aadarshjha / flappy

Q-Learning In Flappy Bird, CS 5891 Final Project

Applying Reinforcement Learning To Infinite-Play And Random Environments: Exploring Optimal Performance In Flappy Bird Through Varying Q-Learning Algorithms

Aadarsh Jha
CS 5891: Reinforcement Learning
December 10th, 2021

Instructions On Running:

Please follow the below instructions to see how both trials of the experiments are run, with respect to Deep Q-Learning and Q-Learning.

• First, git clone https://github.com/aadarshjha/flappy.git or download this repository from the above link.

• Then, cd flappy/ and follow the below instructions to run the experiments.

Deep Q-Learning

• To execute testing and training of the Deep Q-Learning pipeline, cd to the deep-q-learning directory.
• Open the deep-q-learning.ipynb, instructions are provided in the notebook itself so as to get you started.
• Note that models are provided from the best performing trial, more are available at request. Please email me: [e-mail]

Q-Learning

• To execute testing and training of the Deep Q-Learning pipeline, cd to the q-learning directory.
• Open the q-learning.ipynb, instructions are provided in the notebook itself so as to get you started.
• Note that models are provided from the best performing trial, more are available at request. Please email me: [e-mail]

Figures

• Due to the number of charts in the paper, some figures may be hard to read. cd figures/ to find high resolutoin .pdf files that are the figures.

References

• As stated in the paper the following repos are drawn inspriation from to execute the Deep Q-Learning and Q-Learning experiments:
  1. Deep Q-Learning: https://github.com/yenchenlin/DeepLearningFlappyBird
  2. Q-Learning: https://github.com/yashkotadia/FlapPy-Bird-RL-Q-Learning-Bot
  3. Enviornment: https://github.com/sourabhv/FlapPyBird