Appointment with... Niklas Kochdumper

Prof. Niklas Kochdumper completed his Bachelor's degree in Mechanical Engineering and his Master's degree in Robotics, Cognition and Intelligence at the Technical University of Munich (TUM). He earned his doctorate in the field of "Formal Verification" in the research group of Prof. Matthias Althoff in Garching, where he focused extensively on methods for system safety assurance at the Chair of Robotics, Artificial Intelligence and Real-Time Systems.

He gained research experience as a postdoctoral researcher at Stony Brook University, NY/USA, and subsequently at the Institut de Recherche en Informatique Fondamentale (IRIF) at the Université Paris-Cité. His research interests include the formal verification of continuous and hybrid systems, reachability analysis, computational geometry, controller synthesis, and the verification of neural networks. From March 2025, he served as a junior professor at Technische Hochschule Ingolstadt (THI). On 1 March 2026, he was appointed Professor of Spacecraft Control at the TUM School of Engineering and Design (TUM ED). In the interview, he talks about his path into academia, his research plans, and why reproducibility remains central in the age of Artificial Intelligence.

ED: How did you become who you are today? 

Prof. Niklas Kochdumper: From the very beginning of my studies, I knew I wanted to go in a technical direction, though I wasn't yet sure exactly which one. Automotive, aerospace, robotics — many fields interested me. That's why I deliberately chose to study Mechanical Engineering at TUM for my Bachelor's, in order to first build a broad foundation in engineering. Specialization was to follow in the Master's.

When choosing my Master's program, two areas fascinated me above all: aerospace and robotics and control engineering. Ultimately, my path led me through computer science and electrical engineering toward research on safe, intelligent systems. With my current professorship, a circle is now closing in a sense: I am working in the space sector and, at the same time, have returned to my alma mater.

What is your first research project at TUM?

A central focus of my research will be reinforcement learning with safety guarantees. AI-based methods have enormous potential, including for space systems. Studies show, for example, that reinforcement learning can be used to deploy resources such as fuel more efficiently — for instance, when collecting space debris.

However, there is a fundamental problem: it is often not transparent what an AI system has actually learned. This leads to significant safety concerns, particularly in critical applications such as spaceflight, where poor decisions can have serious consequences.

This is precisely where I come in. My goal is to build a bridge between powerful AI and reliable safety. To this end, I am developing formal safety concepts, such as so-called safety filters. These review the AI's decisions and intervene with corrections when an action is potentially unsafe. In the long term, such approaches should help ensure that AI-based methods can be deployed reliably even in safety-critical applications.

What change do you hope for in the future?

One of the greatest challenges, particularly in the context of AI, is the reproducibility of scientific results. The more complex the methods become, the more difficult it is to independently retrace and confirm results. Yet this is precisely one of the central pillars of scientific work.

I therefore hope that the trend toward open and transparent research will continue to grow. This includes researchers not only publishing their results, but also making the underlying code and necessary scripts available. In recent years, there has already been meaningful progress in this area: many conferences now explicitly require the submission of program code, and reproducible results are frequently highlighted with a special designation.

This is an important step in the right direction. Looking ahead, I could envision such requirements becoming even more binding. For scientific progress can only be secured in the long term if research results remain traceable and verifiable — including, and especially, in the age of Artificial Intelligence.