Tetris-reinforcement learning

This repository contains two agents. The first agent-algorithm, has a pre-programmed behaviour. The second agent-rl, uses DQN Reinforcement Learning to learn how to play.

• To execute the first agent, run the file tetris.py. Space Bar to drop the tetromino, R to restart, Q to Quit, Y to change to manual play(A,S Move)(K,L Rotate).
• To execute the second agent, run the file test.py, this will load an agent from the folder models. I have pre-selected the best-performing agent, which cleared more than 100k lines.

Project Structure

Pre-programmed Agent: Uses predefined rules to play Tetris.

Reinforcement Learning Agent: Learned to play Tetris using DQN, capable of clearing over 100,000 lines.

Interactive Controls for the algorithm version:

• Space Bar: Drop the tetromino
• R: Restart
• Q: Quit
• Y: Change to manual play (A, S to move; K, L to rotate)