End-to-end game design using EPCG

Abstract

Procedural content generation (PCG) algorithms have been utilized for automating the creation of game content such as levels, assets, and narratives. One specific type, Exhaustive PCG (EPCG), systematically generates all possible variations of content before selecting the best, embodying a form of exploratory creativity. While EPCG has been applied to modifying or creating entire levels in games, there has been no research into employing EPCG for generating games end-to-end. This work explores the application of EPCG to game design, specifically focusing on the tangram puzzle genre. The game design process involves collaboration between human designers and computational algorithms. Our aim is to create a puzzle curriculum with multiple chapters, each introducing a new combination of constraints. Through this work, we demonstrate the potential of EPCG in driving comprehensive game design. Additionally, we introduce the concept of differential entropy, which assists in teaching puzzle inference rules, and utilize it to develop a puzzle curriculum that encourages players to understand the puzzle more deeply.