Machine Learning and Simulation for Future Autonomous Mobility

Level: MA/(BA)

In the context of future mobility a testbed for development applications is necessary. Different problems such as intermodal vehicle routing, parking or speed recommendations systems have traffic as a common factor. A gym-like environment based on a popular traffic macro-simulation should be extended so that experiments with different methods and problems can be conducted. For the method reinforcement learning or other optimization approaches are possible.

Tasks:

The focus of the proposed thesis can lay in one or multiple topics such as:

  • Multi-agent reinforcement learning methods ( eg. scheduling/routing of delivery robots)
  • Regression methods ( eg. traffic or parking demand forecasting) 
  • Simulation and modeling of traffic based on SUMO and other co-simulations

Requirements:

  • Very good understanding of machine learning through experience and coursework. (at least 2 good completed courses and some practical experience with e.g. Keras/Tensorflow, pytorch, FLAX …) 
  • Practical experience in co-working on Python/C++ projects. Be able to write good documentation and using tests
  • The motivation to contribute to a scientific publication 

References:

RL Intro Book: Sutton, Richard S., and Andrew G. Barto. Reinforcement learning: An introduction. MIT press, 2018.

RL Intro Lecture: https://www.youtube.com/watch?v=2pWv7GOvuf0&list=PLqYmG7hTraZDM-OYHWgPebj2MfCFzFObQ

Traffic Marco-Simulation: https://www.eclipse.org/sumo/

Autonomous Driving Simulation: https://carla.org/

Traffic Forecasting Challenge: https://www.iarai.ac.at/traffic4cast/

Vehicle Routing Challenge: https://euro-neurips-vrp-2022.challenges.ortec.com/

Interested?

Send your application with a convincing set of documents to

Patrick Grzywok

patrick.grzywok@gt-arc.com