MKII_AI

An Artificial Intelligence agent developed to play Mortal Kombat II (Genesis) using the Stable-Retro library.

This project explores Reinforcement Learning techniques applied to classic video games, enabling agents to learn decision-making strategies based on the game state.

---

Project Goals

The main objectives of this project are:

• Train an agent to play Mortal Kombat II
• Explore different state representations (RAM vs. image-based observations)
• Evaluate agent performance throughout training
• Provide a foundation for Reinforcement Learning experiments in retro games

---

Technologies Used

• Python 3
• Stable-Retro
• Stable-Baselines3
• Gymnasium
• NumPy

---

Development Setup

Linux / WSL 2

First, install the required system dependencies:

sudo apt-get update
sudo apt-get install python3 python3-pip git zlib1g-dev libopenmpi-dev ffmpeg freeglut3-dev

Using the Stable-Retro UI

If you want to use the Stable-Retro graphical interface:

git clone https://github.com/Farama-Foundation/stable-retro.git
cd stable-retro

pip3 install -e .

sudo apt-get install capnproto libcapnp-dev libqt5opengl5-dev qtbase5-dev zlib1g-dev
sudo apt install cmake

cmake . -DBUILD_UI=ON -UPYLIB_DIRECTORY
make -j$(grep -c ^processor /proc/cpuinfo)

Run the interface with:

./gym-retro-integration

Without the UI

If you only need the library:

pip3 install stable-retro

Clone the Repository

git clone https://github.com/Antonio-Secchin/MKII_IA.git
cd MKII_IA

Install Python Dependencies

pip install -r requirements.txt

Import the Game ROM

A legal copy of the Mortal Kombat II (Genesis) ROM is required.

Place the ROM file in a folder and run:

python3 -m retro.import .

---

Research Topics

This project investigates several Reinforcement Learning challenges, including:

• Learning from RAM observations
• Learning from image observations
• Reward engineering
• Agent evaluation and benchmarking

---

Publication

This repository accompanies my Bachelor's Thesis:

"Comparação de Técnicas de Aprendizado por Reforço para o Treinamento de Agentes Inteligentes em Jogos de Luta"

Available in:

• Thesis PDF (Docs/Comparação de Técnicas de Aprendizado por Reforço para o Treinamento de Agentes Inteligentes em Jogos de Luta.pdf)

Disclaimer

This repository does not distribute copyrighted game ROMs. Users must provide their own legally obtained copy of Mortal Kombat II.