A Multitask Evolutionary Framework for Procedural Content Generation in Games

Abstract

Procedural Content Generation (PCG) has become a widely used technique for creating content in various types of games. In recent years, many approaches have been proposed to tackle different PCG problems. Despite the diversity of these methods, most existing generators are designed for specific games, even though many games share structural similarities. In this study, we explore the use of Evolutionary Multitask Optimization (EMTO) for PCG. We introduce a framework based on a Multifactorial Evolutionary Algorithm (MFEA) that systematically groups PCG problems by the compatibility of their search spaces, allowing for the simultaneous optimization of multiple problems. To assess the effectiveness of the proposed framework, we conducted experiments on a range of PCG benchmark problems and compared the results to a single-task evolutionary approach. The findings highlight both the potential benefits and the challenges of applying EMTO to PCG, particularly regarding task compatibility.