Researchers found a way to train a deep-learning algorithm to generate levels for games that would be compelling.
Researchers found a way to train a deep-learning algorithm to generate levels for games that would be compelling. Edoardo Giacomello and his colleagues at the Politecnico di Milano in Italy showed how you can use science to build levels for DOOM.
The team states that this approach lets create levels in an automated way and that the technique could really change the way game content is created.
The whole thing begins with 1,000 Doom levels from a repository of publicly available games (all the official levels from Doom and Doom 2, plus 9,000+ levels created by the gaming community).
The levels have been processed to generate a set of images that represent the most important features (the walkable area, walls, floor height, objects, and more). The team has also set up a vector that captured important features of the level in numerical form (the size, area, and perimeter of rooms, the number of rooms).
A deep-learning technique called a generative adversarial network has been used to study the data and learn how to generate new levels.
After 36,000+ iterations, these networks could produce levels of good quality. “Our results show that generative adversarial networks can capture intrinsic structure of DOOM levels and appears to be a promising approach to level generation in first-person shooter games,” stated Giacomello and his team.
The levels are not perfect, though, coming with noisy data that is inevitably generated with this kind of approach.
Still, this is something else. “Levels are of paramount importance, especially in first-person shooter and platform games, as they greatly affect the player experience. Human designers can focus on high-level features by including specific types of maps or features in the training set.”
You can learn more about the work here.