This project is the result from an excerise from the lecture "Interactive Machine Learning" at the University of Augsburg. The goal of the excersice was to train a deep reinforcement network to play flappy bird for as long as possible.
To start the application download the project and run the main inside flappyD.py.
I was able to achieve a perfectly trained model after ~33.000 iterations of simulations. Afterwards I rerun the application several times without the bird dying. The highest score achieved by this model was ~550.000. However, the model most likely is able to flapp the bird for an infinite time without dying.
You can find the model under 'flappy_weight.h5'.