FlappyBirdySmart

This is a Unity project I made to get a better understanding of implementation of Unity's ML-Agents library(v0.15)

Installation

1. Unity 19.3.0f6
2. MLAgents Toolkit 0.15
3. Python 3.7
4. Tensorflow 2.0.1

Overview of the game mechanics

• A replica from the popular game Flappy Bird
• The bird can sort of fly up to avoid the obstacles
• The gravity pulls the bird down
• The obstacles are spawned randomly with variation in gaps between them

Assets:

• The backgrounda and obstacles (pipes) created using GIMP
• The bird sprites used from xstrack1's assets

ML Agents(v0.15)

• Unity's mlagents toolkit, made for creating a game AI
• Please refer to the official description and setup here

Performance

• The Ai currently plays the game without crashing for about 400 seconds, still further training is required and more tweaking of the hyperparameters

To See the AI play the game

• On the agent script in the editor make sure the model has been added just hit play andd watch it fly! Note: there are multiple models in the assets and currently i have added the model named FlappyBirdMoreSensorsMoreGravity.nn, not all the models are compatible as i have made changes in vector observations with different models to get the optimum performance