This is a project for utilizing reinforcement learning in Flappybird, allowing players to compete with an DQN-based AI agent.

(Left AI, Right Player)

·Epoch 1000 Score：1-2

·Epoch 10k Score：20+

·Epoch 100w Score：200+

Q-Learning

Q-Learning iteration function

Using neural network to fit Q(s, a)

Main requirements:

torch
cv2
pygame

Space→Jump