tetris rl

an implementation of an RL agent for tetris

Tetris Reinforcement Learning Agent

We developed an open-source reinforcement-learning agent for the classic game Tetris, implemented in Python and available on GitHub here

The project implements a RL framework that learns to play Tetris by interacting with a custom game environment and optimizing a policy that maximizes long-term score. Key features include:

  • A game environment that simulates Tetris gameplay, piece drops, line clears, and scoring.
  • A reinforcement-learning loop (e.g., Deep Q-Learning or policy gradients) where the agent receives observations (board state + next piece) and selects actions (rotate/move/hard-drop) to optimize future reward (cleared lines, survival time).
  • Training dashboards and evaluation scripts that demonstrate the agent’s learning curve over time, metrics such as lines cleared, survival time, and average drop height.
  • Modular code structure to facilitate reinforcement-learning experimentation: different network architectures, reward shaping schemes, state‐representation options (e.g., board heights, aggregate bumpiness, holes), and training hyper-parameters.