AT-ST Walker Agent using Unity ML-Agents

About the project

This machine learning agent (reinforcement learning agent) was developed for the Intelligent Systems module of the University of Applied Sciences Duesseldorf, Germany.

• Find the full agent concept

  • in German (original): ATST-Walker_Konzept.pdf
  • in English (DeepL translated): ATST-Walker_Concept_EN.pdf

• Find the full agent documentation

  • in German (original): ATST-Walker_Doku.pdf
  • in English (DeepL translated): ATST-Walker_Documentation_EN.pdf

Technical details

• Unity Version

  • 2020.3.7f1

• Required packages

  • ML Agents (2.0.0-exp.1)
  • ML Agents Extensions (0.4.0-preview)

• Project repository

  • Assets/
    • AT-ST_Walker/
    • ML-Agents/

ML-Agents Set-up

Compatible Versions

Version	Python Package	Unity Package
Release 17	0.26.0	2.0.0

Source-Files: https://github.com/Unity-Technologies/ml-agents/tree/release_17

Installing ML-Agents dependencies in Unity

• If you open this project in Unity, this error might occur:

• To solve it, download the Source-Files of ML-Agents Release 17 mentioned before.

• In Unity:

  • Click "Continue" (Error window) and import missing packages manually over Window > Package Manager.

  

  • Click the plus and choose Add package from disk...

  

  • Find and open the package.json in the previously downloaded and unzipped package at ml-agents > com.unity.ml-agents.extensions.

  

Installing ML-Agents dependencies for Python

• In your Python env use
  • python -m pip install mlagents==0.26.0 to install the correct ML-Agents version, and
  • pip3 install torch -f https://download.pytorch.org/whl/torch_stable.html to download PyTorch.

Results

First (buggy) test

This is not the final result, just the output after the first incorrect test run and I love it

Final result: Youtube Video