Skip to main content

Welcome to Reinforce Tactics

Reinforce Tactics Logo

Reinforce Tactics is a modular turn-based strategy game built specifically for reinforcement learning research and experimentation. This project combines classic tactical gameplay with modern RL capabilities, providing a rich environment for developing and testing reinforcement learning algorithms.

🎮 What is Reinforce Tactics?​

Reinforce Tactics is a 2D turn-based strategy game featuring:

  • Turn-based tactical gameplay with multiple unit types
  • 4 unit types: Warrior, Mage, Cleric, and Archer (each with unique abilities)
  • Combat system with attacks, counter-attacks, paralysis, and healing
  • Economic system with income from controlled structures
  • Structure capture: Towers, Buildings, and Headquarters
  • Save/Load system for continuing games
  • Replay system for watching past games
  • AI opponents: SimpleBot and LLM-powered bots (GPT, Claude, Gemini)
  • Full Gymnasium integration for RL training
  • Headless mode for fast training without rendering
  • Multiple training algorithms: PPO, A2C, DQN via Stable-Baselines3
  • Docker support for easy deployment

🤖 Why Reinforce Tactics?​

This project is designed to be:

  • Research-Friendly: Modular architecture makes it easy to extend and customize
  • RL-Ready: Standard Gymnasium interface for seamless integration with RL libraries
  • Educational: Clear code structure perfect for learning RL and game development
  • Performant: Headless mode enables fast training without rendering overhead

🚀 Quick Start​

Installation​

# Basic Installation (GUI Mode)
pip install pygame pandas numpy

# Full Installation (with RL)
pip install pygame pandas numpy gymnasium stable-baselines3[extra]

Play the Game​

python main.py --mode play

Train an RL Agent​

python main.py --mode train --algorithm ppo --timesteps 1000000 --opponent bot

Use as Gymnasium Environment​

from reinforcetactics.rl.gym_env import StrategyGameEnv

# Create environment
env = StrategyGameEnv(
    map_file='maps/1v1/test_map.csv',
    opponent='bot',
    render_mode=None
)

# Standard Gym API
obs, info = env.reset()
action = env.action_space.sample()
obs, reward, terminated, truncated, info = env.step(action)

📚 Documentation Structure​

This documentation is organized into several sections:

  • Getting Started (this page): Overview and quick start guide
  • Implementation Status: Current state of the project and completed features

🔗 Useful Links​

💡 Contributing​

Contributions are welcome! Whether you're interested in:

  • Improving RL algorithms
  • Adding new unit types or game mechanics
  • Enhancing documentation
  • Fixing bugs

Feel free to open issues or submit pull requests on GitHub.

📜 License​

This project is licensed under the MIT License - feel free to use and modify!