vsa

Visual Steering Agent

#Version and Installation

Neccesary tools to install and run mlagents package

• Unity version: 2020.3.12f1
• Python version: 3.9.9 (Python 3.9.10+ seems to be unstable with ml-agents packages)
• Upgrade pip using command python -m pip install --upgrade pip
• pip version: 22.0.3
• Install torch using command pip install torch~=1.7.1 -f https://download.pytorch.org/whl/torch_stable.html
• numpy version: 1.22.2

To install the mlagents Python package, activate your virtual environment and run from the command line:

python -m pip install mlagents==0.28.0

Install Unity ML agent package in Unity:

• Open Window -> Package Manager
• Choose Packages: Unity Registry in the toolbar in Pakage Manager window
• Find ML Agents in list of packages, install preview version: 1.9.1-preview
• To access preview packages, follow this instruction: https://medium.com/@jeffreymlynch/where-are-the-missing-preview-packages-in-unity-2020-

To run training:

• Make sure agents script is in Default Mode with no input brain.
• Create virtual enviroment inside project folder using command: python -m venv venv
• Activate virtual enviroment: venv\Scripts\activate
• Start training by running command: mlagents-learn when you are inside your virutal enviroment
• If you pause/stop and rerun the command again, you will run into this error: "Previous data from this run ID was found". You can use mlagents-learn --resume to resume, mlagents-learn --force to override previous data or mlagents-learn --run-id={new id} to create a new training brain.