Mario AI
This was work done for Monash University's Artifical Life, Artificial Intelligence and Virtual Environments unit in 2024.
Our task was to program an evolutionary algorithm that can play and speed run levels from Super Mario Bros. The program was written in Java.
​
Project Duration: 4 Weeks
The base code is a modified version of the Super Marios Bros simulation developed by Ahmed Kalifa. We were tasked to program a game-playing controller for the Mario agent in any way we wanted, as long as it used evolutionary principles.
​
The AI was tested against 15 levels of Super Mario Bros.
13/15 were able to be completed in under 60 seconds.
Demonstration of the evolutionary controller completing level 1-1.
The controller does not do any prior training. It finds solutions as the game runs.