Flappy Bird for Gymnasium

This repository contains the implementation of Gymnasium environment for the Flappy Bird game. The implementation of the game's logic and graphics was based on the flappy-bird-gym project, by @Talendar.

State space

The "FlappyBird-v0" environment, yields simple numerical information about the game's state as observations or RGB-arrays (images) representing the game's screen.

FlappyBird-v0

• the last pipe's horizontal position

• the last top pipe's vertical position

• the last bottom pipe's vertical position

• the next pipe's horizontal position

• the next top pipe's vertical position

• the next bottom pipe's vertical position

• the next next pipe's horizontal position

• the next next top pipe's vertical position

• the next next bottom pipe's vertical position

• player's vertical position

• player's vertical velocity

• player's rotation

• or RGB-array (image) representing the game's screen

Action space

• 0 - do nothing
• 1 - flap

Rewards

• +1 - every frame it stays alive
• +100 - successfully passing a pipe
• -100,0000- dying

Installation

To install flappy-bird-gymnasium, simply run the following command:

$ pip install flappy-bird-gymnasium

Usage

Like with other gymnasium environments, it's very easy to use flappy-bird-gymnasium. Simply import the package and create the environment with the make function. Take a look at the sample code below:

import flappy_bird_gymnasium
import gymnasium
env = gymnasium.make("FlappyBird-v0", render_mode="human")

obs, _ = env.reset()
while True:
    # Next action:
    # (feed the observation to your agent here)
    action = env.action_space.sample()

    # Processing:
    obs, reward, terminated, _, info = env.step(action)
    
    # Checking if the player is still alive
    if terminated:
        break

env.close()

Playing

To play the game (human mode), run the following command:

$ flappy_bird_gymnasium

To see a random agent playing, add an argument to the command:

$ flappy_bird_gymnasium --mode random

To test the DQN related algorithms for CS184 project, you can run:

$  python flappy_bird_gymnasium/my_priority.py -t
$  python flappy_bird_gymnasium/my_dqn.py
$  python flappy_bird_gymnasium/my_dueling_dqn.py -t

For the policy gradient related methods for our project, please refer to this link (https://github.com/JerryGao9527/Flappy-Bird-Gameplay-with-RL).