rjkilpatrick/mario-rl

PyTorch lightning implementation of reinforcement learning for the NES version of Super Mario Bros

Super Mario Bros

Can a computer learn to play super mario bros? Yes.

Quick Start

git clone https://github.com/rjkilpatrick/mario-rl
cd mario-rl
conda install
conda activate rjkilpatrick-mario-rl
jupyter notebook .

The Project

In this case we use a process called Q-learning which uses deep learning to approximate a reward function. By optimizing a reward function (which is the premise of most games), you can 'play' a game.

Licensing

This project was inspired by: https://blog.paperspace.com/building-double-deep-q-network-super-mario-bros/. Code under MIT License.