TetriMind

A Tetris Game with Reinforcement Learning AI Agent

Game Demo

Training Demo

Setup

• Clone the repository and install dependencies from requirements.txt

  pip install -r requirements.txt
  

• Open the folder in Visual Studio Code and run:

  python3 project/main.py
  

Features

• Classic Tetris with responsive controls
• Player vs AI mode with difficulty levels
• Realistic AI moves
• Versus AI garbage lines
• T-spins
• Retrainable Tetris Agent (Genetic Algorithm)

Results

Genetic Algorithm

• Mutation rate: 0.05
• Mutation step: 0.2
• Reward/Fitness: pieces survived + lines cleared
• Feature function: holes, bumpiness, weighted height, cumulative height, relative height, vertical hole tunnels, max well depth, sum wells, weighted filled cells, landing height, hole depth, row hole

GA Version 1.0

• Generations: 17
• Population: 60
• Games: 1,020 gameovers
• Standard 7-Bag w/ changing seeds per generation

GA Version 2.0

• Generations: 24
• Population: 240 (4x bigger sample)
• Games: 5,760 gameovers
• 11-Bag: 7Bag but 3x more S & Z hard mode pieces
• Changing seeds

Tech Stack

Language: Python
Main Library: pygame
Optional Library: pyinstaller (exe bundling), matplotlib (training)

Useful Links:

• Reinforcement Learning on Tetris
• Evolving Tetris AI based on genetic algorithms