/ml-agents

Unity Machine Learning Agents

Primary LanguageC#Apache License 2.0Apache-2.0

Unity ML - Agents (Beta)

Unity Machine Learning Agents allows researchers and developers to create games and simulations using the Unity Editor which serve as environments where intelligent agents can be trained using reinforcement learning, neuroevolution, or other machine learning methods through a simple-to-use Python API. For more information, see the documentation page.

For a walkthrough on how to train an agent in one of the provided example environments, start here.

Features

  • Unity Engine flexibility and simplicity
  • Multiple observations (cameras)
  • Flexible Multi-agent support
  • Discrete and continuous action spaces
  • Python (2 and 3) control interface
  • Visualizing network outputs in environment
  • Tensorflow Sharp Agent Embedding [Experimental]

Creating an Environment

The Agents SDK, including example environment scenes is located in unity-environment folder. For requirements, instructions, and other information, see the contained Readme and the relevant documentation.

Training your Agents

Once you've built a Unity Environment, example Reinforcement Learning algorithms and the Python API are available in the python folder. For requirements, instructions, and other information, see the contained Readme and the relevant documentation.