Flappy Bird hack using Deep Reinforcement Learning (Deep Q-learning).

Overview

Using Deep Q-Network to Learn How To Play Flappy Bird

7 mins version: DQN for flappy bird

Overview

This project follows the description of the Deep Q Learning algorithm described in Playing Atari with Deep Reinforcement Learning [2] and shows that this learning algorithm can be further generalized to the notorious Flappy Bird.

Installation Dependencies:

  • Python 2.7 or 3
  • TensorFlow 0.7
  • pygame
  • OpenCV-Python

How to Run?

git clone https://github.com/yenchenlin1994/DeepLearningFlappyBird.git
cd DeepLearningFlappyBird
python deep_q_network.py

What is Deep Q-Network?

It is a convolutional neural network, trained with a variant of Q-learning, whose input is raw pixels and whose output is a value function estimating future rewards.

For those who are interested in deep reinforcement learning, I highly recommend to read the following post:

Demystifying Deep Reinforcement Learning

Deep Q-Network Algorithm

The pseudo-code for the Deep Q Learning algorithm, as given in [1], can be found below:

Initialize replay memory D to size N
Initialize action-value function Q with random weights
for episode = 1, M do
    Initialize state s_1
    for t = 1, T do
        With probability ϵ select random action a_t
        otherwise select a_t=max_a  Q(s_t,a; θ_i)
        Execute action a_t in emulator and observe r_t and s_(t+1)
        Store transition (s_t,a_t,r_t,s_(t+1)) in D
        Sample a minibatch of transitions (s_j,a_j,r_j,s_(j+1)) from D
        Set y_j:=
            r_j for terminal s_(j+1)
            r_j+γ*max_(a^' )  Q(s_(j+1),a'; θ_i) for non-terminal s_(j+1)
        Perform a gradient step on (y_j-Q(s_j,a_j; θ_i))^2 with respect to θ
    end for
end for

Experiments

Environment

Since deep Q-network is trained on the raw pixel values observed from the game screen at each time step, [3] finds that remove the background appeared in the original game can make it converge faster. This process can be visualized as the following figure:

Network Architecture

According to [1], I first preprocessed the game screens with following steps:

  1. Convert image to grayscale
  2. Resize image to 80x80
  3. Stack last 4 frames to produce an 80x80x4 input array for network

The architecture of the network is shown in the figure below. The first layer convolves the input image with an 8x8x4x32 kernel at a stride size of 4. The output is then put through a 2x2 max pooling layer. The second layer convolves with a 4x4x32x64 kernel at a stride of 2. We then max pool again. The third layer convolves with a 3x3x64x64 kernel at a stride of 1. We then max pool one more time. The last hidden layer consists of 256 fully connected ReLU nodes.

The final output layer has the same dimensionality as the number of valid actions which can be performed in the game, where the 0th index always corresponds to doing nothing. The values at this output layer represent the Q function given the input state for each valid action. At each time step, the network performs whichever action corresponds to the highest Q value using a ϵ greedy policy.

Training

At first, I initialize all weight matrices randomly using a normal distribution with a standard deviation of 0.01, then set the replay memory with a max size of 500,00 experiences.

I start training by choosing actions uniformly at random for the first 10,000 time steps, without updating the network weights. This allows the system to populate the replay memory before training begins.

Note that unlike [1], which initialize ϵ = 1, I linearly anneal ϵ from 0.1 to 0.0001 over the course of the next 3000,000 frames. The reason why I set it this way is that agent can choose an action every 0.03s (FPS=30) in our game, high ϵ will make it flap too much and thus keeps itself at the top of the game screen and finally bump the pipe in a clumsy way. This condition will make Q function converge relatively slow since it only start to look other conditions when ϵ is low. However, in other games, initialize ϵ to 1 is more reasonable.

During training time, at each time step, the network samples minibatches of size 32 from the replay memory to train on, and performs a gradient step on the loss function described above using the Adam optimization algorithm with a learning rate of 0.000001. After annealing finishes, the network continues to train indefinitely, with ϵ fixed at 0.001.

FAQ

Checkpoint not found

Change first line of saved_networks/checkpoint to

model_checkpoint_path: "saved_networks/bird-dqn-2920000"

How to reproduce?

  1. Comment out these lines

  2. Modify deep_q_network.py's parameter as follow:

OBSERVE = 10000
EXPLORE = 3000000
FINAL_EPSILON = 0.0001
INITIAL_EPSILON = 0.1

References

[1] Mnih Volodymyr, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex Graves, Martin Riedmiller, Andreas K. Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, Shane Legg, and Demis Hassabis. Human-level Control through Deep Reinforcement Learning. Nature, 529-33, 2015.

[2] Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, and Martin Riedmiller. Playing Atari with Deep Reinforcement Learning. NIPS, Deep Learning workshop

[3] Kevin Chen. Deep Reinforcement Learning for Flappy Bird Report | Youtube result

Disclaimer

This work is highly based on the following repos:

  1. [sourabhv/FlapPyBird] (https://github.com/sourabhv/FlapPyBird)
  2. asrivat1/DeepLearningVideoGames
Owner
Yen-Chen Lin
PhD student at MIT CSAIL
Yen-Chen Lin
A game that depicts a real astronaut's struggles

Interstel-quickscooping-game Right from the beginning of our (i.e, me and me alone) journey in the creation of this game, our goal was to give a game

Sharath V 3 Jul 12, 2021
SuperChess is a GUI application for playing chess.

About SuperChess is a GUI application for playing chess. It is written in Python 3.10 programming language, uses PySide6 GUI library, python-chess lib

Boštjan Mejak 1 Oct 16, 2022
It calculates the Nim sum of a nim game.

nim-sum-calculator It calculates the Nim sum of a nim game. The rules of Nim The traditional game of Nim is played with a number of coins arranged in

2 Jan 02, 2022
XO game with server, client and visualizer for AI bots.

XO game with server, client and visualizer for AI bots.

Ali 4 Jul 14, 2022
Battle of Saiyans: Goku v Vegeta is a 1 v 1, (Player vs CPU) 2D Martial arts fighting game

Battle of Saiyans: Goku v Vegeta is a 1 v 1, (Player vs CPU) 2D Martial arts fighting game inspired by the popular anime series Dragon Ball Z The game

ARZ 3 Feb 16, 2022
A rhythm-based game that automatically generates obstacles based on a song's features.

DISCLAIMER: This is my first coding project, created in December 2019. The game may not be optimized, and looking back on it, there are a lot of chang

Kenneth Huang 1 Dec 27, 2021
TicTacToc - Simple TicTacToc game played by minimax algorithm

TicTacToc simple TicTacToc game played by minimax algorithm. This app is based o

5 Apr 05, 2022
Simple Covid-19 shooter game in python.

Covid_game 🍹 Simple Single Player Covid Game Using Python. 🍹 Has amazing background music theme. 😄 Game Instructions: Initial Health is 5, try to s

Tanya Yadav 2 Aug 05, 2022
a game of life implementation in python

gameoflife-py python implementation of game of life Installing As long as you have bash and curl installed and are on Linux the install script should

Raghav 5 Jun 09, 2021
This Country Hangman game written in Python.

country-name-guess-hangman-game This Country Hangman game written in Python. Visit https://example.com to play the game. Description How to play this

Aytaç Kaşoğlu 1 Oct 30, 2021
Super Mario Kart November 1991 Prototype Repair by MrL314

Super Mario Kart November 1991 Prototype Repair by MrL314

MrL314 51 Dec 26, 2022
Snake (PyGame-based) port for Minecraft:Bedrock Edition using PEWSAPI

Snake_PEWSAPI Snake (PyGame-based) port for Minecraft:Bedrock Edition using PEWSAPI And we are not going to make any change to the original Snake sour

Azuki 1 Mar 17, 2022
A Tetris Game for programming education

Tetris Game プログラミング学習を目的とした、ブロックを操作してスコアを競うゲームです。 FAQはこちら。 tutorialはこちら。 実行環境準備 Mac環境 Finder→Application→Utility→Terminalから、ターミナルを起動して以下コマンドを実行する。 # i

11 Dec 01, 2022
Simple python 3D vector3 math library wrapping some types from GLM library using pybind11.

vmath Simple python 3D vector3 math library wrapping some types from GLM library using pybind11. Description Both pure python version and C++ version

6 Aug 02, 2022
This is an interactive MiniMap made with Python, PyQT5 & Pytesseract for the game

NWMM-New-World-MiniMap Features: Automatically grabs position from "New World" Instance Live visualisation of player position on MiniMap Circular & re

Nezzquikk 18 Sep 21, 2022
A 16x16 clone of Minecraft Classic, written in Python with the Ursina Engine

VoxelCraft A 16x16 clone of Minecraft Classic, written in Python with the Ursina Engine Features:Trees, Water(But there's no gravity, so if you break

2 Jun 23, 2022
A Minecraft clone written in python and pyglet.

PyCraft A Minecraft clone written in python and pyglet. Running PyCraft To run PyCraft, run the following code: git clone https://github.com/TheWebCra

The WebCrafters 17 Dec 29, 2022
offline bot for game on chrome

Бот офлайн игры браузера CHROME В автоматическом режиме запускает браузер Chrome под ОС windows, так же автоматически определяет разрешения экрана, на

Andrej Marinchenko 19 Dec 17, 2022
Utility to find games owned by all (or at least some) of the passed players.

SteamCommonGameFinder Utility to find games that are owned by all (or at least some) of the players you pass into this programm. You can already find

Daniel O'Grady 4 Jan 04, 2022
Brawl Stars v31.96 server emulator written in Python.

Brawl Stars v31 Brawl Stars v31.96 server emulator written in Python. Requirements: Python 3.7 or higher pymongo dnspython colorama MongoDB configurat

9 Nov 26, 2021