SNAKE-RL

Welcome on the repo presenting my own implementation of Reinforcement Learning using for snake control.

The agent part structure is inspired by Patrick Loebers work.

Game mode

Without wall transparency

With wall transparency

Run DEMO

docker-compose up

Runs a demo with using model

Runs using pretrained model from:

/snake-rl/src/models/q_learning_net_without_wall_transparency.pth

Structure

TRANSMITER

It's a simple flask project enabling the communication between environment state and snake-visualizer app.

It contains two endpoints:

/publish-environment @Post

Gets json data and emits it using socketio for listeners observing new_environment event

/health @Get

Returns ok if service is available.

SNAKE-VISUALIZER

Simple React + Three.js application for visualizing snake in 2D grid of boxes.

It uses following json properties:

• x_rows (int) - gives number of columns in x direction
• y_rows (int) - gives number of columns in y direction
• is_sphere (bool) - if true displays snake grid as sphere else as 2d board
• is_alive (bool) - status of snake
• reward (int) - reward in the current game
• score (int) - score in the current game
• number_of_steps (int) - total number of steps in the game
• number_of_steps_without_food (int) - number of steps since the last chunk of food was found
• number_of_games (int) - total number of games

SNAKE-RL

Please read detailed description of the snake concept in designated folder:

Snake agent docs