🧠 Quoridor-RL: A Multi-Agent RL Environment for the Quoridor Board Game

Quoridor-RL is a Python-based simulation environment for the Quoridor board game, compatible with the Gymnasium API and multi-agent RL training using Ray RLlib. It also includes a 3D interactive web viewer built with Three.js, and supports human interaction, imitation learning, and deployment on Hugging Face Spaces.

✨ Features

• ✅ Fully rule-compliant 2- and 4-player Quoridor engine
• ♻️ Multi-agent training support (Ray RLlib)
• 🎮 Human-agent interaction mode
• 📈 Episode logging for imitation learning
• 🧠 Gymnasium-compatible environment
• 🌐 3D visualization in the browser (Three.js)
• 🚀 Hugging Face Space ready

📦 Installation

Python Game Engine

pip install quoridor-env

JavaScript Viewer

cd web-viewer/
npm install
npm run dev

🧪 Quick Start: Training with Ray RLlib

from ray.rllib.algorithms.ppo import PPOConfig
from quoridor.env import QuoridorEnv

config = PPOConfig().environment(env=QuoridorEnv, env_config={...})
algo = config.build()
algo.train()
See training_examples/ for full MARL setups.

🌍 Web Viewer The web UI is built with Three.js and supports:

• Live replay of RL episodes
• Human-vs-agent gameplay
• Drag-and-drop wall placement
• API polling or WebSocket update

Launch

cd web-viewer/
npm run dev

📊 Observation & Action Spaces Observation Shape: (C, 9, 9) Channels:

• Player positions
• Opponent positions
• Goal rows
• Horizontal/vertical walls

Action Discrete:

• 0–3: move in cardinal directions
• 4–N: place wall at encoded location and orientation

See docs/observation_action.md for details.

🧠 Imitation Learning Support Episodes are logged to JSONL or .npz with:

• observations, actions, rewards, next_observations, dones

Compatible with HumanCompatibleAI/imitation:

imitation-train expert_demos.npz --algo=bc

🚀 Hugging Face Space You can deploy the environment and viewer together on Hugging Face Spaces.

Example layout:

huggingface-space/
├── app.py     # Gradio or Flask API server
├── assets/    # Quoridor model data
└── web/       # Built JS assets
📜 License
MIT License. See `LICENSE`.

🤝 Contributing Pull requests and feature suggestions are welcome! Please open an issue to discuss major changes first.