Stefan Stojanovic

I am a 4th year PhD student at the Division of Decision and Control Systems (DCS)of the School of Electrical Engineering and Computer Science. I carry my research under the supervision of Prof. Alexandre Proutiere. Prior to that, I completed a MSc degree in Electrical Engineering and Information Technology from ETH Zurich with distinction.

Research

My primary research area is reinforcement learning (RL). Currently, I am interested in self-supervised RL, which involves learning to act without specific reward signals. For example, we recently introduced and studied the framework of switching successor measures, which enables hierarchical zero-shot reinforcement learning. On the theoretical side, we have recentlyinvestigated conditions under which structures essential for solving a task naturally emerge in self-supervised RL. Previously my focus has been on establishing theoretical guarantees for problems with a specific low-dimensional structure in settings such as model-based RL, contextual bandits, and model-freeRL.

If you are interested in my publications (Neurips, ICML, ALT...) please check my Google Scholar page. My research is supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP-AI), funded by the Knut and Alice Wallenberg Foundation.

Teaching & Supervision

I have been a teaching assistant for the Reinforcement Learning course during 2023-2025. In addition to the regular teaching activities, I designed and developed a new set of exercise sessions and educational materials for the course, available here.

I have also supervised bachelor's and master's theses related to my research area. Here are some examples of the work I've supervised:

• "Goal-Conditioned Reinforcement Learning for Autonomous Perceptive Earthworks in Particle-Based Soil Simulation" by Andrea Cucchietti (co-supervised with ETH Robotic Systems Lab)

• "Energy-Aware Adaptive Video Streaming Using Deep Reinforcement Learning" by Baptiste Boutaud de la Combe (co-supervised with InterDigital)

• “Adaptive Reinforcement Learning for Real-World Systems with Delays” by Iga Pawlak (co-supervised with ABB)

• “Frameskipping and Exploration Strategies for Deep Q-Networks” by Niklas Rolin and Vaka Soleyjardottir

• “Multi-Agent Control in Warehousing: A Deep Q-Network Approach” by Adam Fischer and Martin Wilen

If you are a bachelor's or master's student at KTH interested in working on Reinforcement Learning, please contact me via email.