Difficulty Curve-Based Procedural Generation of Scrolling Shooter Enemy Formations

Abstract

Listen

Translate article

In any game where the player must face large formations of enemies, the way the formations are formed significantly influences the difficulty level of the game. As proper difficulty levels are essential to player’s enjoyment, the enemy formations need to be arranged properly. In this paper we experimented with how to procedurally generate enemy formations in a vertical scrolling shooter game while keeping the difficulty levels proper. Generations of the formations were done with genetic algorithm where the fitness function calculated two variables: difficulty curve, which reflects a formation’s difficulty level, and enemy variety. The difficulty curve took into account difficulty points, which measured the danger levels of on-screen enemies throughout the game’s duration. The fitness of difficulty curve of an enemy formation was acquired through point-by-point comparison between the curve and a human-designed one, which represented the ideal difficulty curve as intended by the game’s developer, and calculation of the root-mean-square error. The genotype of an individual was represented by a 5x40 grid containing enemy genes and empty space genes. We ran 300 generations in total and each was done in 100 iterations and had 40 individual genotypes. The results show that the amount of enemy genes in initial genotypes influences the population’s fitness progression. We then discuss the implications of our findings and possible directions of future researches.