Torcs Machine Learning Driver

The machine learning driver is developed to control a race car on the Torcs simulation platform. In this repository, you can find the training algorithm in build_pipeline/build_pipeline.py. The data used for this is located in the csv folder. The details of how this data was collected are explained in the README inside the data_pipeline folder.

A working final version of the model can be found in the models/final_model/ folder.

Contents

• Build_pipeline/ Contains the script used to train the model.
• csv/ Contains the training data used for the model.
• driver/ Contains the program used to run the model on the Torcs server.
• models/ Contains other models trained outside the latest model.
• models/final_model/: Contains the most recent model.
• notebooks/ Contains Jupyter notebooks used during the development of the model.
• Dockerfile: Contains the instructions for building the Docker image.
• NeuralNet.py: Contains classes and functions used by multiple scripts in the project.
• NeuralNetSettings: Contains the settings used for the model, including structure and the data fields used.

Contributors

• Raekwon Killop
• Raymond Blok
• Arthur Struik

Using the Model

You can use the model by either building or pulling the Docker container. Below are the instructions for both methods.

Method 1: Docker Build

1. Clone this repository to your local machine.

2. Navigate to the repository directory.

3. Run the following command to build the Docker container:

   docker build -t mldriver .

Method 2: Docker Pull

You can also directly pull the prepared Docker image from Docker Hub.

1. Run the following command to pull the Docker image:

   docker pull raymondblok/mldriver

Running the Container

Once you have built or pulled the Docker container, you can run it with:

docker run -it [raymondblok/]mldriver --hostname "torcs server ipaddress" --port "torcs server port"