Evolutionary Procedural Content Generation for an Endless Platform Game

Abstract

Making innovative, cohesive and appealing games has become inherently more difficult given the ever increasing competition in the digital games' market. Manually creating game content is expensive and time-consuming. Therefore, alternative approaches for game content creation are relevant for increasing the efficiency of the game development process. This is where procedural techniques step in. Even though they have been used by commercial games since the 1980s, it was only in recent years that this kind of approach has been given the righteous attention in the academic context. In this work, we propose a procedural content generation approach for creating infinite environments for a 2D platform runner game. The approach consists of a Genetic Algorithm that innovatively takes into account environment aesthetics as well as game's physics and rules in its fitness function. Therefore, the created environments should be pleasant and possible to be overcome by the player. An instantiation of the approach was developed using the Godot Game Engine. Time viability for in-game real-time generation and convergence to high/stable fitness values were experimentally evaluated. Our tests indicated parameter ranges that performed best in terms of environment quality and processing time were mutation rates between 0.5 % and 1 % aligned with a population ranging from 50 to 100 individuals. This approach is expandable to other games that have a tileman-based environments.