Skip to content

Latest commit

 

History

History
29 lines (19 loc) · 1.69 KB

README.md

File metadata and controls

29 lines (19 loc) · 1.69 KB

NEAT Network Implementation

This is a simple implementation of the NEAT network. The only purpose of this project is to learn how to implement a NEAT network, learn about physics and game development, and have fun of course :3.

I don't use any framework or library, everything is written from scratch with vanilla javascript and typescript.

The project is divided into two parts: the neuralNetwork and the game. The neuralNetwork is the brain of the bot, and the game is the environment where the bot is playing. Both parts have their own config.ts file.

The neuralNetwork is a simple implementation of a NEAT network, and the game is a simple implementation of a game in 2D with physics made with verlet integration, specifically Non-constant time differences Størmer–Verlet with Taylor series that you can find here.

I tried so hard to make the code as simple and readable as possible for anyone who wants to understand how it works or use it as a guide to implement their own network.

How to use

Install and run

  1. Clone the repository
  2. Run npm install
  3. Run npm run dev
  4. Open the browser and go to the url that vite provides

Usage

  • Start/Stop toggles the training mode
  • Simulate toggles the rendering mode
  • Both modes can't be active at the same time
  • The Renderer class can be instantiated with a Game object, without an agent and it will render the game with controls for the user (Arrow keys for the bot movement).

Note

For now the project is not complete, and there are a lot of things to do, such as adding a better UI, and maybe a better game. But I hope you enjoy it anyway!