Google-Research-Football-RL ⚽

Creating a AI-agent that can play football in the google research football environment.Project based on Kaggle Competition .Undergraduate Thesis for Computer Science and Engineering - University of Ioannina.

Read the paper - here 📄

Setup the environment

• All the stuff you need to know about environment and python libriaries is - here

• Observations and Actions list is - here

• Compile the GRF-Engine instructions

Interact with GRF environment

You can interact with GRF environment and learn how to use it through Simple_Observations.ipynb

Learning & Playing 📖

Implemented algorithm Deep Q-Learning(DQN) based on Deepmind's paper

We need :

• The Experience Replay Buffer
• Epsilon Greedy action selection algorithm

We dont need:

• Target neural network

Scenarios

1. Academy_empty_goal (agent learns to score doing the minimum number of steps) Empty_goal.ipynb
2. 1vs1 (agent learn to score with opponent) 1vs1.ipynb