Unity 机器学习代理工具包 (ML-Agents) 是一个开源项目,它使游戏和模拟能够作为训练智能代理的环境。
您最多选择25个主题 主题必须以中文或者字母或数字开头,可以包含连字符 (-),并且长度不得超过35个字符
 
 
 
 
 
vincentpierre fef449f0 Now collecting the states in the PlayerCoreBrain in order to test and debug the CollectState method. 7 年前
..
Assets Now collecting the states in the PlayerCoreBrain in order to test and debug the CollectState method. 7 年前
ProjectSettings Initial commit 7 年前
README.md Update README.md 7 年前

README.md

Unity ML - Agents (Editor SDK)

diagram

Unity Setup

Make sure you have Unity 2017.1 or later installed. Download link available here.

Building a Unity Environment

  • (1) Open the project in the Unity editor (If this is not first time running Unity, you'll be able to skip most of these immediate steps, choose directly from the list of recently opened projects and jump directly to )
    • On the initial dialog, choose Open on the top options
    • On the file dialog, choose ProjectName and click Open (It is safe to ignore any warning message about non-matching editor installation")
    • Once the project is open, on the Project panel (bottom of the tool), click the top folder for Assets
    • Double-click the scene icon (Unity logo) to load all game assets
  • (2) File -> Build Settings
  • (3) Choose your target platform:
  • (opt) Select “Developer Build” to log debug messages.
  • (4) Set architecture: X86_64
  • (5) Click Build:
    • Save environment binary to a sub-directory containing the model to use for training (you may need to click on the down arrow on the file chooser to be able to select that folder)

Example Projects

The Examples subfolder contains a set of example environments to use either as starting points or templates for designing your own environments.

  • 3DBalanceBall - Physics-based game where the agent must rotate a 3D-platform to keep a ball in the air. Supports both discrete and continuous control.
  • GridWorld - A simple gridworld containing regions which provide positive and negative reward. The agent must learn to move to the rewarding regions (green) and avoid the negatively rewarding ones (red). Supports discrete control.
  • Tennis - An adversarial game where two agents control rackets, which must be used to bounce a ball back and forth between them. Supports continuous control.

For more informoation on each of these environments, see this documentation page.

Within ML-Agents/Template there also exists:

  • Template - An empty Unity scene with a single Academy, Brain, and Agent. Designed to be used as a template for new environments.

Agents SDK Package

A link to Unity package containing the Agents SDK for Unity 2017.1 can be downloaded here :

For information on the use of each script, see the comments and documentation within the files themselves, or read the documentation.

Creating your own Unity Environment

For information on how to create a new Unity Environment, see the walkthrough here. If you have questions or run into issues, please feel free to create issues through the repo, and we will do our best to address them.

Embedding Models with TensorflowSharp [Experimental]

If you will be using Tensorflow Sharp in Unity, you must:

  1. Make sure you are using Unity 2017.1 or newer.
  2. Make sure the TensorflowSharp plugin is in your Asset folder. A Plugins folder which includes TF# can be downloaded here.
  3. Go to Edit -> Project Settings -> Player
  4. For each of the platforms you target (PC, Mac and Linux Standalone, iOS or Android):
    1. Go into Other Settings.
    2. Select Scripting Runtime Version to Experimental (.NET 4.6 Equivalent)
    3. In Scripting Defined Symbols, add the flag ENABLE_TENSORFLOW
  5. Restart the Unity Editor.