Using an Artificial Neural Network for the Task of Road Following

This repository contains code to train a neural network for the task of road following. The network is provided with street images and outputs the corresponding steering wheel angles to drive a car on the street autonomously. The algorithm is based on the neural network described in the paper "ALVINN: An Autonomous Land Vehicle in a Neural Network" (1989) by Pomerleau. This project was created during the seminar "Algorithms for Imitation Learning" at the University of Stuttgart.

Installation

• Create a data/ folder in the root of the repository with the following data set files:
  • data/track_data_2.h5
  • data/camera/2016-06-08--11-46-01.h5
  • data/log/2016-06-08--11-46-01.h5
• Create a pip env: python3 -m venv env
• Activate the environment: source env/bin/activate
• Install dependencies: pip install -r requirements.txt
• Start the jupyter notebook: jupyter notebook
• Now you can view the notebook Seminar_ImitationLearning_FabianHauck.ipynb

OR use Docker to run the notebook with GPU support

• Copy the data sets in the respective folders as described above
• Build and run the Docker container with docker/start.sh from the repository root
• The notebook is now available under http://localhost:8888
• The access token is b0355f51bc6f93f72553da74bb6548801e64b2f9689ad96c
• Now you can view the notebook Seminar_ImitationLearning_FabianHauck.ipynb

Data Sets

1. The file track_data_2.h5 can be found in this repository.

2. The other files are part of the comma.ai driving data set.