Maze Generation Based on Difficulty using Genetic Algorithm with Gene Pool

Abstract

Game level design is one of the most important element of developing an enjoyable video game. Besides, game with difficult and dynamic level can make players more exciting. This paper presents a new method of generating a video game level using a genetic algorithm. The proposed method is called gene pool integrates learning. This method implemented in feature selection so that this method is general enough to be used for multiple different types of games. This paper uses some training data to scan good patterns and store all of them in a gene pool. Furthermore, the genetic algorithm is used to find the combination of patterns that can produce the best result. The gene pool also records the quality of each gene so it can learn the pattern which most commonly found in multiple levels. For testing, this research develops a custom game with complicated rules that are hard to represent by a simple 2D array compared to the previously attempted work. The result of this research shows that the method can generate many complicated levels at once. Overall, levels generated using this method on average requires almost 3 times more steps to solve than the dataset.