Name		Name	Last commit message	Last commit date
parent directory ..
assets/sprites		assets/sprites
game		game
model		model
README.md		README.md
requirements.txt		requirements.txt
test.py		test.py
test_flappy_bird.sh		test_flappy_bird.sh
train.py		train.py
train_flappy_bird.sh		train_flappy_bird.sh

README.md

Playing Flappy Bird Using Deep Reinforcement Learning

Implementation of DQN to play the Flappy Bird game with Oneflow framework .

This work is based on: uvipen/Flappy-bird-deep-Q-learning-pytorch and Ldpe2G/DeepLearningForFun

screenshots

Environment

Spec
Operating System	Ubuntu 18.04
GPU	Nvidia A100-SXM4-40GB
CUDA Version	11.2
Driver Version	460.73.01
Oneflow Version	branch: master, commit_id: 90d3277a098f483d0a0e68621b7c8fb2497a6fc2

Requirements

python3
- pygame
- numpy
- opencv
Oneflow

Pretrain Model

DQN_FlappyBird.zip

Play with pretrain model

bash test_flappy_bird.sh

Train from scratch

bash train_flappy_bird.sh

It will take about at least 40000 time steps before the bird can learn to play the game, be patients :).