FLAPPY BIRD AUTOMATION USING DEEP REINFORCEMENT ALGORITHM

FLAPPY BIRD AUTOMATION USING DEEP REINFORCEMENT ALGORITHM

Arya Sharma, Naman Malhotra, Sameer, Sabahul, Abhinav

Department of Information Science and Engineering, Chandigarh university,

Mohali, Punjab – 140413

ABSTRACT

It is fascinating to see the increasing interest in game learning within the field of Artificial Intelligence. One noteworthy game that has been developed in this context is Flappy Bird. In this game, the agent   learns how to manoeuvre through obstacles and aims to maximise its score by receiving rewards and punishments. Notably, the agent is not provided with any prior knowledge regarding the game environment. In contrast to using raw pixels as input, the agent is trained using domain-specific features. These features include parameters such as the bird's speed, the distance between pipes, and the height of the pipes. The paper discusses the application of the reinforcement deep learning algorithm and the use of domain-specific feature extraction for training an agent to play Flappy Bird. Through this approach, the agent can autonomously learn how to navigate the game and maximise its score without prior knowledge about the game environment. This research sheds light on the exciting advancements in game learning and the potential of domain-specific approaches in training AI agents.

KEYWORDS-featureextraction, pygame,reward,fitness,reinforcement learning.