Education

Postdoctoral research

Chalmers University of Technology: Oct. 2025-present

I am currently a postdoctoral researcher in the Division of Systems and Control at Chalmers University of Technology, where I focus on interpretable methods for robot-decision making and human-centered collaboration.

Hochschule Bonn-Rhein-Sieg (H-BRS): Feb. 2023-Sept. 2025

I was a postdoctoral researcher in the Institute for AI and Autonomous Systems at H-BRS, where my work focused on cognitive robotics, in particular robot learning (specifically using imitation learning) and robot adaptation.

PhD studies: Sept. 2017 - Jan. 2023

Completed dissertation (Dr. rer. nat.) at RWTH Aachen University with a magna cum laude (1.0)

Dissertation title: Skill generalisation and experience acquisition for predicting and avoiding robot execution failures

Advisors: Prof. Paul G. Plöger (H-BRS) and Prof. Gerhard Lakemeyer (RWTH Aachen University)

Master's studies: Oct. 2013 - Mar. 2016

M.Sc. in Autonomous Systems - Intelligent Robots at H-BRS

Thesis title: Improving the Manipulation Skills of Service Robots by Refining Action Representations

Thesis advisors: Prof. Paul G. Plöger, Prof. Kurt-Ulrich Witt, and Anastassia Kuestenmacher

Bachelor studies: Sept. 2009 - June 2013

B.Sc. in Information Science and Technology - Information Systems, Visualisation, Multimedia, and Animation: University of Information Science and Technology (UIST), Ohrid, North Macedonia

Research interests

From a general point of view, my research is concerned with robots that are one day supposed to support us in our everyday lives, for instance as household assistants. I am particularly interested in the following aspects of robotics:

Robot introspection, fault detection and diagnosis	Cognitive robotics
When I first joined the robotics field, I didn’t have robot faults on my radar, but, thanks to my eventual PhD advisor, I started working on different aspects of robot fault analysis quite early, so much that this has become my primary research passion. I believe that robots can only be useful in everyday applications if they are equipped with appropriate mechanisms for detecting their own failures and causally understanding those, potentially over a long time horizon - after all, we expect that any intelligent person can do this as well, so there is no reason to hold robots to a different standard if they are supposed to be integrated into our daily lives. My particular interest here is on execution failures, namely failures that a robot encounters while executing various skills, particularly in the context of learning-based skills.	I don’t know if we should necessarily aim to emulate human intelligence in robotics, particularly because our own intelligence is far from perfect (for instance, we tend to forget a lot). In any case, I think that autonomous robots cannot be flexible enough in complex environments without some cognitive mechanisms that we take for granted, such as the ability to memorise important aspects about the world, reason about their own actions as well as the actions of other actors in the world, or adapt their execution to a variety of contexts. From this point of view, cognitive robotics is, in my opinion, the right way to go so that we have useful autonomous robots in the long run.
Lifelong robot learning	Simulation-based robot learning and reasoning
Unless the application domain is narrow and well-defined, I think that it is very unrealistic to expect that a robot can be fully programmed or trained before being deployed; our environments are too complex and dynamic for that to be feasible. An everyday robot should thus be able to acquire new knowledge and new skills throughout its operation. I particularly believe that lifelong learning can only be successful if it is intertwined with, on the one hand, introspection (which, in turn, requires causal understanding about a robot's own actions) and, on the other hand, goal-directed communication with other agents (be it people or other robots), as it will otherwise be difficult to do lifelong learning efficiently. In this respect, I see introspection and learning as closely related to each other.	My specific interest here is on how a simulation can be used to allow a robot to try out its execution in scenarios that are otherwise time-consuming to set up in the real world. I think this is particularly important in the lifelong learning context, as a robot should be able to verify that the learning process is not causing performance degradation in some important aspects, such as safety.