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。

unity3d unity unity-tech reinforcement-le deep-learning deep-reinforcement-learning neural-networks

文件历史

GitHub 7f0bb6bd Yamato inference tests (#4066 ) * better errors for missing constants * run inference in yamato after training * add extension * debug subprocess args * fix exe path * search for executable * fix dumb bug * -batchmode * fail if inference fails * install tf2onnx on yamato * allow onnx for overrides (expect to fail now) * enable logs * fix commandline arg * catch exception from SetModel and exit * cleanup error message * model artifacts, logs as artifacts, fix pip * don't run onnx * cleanup and comment * update extension handling		4 年前
..
mlagents	Merge pull request #4134 from Unity-Technologies/develop-removebrainnamepolicy	5 年前
tests	Yamato inference tests (#4066)	4 年前
README.md	[refactor] Store and restore state along with checkpoints (#4025)	5 年前
setup.py	[refactor] Structure configuration files into classes (#3936)	5 年前

README.md

Unity ML-Agents Trainers

The mlagents Python package is part of the ML-Agents Toolkit. mlagents provides a set of reinforcement and imitation learning algorithms designed to be used with Unity environments. The algorithms interface with the Python API provided by the mlagents_envs package. See here for more information on mlagents_envs.

The algorithms can be accessed using the: mlagents-learn access point. See here for more information on using this package.

Installation

Install the mlagents package with:

pip3 install mlagents

Usage & More Information

For more information on the ML-Agents Toolkit and how to instrument a Unity scene with the ML-Agents SDK, check out the main ML-Agents Toolkit documentation.

Limitations

mlagents does not yet explicitly support multi-agent scenarios so training cooperative behavior among different agents is not stable.
Resuming self-play from a checkpoint resets the reported ELO to the default value.