A Procedural Generation Platform to Create Randomized Gaming Maps using 2D Model and Machine Learning

Abstract

As a video game developer, the most difficult problem I ran into was creating a map for the game, as it was difficult to create non repetitive and original gameplay [1]. My project proposes a solution to this problem as I use Answer Set Programming to create a program to procedurally generate maps for a video game [2]. In order to test its reliability, I allowed it to generate around 10,000 maps, stored the data of each of the maps, and used the common trends I find in the data to find problems with the program and fix it in the future. In developing a video game, level creation consumes a major portion of the development total time and level procedural generation techniques can potentially mitigate this problem. This research focused on developing a VVVVVV style level generator using Answer set programming for the game Mem.experiment which was developed at the same time. VVVVVV is a 2D puzzle platformer that uses changes in direction of gravity instead of jumping for the player's vertical movement [3]. During the development of the level generator, 10,000 levels were created. I found out that the average total time it took between generations is 45 seconds, and the average time for ASP to generate a map is 12 seconds [4]. This means that the process of displaying the generation took between 2x - 3x longer than generating the ASP solution.