

A new research paper on arXiv proposes combining Wave Function Collapse (WFC) with reinforcement learning (PCGRL) to generate game levels that satisfy both local visual constraints and global playability. The method constrains the action space of a PCGRL generator using constraints learned by WFC, enabling adherence to local patterns while achieving global properties via reward functions. Experiments vary input number/type and test random collapse of starting states and exclusion of rare patterns. The approach addresses the trade-off between visual quality and functional guarantees in procedural content generation.