Mini project: Implementation of Deep Q-learning algorithm to create an AI agent capable of scoring high on Flappy Bird game
The algorithm implemented is fully described in this paper: Playing Atari with Deep Reinforcement Learning (Deep Q-learning with experience replay, page 5)
A demo video can be found here:
- gym
- gym_ple
- Pygame-Development-Environment
- Pygame
- keras
To train from scratch, just run the command python flappy_bird.py -r
You can also download pretrained weights to ./model
folder in the current dir and run the command python flappy_bird.py -l -r
to see how the trained AI plays.
The training is undertaken on a machine with GTX860 gpu. So you need a GPU to reproduce the results.