unity-ml-racing-track

Sample project using Unity ML-Agents in an autopilot racing car.

Project history

Check the project diary here.

Environment setup

1. Install Unity ML-Agents and dependencies (steps here).

ML-Agents related

2. Train the model

• Activate the virtual environment in the repo root folder
• Run mlagents-learn config/autopilot_config.yaml --run-id=Autopilot --force. After the script starts it will ask you to run the project in Unity.
• Press Play in Unity

3. Review training statistics

• Activate the virtual environment in the repo root folder
• Run tensorboard --logdir results --port 6006
• Open a browser window and navigate to localhost:6006

Steps into development

• ML-Agents getting started
• Create a learning environment
• ML Agents walkthrough

Lectures

ML-Agents

• How to read TensorBoard?
• How to read TensorBoard? II
• Configuration file in ML-Agents

Unity

• Car movement in Unity with Wheel Colliders
• ML Agents in car movement
• Autonomous car parking
• Ignore collitions while training?

Geometry

• Dot vector applications in games

ML in gaming

• NN usage in Colin McRae's Rally 2