Model Predictive Control & Machine Learning for Trajectory following of Autonomous Vehicles

Lehrstuhl für Fahrzeugtechnik
Semesterarbeit / Masterarbeit /
experimentell / theoretisch /  

At the Chair of Automotive Technology, we are developing a full open-source software stack to operate autonomous vehicles not only in a motorsport context but also on public roads. The AV 2.0 software project represents a scalable and modular platform. The main modules are Perception, Planning, and Control. This enables us to quickly evaluate novel concepts and to operate our different experimental vehicle types with the same software structure, e.g. Multivan, race car, or truck.

We, at TUM Control, are working on a groundbreaking learning-based Model Predictive Controller for trajectory following of autonomous vehicles in different environments. With our algorithms, we are striving for decreasing the manual MPC Design effort by automating the process with machine learning techniques. Adapting the MPC parameters online promises an increased closed-loop performance when faced with changes. Our research focuses also on ensuring a robust control of the vehicle dynamics against uncertain parameters and external disturbances.  

At the TUM AV Control team, you’ll get the possibility to develop your skills and personality in very exciting and big-impact projects by working on developing the first-ever L5-capable MPC controller. For that, we develop algorithms not only in a software simulation but also in a hardware simulation as well as deploying them on real vehicles.

In an exchange with other team members, you have the opportunity to apply the theories you’ve learned in your studies as well as getting to know new algorithms and expand your programming skills.  At the end of your semester thesis, you have the chance to become an expert in your field and get ready for professional life.   

Do you want to put your own mark on the future of autonomous vehicles with your ideas and concepts? You will be part of a team conducting state-of-the-art research in the trajectory control of autonomous vehicles.

If you are interested in a student research project, feel free to send me an initiative application and we will arrange a call to discuss the topics. Just send me an e-mail with a short motivation, curriculum vitae, and a recent transcript of records.

Currently, one of the following topics can be addressed in your thesis:

  • Model Predictive Control (MPC) problem formulation for trajectory following of autonomous vehicles:
    • Stochastic MPC
    • Tube MPC
    • Nonlinear MPC
  • Learning-based & adaptive MPC:
    • MPC Prediction Models:
      • Online learning/ online parameter identification for an adaptive dynamics model
      • Enhancing MPC prediction- / uncertainty description models with online available data
      • Approximation of a non-linear vehicle dynamics model for real-time feasibility, e.g. using neural networks 
    • MPC cost function:
      • Deep Reinforcement Learning of MPC parameters
      • Bayesian Optimization of MPC cost function weights
  • Robust Vehicle Dynamics Control:
    • Investigation and modeling of uncertainties and disturbance in the vehicle context (e.g. crosswind)
    • Sensitivity analysis of the uncertainties and disturbances on the system state.

The student research project/ thesis will handle upon agreement a single or multiple bullet point work packages.   

  • Motivation to familiarize yourself with new topics and to try new ideas
  • Ideally previous theoretical knowledge in Model Predictive Control
  • Ideally previous experience with Python/C++, Git, ROS2
FTM Studienarbeit, FTM AV, FTM Zarrouki, FTM Informatik, FTM AV Safe Operation
Möglicher Beginn
Baha Zarrouki, M.Sc.
Raum: MW3505
Tel.: +49 (89) 289 - 10498