DQN-for-FlappyBird

An implementation of DeepMind's Deep-Q-Network agent to play the notorious FlappyBird game.

Setting up the environment

Here are the steps that you need to follow to train or test the model in action. Just run the following commands:

• First make a python virtual environment

  virtualenv env -p python3
  

• Install the dependencies as listed in the requirements.txt

  pip install -r requirements.txt
  

• Run the python script deep_q_network.py, if you have checkpoints stored in the saved_networks directory, the latest checkpoint will be used to play the game.

  python deep_q_network.py
  

Model used

Here is the model architecture that I used to train the model. The model was conceptualized by Deep Mind for playing ATARI Games. This architecture is based on that.

Model in action / Gameplay

What just happend?

This gif is a demo of the DQN Agent trained to play this game. To know about the DQNs try following this blog post.