Super Mario Maker doesn’t come out until September, but it’s already been rendered obsolete (as have humans). Researchers at the Georgia Institute of Technology have engineered a computer program that watches YouTube Let’s Plays of simple two-dimensional video games, then creates its own level sections based on the knowledge it’s gained.
The program studies sprite movement to analyze where players spend most of their time in a game level, and uses this knowledge to calculate level layouts. It studies the placement of blocks, bricks and pipes to determine where they should be placed, and where they shouldn’t. It uses spatial analysis to know how far is too far to make an in-game jump.
After viewing 17 game videos, the program was able to generate 151 original level sections for Super Mario Bros. This number soon increased to 334. “An initial evaluation of our approach indicates an ability to produce level sections that are both playable and close to the original without hand coding any design criteria,” said lead researcher Matthew Guzdial.
The computer isn’t up to generating entire levels yet, but the researchers at the Institute are working on it. When the program is perfected, says associate professor Mark Riedl, the knowledge gained from its application will help game designers better generate procedurally generated worlds, which up to this point have had the downside of turning out repetitive and boring.