Deep Reinforcement Learning implementation in Keras of an AI controlling the popular Flappy Bird videogame, using Asynchronous Advantage Actor Critic (A3C).
By Federico Berto: Bachelor's degree thesis for Joint double degree for Tongji University and University of Bologna. For more details, please contact me at: [email protected]
Original implementation: For more information about the project details, see this blog post associated with this project.
See LICENSE for further details Copyright (c) 2019 Federico Berto based on: Copyright (c) 2017 Shalabh Singh
- Python 3.5
- Keras 2.0
- pygame
- scikit-image
Clone the repository or download it. To test the pretrained model, run the "test.py" file. To retrain the model from scratch, run the "train_network.py" file and the trained models at different stages will be saved in "saved_models" folder.
This work is based on the following repos and blogs-
- https://github.com/yenchenlin/DeepLearningFlappyBird
- https://github.com/jaara/AI-blog
- http://karpathy.github.io/2016/05/31/rl/
- play.py --> press SPACE to flap the bird
- test.py --> test one network model
- train_network.py --> train a network
Don't esitate to contact me for ideas, pull requests and more :)