/FlappyBirdRL

Primary LanguagePython

FlappyBirdRL

DQN applied to Flappy Bird with pygame and PyTorch.

  • FlappyBird_original/train.py runs the baseline DQN algorithm without fixed target Q network
  • FlappyBird_original/double_train.py runs the Double DQN version
  • FlappyBird_original/double_train_pr.py runs the Double DQN with Prioritized Replay
  • FlappyBird_original/inference.py runs inference of a trained model
  • FlappyBird_reward_shaping/*train*.py runs the aforementioned algorithms with the modified reward that quantifies the closeness of the bird to the target pipe gap when it crashes a pipe.

Credit:

  1. https://github.com/uvipen/Flappy-bird-deep-Q-learning-pytorch