This project was about evolving an agent which had to be able to play Flappy Bird game as long as possible without hitting any pipes. The agend was controlled by a feed-forward neural network and the neural network was trained by using reinforcement learning and evolutionary algorithms (more specifically, differential evolution and NEAT).
Here you can find a more detailed report about the techniques and the method used.
The left image shows how a random agent performs, while the right image shows how a final evolved agent performs.
The project was tested on Ubuntu 16.04 and it is compliant with Python 3.6. We used Pipenv in order to manage the various dependencies. They can be installed by issuing:
cd evolutionary-FlappyBird
pipenv install
pipenv shellThe file evo_flappy.py can be used for training/evaluating the Flappy Bird agent. By issuing python3 evo_flappy.py -- help, the list of possible actions and flags will be displayed. For instance, the command shown below will train an agent
using Differential Evolution and then it will play the game with the best individual found.
python3 evo_flappy.py --MODE_AGENT --MODE_LEARN --PLAY_BEST --MODE_NO_SCREEN --EA="DE"Giovanni De Toni: giovanni.detoni@studenti.unitn.it
Seyed Mahed Mousavi: seyedmahed.mousavi@studenti.unitn.it
Andrei Catalin Coman: andreicatalin.coman@studenti.unitn.it