pong-rl

Reinforcement Learning with Pong in the browser via TensorFlow.js by @cpury 🚀

Demo: https://cpury.github.io/pong-rl/

With the default options, watch it learn the game in ~10 minutes!

Blog post coming soon 📝

Color scheme: https://coolors.co/3a3c3f-e5e5e6-f06543-2ec0f9-e7e247

Implements a Pong-like game as well as a controller interface to add arbitrary controllers to the game.

• KeyController to controll via keyboard
• DumbController a simple controller that just follows the ball with some hiccups
• DQLController a first Deep-Q-Learning controller that learns on hard-coded features (position of the ball, force of the ball and positions of the paddles).
• VisualDQLController Deep-Q-Learning directly on the pixels. SLOW!

• Three difficulty levels
• Live mode: When disabled, learning happens without drawing or waiting for user

• TensorFlow.js for neural networks
• Bulma as CSS framework
• Parcel.js for bundling

First, install dependencies:

npm install

To run in hot module reloading mode:

npm start

To create a production build:

npm run build-prod

Contributions welcome! If you'd like to add a new controller, add a new class that inherits from BaseController and create a pull request.