Spring til indhold.
Forside

Room 0.2.13

Aalborg Universitet

PhD Defence by Jakob Haahr Taankvist

On Wednesday 9th of April 2025, Jakob Haahr Taankvist, will defend his thesis: "Safe, Optimal and Compact Strategies for Cyber Physical Systems".

Room 0.2.13

Selma Lagerløfs Vej 300
9220 Aalborg Øst

  • 09.04.2025 Kl. 13:00 - 16:30

  • English

  • On location

Room 0.2.13

Selma Lagerløfs Vej 300
9220 Aalborg Øst

09.04.2025 Kl. 13:00 - 16:30

English

On location

Aalborg Universitet

PhD Defence by Jakob Haahr Taankvist

On Wednesday 9th of April 2025, Jakob Haahr Taankvist, will defend his thesis: "Safe, Optimal and Compact Strategies for Cyber Physical Systems".

Room 0.2.13

Selma Lagerløfs Vej 300
9220 Aalborg Øst

  • 09.04.2025 Kl. 13:00 - 16:30

  • English

  • On location

Room 0.2.13

Selma Lagerløfs Vej 300
9220 Aalborg Øst

09.04.2025 Kl. 13:00 - 16:30

English

On location

All interested parties are welcome. After the defense Department of Computer Science will host a small reception in cluster 1, registration is not necessary. 

Abstract

Cyber physical systems are systems with two components, a controller and an environment. The two communicate via actuators and sensors. The controller is usually discrete, and the environment is typically dynamic. 

The job of the controller is to control the environment, which may behave in unpredictable ways. E.g. a traffic light aims to control the traffic, but the traffic may not always behave as expected.

Examples of cyber physical systems are plentiful and includes traffic lights, cars, vehicles in general, airbags, ovens, farms, factories, etc. Common for these are that if the controller goes wrong the consequences can be fatal, thus it is crucial that the controller is working as expected, without error.


Similarly it is also important that the systems are efficient, while staying safe. A traffic light could just give green to one direction for one minute, then to the next for one minute etc. This is easy to verify as safe. However, if there are no cars in a direction or if one direction has a much heavier load than another, this would be hugely wasteful.

Cyber physical systems, as mentioned have dynamic behavior in the environment, a particularly important dynamic feature is time. 


In this dissertation we discover ways to use machine learning techniques to learn strategies for cyber physical systems with time and other dynamic behavior. The strategies generated are not only safe, but also optimal. The main idea is to learn the strategy under a strategy which is synthesized to be safe, and to keep enforcing the safe strategy as the learned optimal strategy makes decisions. The techniques are implemented in Uppaal Stratego, making then easy to use and do experiments with.

The sizes of strategies are also explored. Some of the strategies generated by Uppaal Stratego are several gigabytes big, we show methods of reducing the sizes of the strategies, making it easier to deploy the strategies in real systems. We also explore various use cases, such as a cruise control system in a car, a floor heating system in a house, and traffic light systems on roads. We learn strategies which are significantly better than the strategies used in the cyber physical systems today.

Attendees

in the defence
PhD supervisors
  • Professor Kim Guldstrand Larsen, Aalborg University.
Co-supervisor
  • Associate Professor Manfred Jaeger, Aalborg University.
Moderator
  • Associate professor Martin Zimmermann, Aalborg University.
Assessment committee
  • Associate Professor Bent Thomsen (chairman), Aalborg University.
  • Professor Dr. Nils Jansen, Ruhr-Universität Bochum.
  • Professor Dr. Ernst-Rüdiger Olderog, Universität Oldenburg.