On 25 January, Mahir Gulzar defended his doctoral thesis „Addressing Real-world Scenarios via Motion Prediction in Autonomous Driving“. To mark the occasion, Professor Jaak Vilo, Head of the Institute of Computer Science, interviewed Mahir about how his journey in the world of research has unfolded.
If you had to describe your PhD in one sentence today, what would it be?
A roller coaster ride of ups and downs with a lifelong experience.
How did you first discover the research topics that eventually led you to your PhD?
After my master’s degree, I started working as a scientific programmer in the autonomous driving lab of University of Tartu. We were developing core functionality of an autonomous vehicle from ground up. There were many potential research areas which emerged from my work in ADL, and I eventually chose one of them.
Was there a specific moment or problem that made you think: this is what I want to work on?
With ADL, I have had the opportunity to work on different modules of autonomy stack since my masters. Predicting behaviors of other traffic agents was something which I didn’t work on earlier and no one else in our research group was exploring this niche. My curiosity of doing something new and novel led me to pursue the PhD in behavioral prediction.
What alternatives did you seriously consider instead of a PhD?
I was kind of looking forward to joining the industry as a research engineer in autonomous driving. There wasn’t any special preference for industry or academia but since I was already working in the University of Tartu with some of the best colleagues and mentors, I preferred to commit long term here and it was totally worth it!
What first exposed you to AI, applied research, or domains like health?
Apart from many other exciting courses, I studied machine learning and deep learning during my master’s in informatics from the Institute of Computer Science. I was first introduced to AI there. The applied research followed up during my master thesis where my developed solution was deployed to a UGV (unmanned ground vehicle). Since then, I have been working on autonomy solutions with a blend of both academic and applied research.
Was this transition planned, or did it happen gradually?
I was originally a game developer. I loved working with simulations. For me the transition from game development to applied robotics was very natural in a sense we work with linear algebra, trigonometry and many other similar disciplines of mathematics and algorithmics in these two domains.
What turned out to be harder than you expected?
Merging theoretical and applied research is exceptionally difficult. Literature would show plenty of state-of-the-art solutions to a specific problem, but when one starts to employ any of them, you realize it’s not a walk in the park. Counterintuitively, sometimes less sophisticated solutions work exceptionally well compared to SOTA in literature.
What was unexpectedly easier?
This might be rare but my first publication was accepted relatively quickly without any revisions. This was unexpected since I thought that this is the first time I am publishing an academic paper and the reviewers won’t go easy on me.
How did real-world data challenge your theoretical assumptions?
Theory with simulations might give you promising results, but real-world data is the actual deal. You might come up with thousands of simulated scenarios to solve a specific problem, but real-world data literally gives you a harsh reality check.
When did you truly realize how messy and imperfect data can be?
This is very funny. I was testing a yielding solution by deploying it on our testing vehicle on a test-site. The solution was working perfectly in simulation. We replaced the simulation input by real world traffic agents on the test-site and realized that we weren’t even detecting the traffic agents early enough for the solution to work. It was analogous to optimize a solution pipeline while overlooking the fact that the input signal itself was too delayed for the system to respond in time.
What does it feel like to work on problems that affect real people?
In many ways, it’s highly motivating because you’re working on something that can directly impact people’s daily lives. At the same time, that motivation comes with a strong sense of responsibility. When dealing with safety-critical systems, it’s not enough for a solution to work in theory. It must be robust and fail-safe to be deployed in the real world.
How do you deal with the pressure of wanting to “get it right”?
There isn’t a single right answer. It often comes down to personal approach. In my case, I rely heavily on systematic validation. That means testing extensively i.e. starting in simulation, moving to controlled test sites, and then gradually expanding to geofenced real-world environments. Even small changes should be monitored closely, because in safety-critical systems, seemingly minor issues can have significant consequences.
What has been the most difficult moment of your PhD so far?
Writing the final thesis and its revisions. Personally, I was not fond of writing at all. Even with papers, I could develop a novel model and validate it with ease, but when it came to articulating the work, I found it much harder than the technical challenges themselves. Maybe it is like that for most people who transition from industry to academia?
How do you deal with rejection—papers, grants, feedback?
Paper rejection is part of the process, and over time you learn not to take it too personally. If you get a rejection just submit it somewhere else, don’t stress out too much. In fact, one of my papers was initially rejected and later accepted at a more competitive venue with only minor revisions. So, such experiences can be a blessing in disguise.
What part of your work do you quietly enjoy the most?
The part I quietly enjoyed the most was the working atmosphere at the Institute of Computer Science. I really value the company of my colleagues and fellow researchers. Whether it was exchanging ideas, troubleshooting problems together, or just having casual conversations. Those moments made the experience much more meaningful than just the work itself.
Is there a place or trip during your PhD that stands out in your memory?
Every year, the institute of computer science organizes a summer school (ESSCaSS). The venues are often selected by voting. The best time I spent during my PhD was summer school 2024 in Roosta Holiday Village. It was a very exceptional experience. I would even recommend every PhD student to participate in summer school where the venue is off campus at least once during their study.
Are you the first in your family to do a PhD?
Indeed, I am the first one in my family to obtain a PhD degree. It’s a big milestone for me, and after all the effort it takes, I believe it’s something me and my family are very proud of.
What was your first impression of Estonia?
I always say that Estonia is like a hidden gem. Many people don’t know about it, but once you get to live here. It gives you a fresh perspective on everyday life. Whether it’s the digital lifestyle, the calm environment, or the balanced work culture, everything feels simple yet very efficient.
