facebook tracking

Master Thesis - Driving scenario generation using deep learning

Scroll to content



An important challenge to release autonomous vehicles to public roads is the safety verification. In order to be able to argue for safety of autonomous driving, a verification machinery will have to combine several evaluation tools such as road testing, scenario-based testing using simulation environment, etc. In the scenario-based testing approach, the aim is to generate a set of scenarios in the simulation environment, or so-called test cases, which can expose the AD vehicle to safety critical situations. The scenario generation process usually starts by defining a set of rules to break down the traffic dynamics into a well-defined set of fundamental actions of the road users which can be identified from data collected during road testing. The extracted scenarios are then statistically modeled in order to be able to fuzz and/or extrapolate them in the simulation environment to find safety critical situations. However, such approach for scenario generation is predominantly limited to set of rules defined by the expert’s knowledge.



Project Description 

The aim of this master thesis is to automate the scenario generation procedure using deep learning methods. The students will be provided with a database of real-world driving scenarios from about hundreds of thousand kilometers of road testing. State-of-the-art deep learning techniques such as Variational Autoencoder (VAE) or Generative Adversarial Networks (GANs) have shown promising results in learning and to extracting basic driving features (scenarios) from road testing. The goal is to extend this approach to more complex scenarios. The resulting models can later be adopted to generate synthetic but physically feasible driving scenarios in simulation environment to evaluate the performance of the autonomous driving software.


We are looking for students with the following skills: * Creative mindset

  • A strong background in mathematics, statistical modelling, machine
  • learning and deep learning
  • Solid programming skills (Matlab, Python)
  • A passion for data analysis and programming

Further information


Final application date: 30, Nov, 2021 .
Please send in individual applications with CV, motivational letter and grade transcripts.
If you wish to partner with someone, simply note that in your application.
Planned start: 10 Jan 2022, with some flexibility.
Duration: 30 ECTS
For questions regarding the project, please contact: Majid K. Vakilzadeh (majid.vakilzadeh@zenseact.com) Andreas Nord (andreas.nord@zenseact.com)


Additional information

  • Remote status

    Flexible remote

Or, know someone who would be a perfect fit? Let them know!

Gothenburg, Sweden

Lindholmspiren 2
417 56 Gothenburg, Sweden Directions View page


Career site by Teamtailor