szmyty / OpenAI-Retro-SuperMarioWorld-SNES

This project contains my implementation of the NEAT-Python algorithm to use a recurrent neural network that enables an AI-controlled Mario to train itself to complete levels in "Super Mario World" on SNES. I used the gym-retro-integration program to create my own variables from the game's RAM values and then used those variables to reward/penalize the AI. If the AI complete's a level, the neural network is saved as the winner.

Geek Repo

Github PK Tool