Polytopia RL

A reinforcement learning platform for The Battle of Polytopia — a turn-based strategy game with complex decision spaces, multi-agent competition, and hundreds of possible actions per turn.

The project includes a Java game engine, a real-time Python bridge via Py4J, and a standard Gymnasium environment compatible with any RL algorithm. Built at HackMIT.

Architecture

• Java Game Engine Full headless simulation — units, cities, tech trees, combat, terrain, and resources.
• Py4J Bridge Real-time Java-Python interop. Actions in, observations out, every tick.
• Gymnasium Environment Standard reset() / step() / render() API. Compatible with PPO, DQN, and other RL algorithms.
• Web Visualizer Live game playback, step-through controls, video recording, and JSON export.

Tech Stack