Used Reinforcement Learning to train an Agent to Play Flappy Bird (CNN currently not included)
Average score over 100 games is approximately 65 after training. Looking to improve the algorithm. Please open an issue if you have any good ideas
- Arduino-controlled "tapper". Needs to account for control delay.
- Using raw pixel data with a CNN
Mnih, Volodymyr, et al. “Human-Level Control through Deep Reinforcement Learning.” Nature News, Nature Publishing Group, 25 Feb. 2015, www.nature.com/articles/nature14236.
- Used a Modified version of the Open-Source Flappy Bird Game from https://github.com/sourabhv/FlapPyBird
- QNetwork implementation inspiration from https://www.youtube.com/channel/UC58v9cLitc8VaCjrcKyAbrw