Deep Q-Learning Player for the Microservice Dungeon Game

This repository implements a Deep Q-Learning algorithm for the Microservice Dungeon game, which was a main part of my bachelor thesis.

Overview

The Microservice Dungeon is a complex game environment designed to challenge AI agents. This project aims to create an intelligent agent capable of playing the game effectively using Deep Q-Learning, a popular reinforcement learning technique.

Key Features

• Implementation of a Deep Q-Learning algorithm
• Custom environment integration with the Microservice Dungeon game
• Accelerated training using a custom Game Server written in Rust
• Detailed experimentation and performance analysis

Quickstart

1. Start the Custom Game Server
2. Install the requirements.txt for your local Python Environment.
3. Start the Training via python main_dqn.py

Experiments and Results

I conducted several experiments to evaluate the performance of our Deep Q-Learning agent. Below are video demonstrations of key experiments:

Experiment 1: Trained Agent vs. Itself

Experiment 2: Trained Agent (Exp. 2) vs. Trained Agent (Exp. 1)

Experiment 3: Trained Agent (Exp. 1) vs. Random Acting Agent