caoandong/Animal-AI-Training

Animal-AI Reverse-Engineering

This is an Unity project that reverse engineer the basic mechanisms of Animal-AI to fascilitate better training. The core features of this project are:

• ACchieves the basic functionalities of the Animal-AI environment, including
  • Randomized initialization of different food (great food: yellow, good food: green, bad food: red)
  • Restart episode when the player eats a food.
  • Records score for each episode. W S A D Control of player movement.
• Displays the masks of the objects in the player's view.
• Simulates the depth-sensing camera in the player's view.
• Records each frame (RBG, mask, depth) in the Asssets/StreamingAssets folder.
• Training using Unity ML agent.

RGB Screenshot:

Mask Screenshot:

Depth Screenshot:

Installation

Prerequisites

1. Unity Editor 2019.2.1f1 - The main game engine
2. Windows 10 Pro
3. Microsoft Visual Studio 2019 Version 16.2.3 - Go to Help -> About Microsoft Visual Studio, and check that the following products are installed:
   • C# tools
   • Visual Studio Tools for Unity

Install From Unity Package

1. Create a new project for Unity 2019.2.1f1.
2. Download and import (Assets > Import Package > Custom Package) the Animal-AI package.
3. Setup the World Manager and the Player:
   • Set the linear and rotational speed for the player.
   • Check the toggle for Take Screen Shot.

Settings for the world manager and the player: