Reinforcement Learning for Catan: DQN/A3C Implementation

Overview

This project presents a PyTorch-based implementation of two reinforcement learning algorithms: Asynchronous Advantage Actor Critic (A3C) and Deep Q-learning. They are applied to a custom-designed, 2-player environment of the board game Catan, fully integrating the game's rules, except for player-player trading.

To train the reinforcement learning agents, you can adjust hyperparameters in the config file.

Achievements

Both of the models have been optimized and tested using Ubuntu on WSL2.

• The A3C model demonstrates significant success, attaining an 87% win rate against a random agent as an opponent.
• The DQN model is currently under development, with anticipated improvements following the refinement of the A3C implementation.

Project Status

Please note that this project is actively being developed. Further updates and enhancements are expected.