The prototype showed off its design chops with Nintendo’s 1985 platformer Super Mario Bros., producing entirely new level layouts that are fully playable by humans.
“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 Matthew Guzdial, the project’s lead researcher and Ph.D. student in Computer Science at Georgia Tech.
When crafting its levels, Georgia Tech’s AI observes uploaded gameplay video and focuses on high-interaction areas, or spots within each level that demand player attention and mastery of skill. This allows the AI to create layouts that lean on established conventions within the platforming genre while crafting new challenges of its own.
The AI can passively learn level design rules from its research, preventing the creation of elements that break designer-issued standards or prevent player progression. Within the 2D world of Super Mario Bros., for instance, the AI has learned how to position warp pipes, and knows which gaps are too large for players to realistically cross.
“Our system creates a model or template, and it’s able to produce level sections that have never been seen before, do not appear random and can be traversed by the player,” said Mark Riedl, associate professor of Interactive Computing at Georgia Tech. “One could say that the system ‘studies’ the design of Mario levels until it is able to create new playable areas.”
Georgia Tech’s researchers hope to eventually port these AI-developed levels to the Super Mario Bros. engine and observe how players approach computer-designed level layouts, as opposed to challenges crafted by human developers.
Nintendo will launch its DIY 2D platformer toolkit Super Mario Maker for the Wii U in September, potentially giving humans an edge over their level-designing AI overlords.
