Autonomous lane following in a simulated environment


The goal of this project will be to implement an autonomous system to keep a car on a lane in a simulated environment based on visual data. The project consists of three main components:

  1. Designing model routes in a simulated environment.
  2. Training street and lane detection
  3. Implementing/assembling a control to adhere to the lane The basis for this project will be an open source self-driving car simulation, a street detection CNN, and a control system.


Part one: Designing basic routes, defining interfaces for interaction with the lane detection and a control system.

Part two: Automated generation of training and test scenarios. A first step here is to build basic street blocks which may be combined with routes. These blocks should then be used to visualize predefined routes in the OpenDrive format.

Street detection:

After familiarization with the neural network, the network will be retrained to work on the simulated data. Based on the performance on this data, the network will have to be adapted and a post-processing added.


After a first PID control is implemented and tested, an advanced control system, for example a fuzzy control will be implemented.

For detailed information please contact Dr. Stefan Held

For the students accepted to this project, there will be a SCRUM course on October 10th and 11th taking place at the ITK office in Martinsried to prepare the participants for the project. If you are not able to take this course then we will not consider your application. If you apply to this project then write in your application that you will be available to take the course.


The results of this project were summarized in a final presentation and explained in detail in the final documentation