RL_FlappyBird

Overview

Using DQN to play Flappy Bird. Implemented by PyTorch.

Improvement: Fixed Q-Network

RL.md: Notes about DQN

Dependencies:

• PyTorch 1.4.0
• pygame 1.9.6

How to Run?

play the game with pretrained weights:

(set FPS to 480, achieve 2000 score in 10 minutes.)

python main.py --play

train:

python main.py

resume training with pretrained weights:

python main.py --resume --ckpt_path 'path'

Disclaimer

This work is based on the following repos:

1. sourabhv/FlapPyBird
2. yenchenlin/DeepLearningFlappyBird
3. xmfbit/DQN-FlappyBird