Flappy Bird Game Based on Reinforcement Learning Q-Learning Algorithm

Abstract

In the field of Artificial Intelligence (AI), game AI is becoming more and more important, and the human-machine training of games is gradually driving the development of the game field. Among them, Flappy Bird is one of game which can controlled by an AI, which deserves more attention. In this work, we used Q-Learning as our main algorithm of the AI. In the flappy bird AI, the algorithm of Q-learning is used for giving the feedback through the environment which corresponding reward according to the actions of the agent. By using this method and after the training of the flappy bird AI, we can get the scores that are much more than human’s record. The highest record of the flappy bird AI is 4, 083. The average score for human is about only 100, but in the flappy bird AI, the score can easily be more than 1, 000. According to all the work we did and all the result we got, we can see that the comparison between the AI and human. In the game area, AI did much better than human in most game. That is the reason that much research is focusing on developing game AI to help us getting deeper in the game field since it is more efficient to use.