Snake Game - QLearning

This project is a reinforcement learning experiment where an AI agent learns to play the classic Snake game using Q-Learning.

Installation

Make sure you have Python 3.7+ installed.

Install the required dependencies:

pip install pygame plotext numpy

Usage

To run the program, go into the home directory where qlearning.py is and run:

python qlearning.py

or

python3 qlearning.py

You can go to qlearning.py and adjust the following quality of life parameters:

1. render_episode -- Visualize the snake game and learning progress in the terminal when training reaches episode: render_episode
2. frame_rate -- Frame rate of snake game visualization

State Representation

The Q-learning agent observes the environment using a compact feature set that enables fast learning:

• Danger indicators relative to snake (Booleans):
  • danger_ahead
  • danger_left
  • danger_right
• Snake’s current direction (4 values): up, down, left, right
• Relative food location (Booleans):
  • food_ahead
  • food_behind
  • food_left
  • food_right

This results in a total of 1536 hypothetical states (2³ danger × 4 directions × 2⁴ food positions). In practice, many of these states aren't reachable, e.g. the food can't be both to the left and right.

Actions

The agent chooses from 3 discrete actions:

• Turn left
• Turn right
• Move straight

Images: