w3ntao / q-bird

Flappy Bird with Q-learning

Flappy Bird with Q-Learning

A simple implementation to Playing Atari with Deep Reinforcement Learning.

Train the bird online at https://w3ntao.github.io/q-bird.
Train the bird locally:

1. host a server with python: $ python3 -m http.server 8000
2. type http://0.0.0.0:8000 in a browser to play

Pseudocode

Q ← {}
state-seq ← []

for each round:
    for each frame:
        S ← current state
        if S in Q:
            if Q(S, flap) > Q(S, do-nothing):
                A ← flap
            else:
                A ← do-nothing
        state-seq ← state-seq + [S, A]

        if termination:
            for [S, A] in reversed(state-seq):
                if S is the 1st (right before termination):
                    Q(S, A) ← (1 - α) * Q(S, A) + α * R
                else:
                    S' ← next state of S
                    Q(S, A) ← (1 - α) * Q(S, A) + α * {R + γ * max[Q(S', flap), Q(S', do-nothing)]}
            state-seq ← []

α: learning rate
γ: discount factor
R: reward
       +1 for survival
    -1000 for death

After 1000 Rounds' Iteration

Credits

https://github.com/enhuiz/flappybird-ql
https://github.com/chncyhn/flappybird-qlearning-bot