Polytopia RL

Project Summary

Polytopia RL is a research framework for training and analyzing reinforcement learning agents in a turn-based strategy environment. We combine the Tribes Java engine (forward model, rules, GUI) with a lightweight Python bridge (Py4J) and Gym-style APIs for rapid prototyping, evaluation, and visualization.

Key features:

• Java engine (Tribes) for deterministic forward modeling and rich game dynamics
• Python bridge exposing reset/step/list_actions/observation JSON
• Multiple renderers: Java GUI, ANSI text, PIL images, and a simple web GUI
• GitHub Codespaces support for one‑click cloud dev environments

Demo (Video)

Watch a short demo here: https://www.youtube.com/watch?v=xPcRPhiPnsg

Project Slideshow

Slides with methods and visuals: https://docs.google.com/presentation/d/1oripsNAIIUnl0mX7U4JyGPdtTLTwYCnO9eWK9A8KGiU/edit?usp=sharing

Simple Web GUI

• Location: pol_env/Tribes/py/simple_web_gui.py
• Usage (local):
  1. Create and activate a venv
     • python3 -m venv pol_env/Tribes/.venv
     • source pol_env/Tribes/.venv/bin/activate
  2. Install requirements
     • pip install -U pip wheel
     • pip install -r pol_env/Tribes/py/requirements.txt
     • For the web GUI: pip install gymnasium imageio numpy
  3. Run the server
     • cd pol_env/Tribes/py
     • python simple_web_gui.py
     • Open http://localhost:8000

Java UI and Python Bridge (quickstart)

• Java UI (after compiling):
  • cd pol_env/Tribes
  • java -cp out:lib/json.jar Play
• Python demo:
  • cd pol_env/Tribes/py
  • python run_gym.py