DeepLearningFlappyBird

Why we included this project

For years this has been the project people point to when they want to watch deep Q-learning do something concrete, and it still holds up as a teaching tool. It feeds raw game-screen pixels into a convolutional network and applies the experience replay and target-value ideas from the original Atari DQN paper to the simpler Flappy Bird game. The readme lays out the algorithm and the screen preprocessing step by step, and the training code is short enough to read in one sitting, so the repo reads like a self-contained tutorial rather than a production system. If you are learning reinforcement learning or just want a small, readable baseline to adapt to your own environments, this is a good place to start.