Autonomous Parking Simulation

This project is about an autonomous parking simulation program with Unity ML-Agent.

Program Download Link

Installation

• Clone the git repository

  git clone https://github.com/hae-sung-oh/AutoParkSimulation.git
  

• Open the project with Unity Hub

  • Unity Hub -> Open -> {Select Cloned Folder} -> Open

  

  

• Add ML-Agent package in the project

  • Window -> Package Manager -> Unity Registry -> ML Agent Install

  

  

  

  

• Install ML-Agent

  • For Ubuntu: Instructions

    conda create -n autopark python=3.10
    conda activate autopark
    pip install tensorflow-gpu
    python -m pip install mlagents==0.30.0
    

  • For Windows: Instructions

Scenes

There are 3 Unity scenes in the /Assets/Scenes folder for corresponding parking situations. The Master scene is for the .exe program.

• Front

• Parallel

• Diagonal

• Master

Training

You can train the agent with Reinforcement Learning by running commands below and hitting the 'Play' button in the Unity Editor.

Note: Don't forget to activate the virtual environment that made in the installation part.

• Default command

  mlagents-learn trainer_config.yaml
  

• Specify RUN_ID (You can label the custom training ID)

  mlagents-learn trainer_config.yaml --run-id={RUN_ID}
  

• Resume existing training with RUN_ID

  mlagents-learn trainer_config.yaml --run-id={RUN_ID} --resume
  

• Run existing training forcefully and overwrite it

  mlagents-learn trainer_config.yaml --run-id={RUN_ID} --force
  

Test

After the training, you can test the result with the following methods.

• Check created Neural Network Model (Behavior Parameter)

  • If you put the NN file (.onnx) created by the training into the agent object, the agent will show the trained behavior.
  • Hierarchy Tab -> Select AgentPrefab Object -> Inspector Tab -> Behavior Parameters -> Model -> Select .onnx File
  

  

• Heuristic Mode (Manual Control)

  • If you change the control mode of the agent to 'Heuristic', you can control the agent manually with W(acceleration), A(steer left), S(back acceleration), D(steer right), and Spacebar(brake) keys.
  • Hierarchy Tab -> Select AgentPrefab Object -> Inspector Tab -> Behavior Parameters -> Behavior Type -> Heuristic Only