Mario-Game-Reinforcement-Learning

Introduction

This project is a reinforcement learning project that uses the PPO algorithm to train an agent to play the game Super Mario Bros. The game is played using the gym-super-mario-bros environment. The agent is trained using the stable-baselines3 library. The agent was trained for 5 million steps and was able to achieve a Total score of 21,650.

Mario Game after training the agent:

• The below gif demonstrates how the agent passes the first level after training it

Training

The agent was trained using the following hyperparameters:

• Total Timesteps: 5,000,000
• Learning Rate: 0.0000003
• number of steps per iteration: 2048

The model was saved each 100,000 steps for evaluation.

Installation

To install the required packages, run the following command:

pip install -r requirements.txt

Usage

To prepare the environment, run the following notebook:

• preprocessing.ipynb

To train the agent, run the following notebook:

• training.ipynb

To test the agent, run the following notebook:

• testing.ipynb

Results

• The agent was able to learn to play the game and achieve a Total score of 21,650.

• The agent performed well on 3,500,000 million steps & was able to finish the first level.

• The agent started to overfit the environment after 3,500,000 million steps and the score started to decrease.