During your studies at the Faculty of Electrical Engineering at CTU, you focused on air traffic control and later robotics. How did you gradually get into research in the field of computer vision and working with satellite data?
During my studies at FEE ČVUT, I saw my future more in the field of research and development, which was confirmed during my thesis work. I enjoyed coming up with my own projects and then defending the results of my work at international conferences in front of equally enthusiastic colleagues, as well as the travel associated with it. I wanted to pursue a PhD and, to be honest, I didn't really think about working for a startup or other company. It's true that before starting my PhD, I successfully completed a few interviews with corporations such as Honeywell, but I did that more for the experience. Before completing my doctoral studies, I changed my field from aerospace to robotics, and since I enjoyed mentoring students, I stayed on at FEE as an assistant professor after returning from an internship in Zurich.
Today, you are Chief Scientist at SpaceKnow. How did you get involved with this company, and what attracted you to this opportunity?
SpaceKnow is a company founded by my friends Pavel Machálek and Jerry Javornický, both of whom are classmates from the English College in Prague, where we all studied. When they approached me with an offer to join them and presented their vision, I knew almost immediately that this was an opportunity I definitely didn't want to miss. I was attracted by the idea of getting to know Silicon Valley and being part of the creation of a company that would be a pioneer in a newly emerging market from the very beginning. I knew that this would be a job I could put my heart into without hesitation, because it wasn't some foreign corporation. In addition, around 2014, the first large neural networks began to gain ground in the field of computer vision, and the use of deep learning for satellite data to create a useful product was an exploration I threw myself into with great enthusiasm. After my first stay in San Francisco and a successful meeting with investors, I knew that SpaceKnow would not only be a job for me, but also a hobby. We founded R&D at SpaceKnow together with my doctoral student—when you have a great partner with whom you enjoy working, such a fundamental decision is very easy to make.
Today, you lead teams that design and deploy cloud solutions based on computer vision and deep learning. What is the biggest professional challenge in your work at the moment?
My role in the company's management means taking care not only of the direction of technology and product development, but also of people development, both of which are very closely intertwined. Nowadays, perhaps the biggest challenge is keeping pace with the dynamic development of AI technologies, realizing where these technologies can be reliably deployed, what their limitations are, and what their business benefits are. And I don't just mean deployment in products, but also use in the daily activities of everyone in the company, from administrative tasks, through the creation of presentations and marketing materials, to data processing, dataset management, and the development of new algorithms.
SpaceKnow is based on the idea of transforming huge volumes of satellite data into understandable and commercially usable information. What do you think is the greatest added value of this transformation?
Having timely, key, and ideally complete information so that people can make important decisions correctly is invaluable, and this applies across industries and across different markets. The price of satellite data still plays an important role, so it is not possible to monitor the world all the time and everywhere. And this is another thing we are optimizing in our product. Not only the right information at the right time, but also at an acceptable cost.
Satellite data will one day be a common commodity
The price of satellite images is falling, but the value of the information is growing. How do you think the satellite analytics market will evolve over the next ten years?
When we started in our field, the commercial use of satellite imagery was completely uncharted territory, and the entire market was just emerging. In fact, this market is still developing very dynamically today. Previously, only large corporations owned satellites and supplied data primarily to governments or government agencies. However, with the arrival of smaller companies and startups operating their own satellite constellations, falling prices and better data availability have opened up the market to companies like ours.
Unfortunately, the price of images is still a limiting factor today. The average customer cannot yet afford daily monitoring of large areas in high resolution. But the trend is clear: prices are gradually falling and the amount of data is growing. In addition to electro-optical sensors, satellites providing hyperspectral or radar data, which are not affected by cloud cover, for example, are also on the rise.
The growing number of satellites and various sensors in orbit makes our work incredibly interesting. New opportunities are constantly opening up for us, and what was true in our field six months ago is now long gone. I believe that in 10 years, satellite data will be a commodity and information from orbit will be integrated naturally and routinely with other data sources, without anyone even thinking about the fact that it comes from a satellite.
SpaceKnow has received significant investment and has even been featured in the Financial Times. What do you think determines whether a technology startup can succeed globally?
To be honest, I can't really imagine a technology startup succeeding today without thinking globally, because the competition often thinks globally, and the global market opens up more opportunities and space to realize one's potential. SpaceKnow is an American company that was founded in Silicon Valley, and our mindset has been global from the beginning. What does that mean? It means that we developed a solution that includes, for example, computer vision algorithms that would work on images anywhere in the world for various applications.
As for investments, I have a slightly different opinion than most people on how investments are glorified. When a company receives a large investment, it is certainly a sign that it has some market value. But that's only part of the story; there's also the other side of the coin, which is usually not mentioned. For example: what did the investment cost the company, what share do investors now control, and what obligations or debts does the investment entail for the company? An investor should primarily be a good partner who will help the company, and for me, choosing the right investor has a strong influence on whether a startup can succeed globally.
Academic research vs. startups, or why companies sometimes don't have time to publish
You also worked as an assistant professor at the Department of Cybernetics at the Faculty of Electrical Engineering of the Czech Technical University in Prague and are a co-founder of the CRAS robotics group. What do you think are the biggest differences between research at a university and research in a startup?
This is a very good question, and it is also something I have thought about for a long time, especially when I was working both as an assistant professor at the Department of Cybernetics and at SpaceKnow. The study of the state of knowledge, research methodology, design, implementation, and evaluation of experiments, as well as careful and objective evaluation of the quality of the output, are all the same in both cases. However, the main difference between research at a university and research in a startup is who evaluates you in the end and what criteria you optimize in your work. In academic research, your output is usually an article or conference paper that is evaluated by the professional community, which is primarily interested in the objective contribution to the state of knowledge and innovation. In contrast, in a technology startup, the most important criterion for successful research is innovation that increases the added value of the product, which is assessed not by researchers but by paying customers.
What role do publishing, openness of results, and participation in conferences (CVPR, ICCV, ICRA, etc.) play today in the technology company environment?
As a long-time reviewer of contributions to CVPR and ICCV conferences, this is an area that I personally continue to follow very actively. It is an important source of information for me, which can be worked with very quickly and efficiently today using tools such as Google Scholar Labs. Previously, I was more interested in the main idea and innovation in scientific articles, but today I focus more on what datasets were used, how replicable the experiments are, and whether open code is available under a suitable license. I must admit that successfully publishing a high-quality paper at one of these top conferences is very time-consuming, and unfortunately, we really don't have the time or capacity for that at SpaceKnow. But who knows, maybe as we grow, there will be more time to write conference papers, because that's definitely a good calling card for a technology company.
My experience from my doctoral studies in Zurich led me to robotics
During your doctorate, you worked at the University of Zurich, for example, and then worked on robotics in the field of Urban Search & Rescue. How did your experience abroad shape you?
As I mentioned, SpaceKnow was founded by my high school classmates, so I also traveled to Zurich to Prof. Rolf Pfeifer's artificial intelligence laboratory to join my high school classmate Matěj Hoffmann, who now leads his own research group at the Department of Cybernetics at the Faculty of Electrical Engineering. I dare say that this experience abroad was absolutely crucial and decisive for me. I definitely recommend that everyone travel to a top university if they have the opportunity. So what was decisive for me? Not only did I manage to complete my dissertation in record time thanks to the robotics experiments in Zurich, but the experience I gained there led me to change my field of study. I moved completely from aeronautics to robotics, and in addition to signal processing algorithms and data fusion, I became enthusiastic about machine learning and computer vision. As a result of my experience abroad, Prof. Vašek Hlaváč noticed me and, after completing my doctorate, accepted me into his top research group, where I was given the unique opportunity to work on international robotics projects in the field of Urban Search & Rescue (USAR). Under the leadership of Prof. Tomáš Svoboda, our group then formed the aforementioned CRAS, which represented CTU brilliantly at the DARPA SubT Challenge. What fascinated me most about these USAR projects was the emphasis on applicability and benefits for the end user, who were elite firefighters and rescue workers from countries such as Germany and Italy, who tested our technology in the most demanding conditions. Writing a scientific article about a new algorithm for robot localization and mapping is one thing, but it's another thing to make the algorithm work reliably in accident or natural disaster conditions, where the robot is traveling through a smoke-filled tunnel, cars are burning, sensors are constantly losing data, and the robot has to reliably detect and locate key objects or potential victims. To be honest, these projects really opened my eyes, and when I was considering working for SpaceKnow, it was the combination of research and practice that appealed to me the most. I didn't want to do research that would only end up in an article.
What did your doctorate give you the most—was it expertise, a way of thinking, contacts, or something else?
Doctoral studies taught me a lot—from specific hard skills to soft skills related primarily to teaching and mentoring students or presenting at large scientific conferences. Thanks to my doctoral studies, I had the opportunity and time to immerse myself mathematically in a topic that interested and entertained me, to such an extent that I was able to discuss it with world leaders in the field and learn from them. It also taught me humility, because when you want to compete with the best, you quickly realize how much you still have to catch up on and that you will probably never succeed. But if I had to name one thing, it would definitely be the ability to create my own system for learning and improving, which I still use in my work today.
What kind of graduates are technology companies interested in?
Do you think that FEE CTU and technical universities in general could better support the creation of technology startups with global ambitions? If so, how?
This is a rather complex topic, which is closely related to the experience and ambitions of the teachers themselves. The more experts there are at FEE CTU who have, for example, successfully built a company, the better they will be able to guide students in this regard, pass on their experience, and motivate them. Personally, I am thrilled that Prof. Pěchouček was elected rector, and I really like the team he has put together. I think this is an absolutely crucial step for the future of CTU.
What type of graduates are companies like SpaceKnow really looking for today?
In terms of knowledge and skills, I am primarily familiar with graduates of study programs such as Cybernetics and Robotics or Open Informatics. There is basically nothing to criticize there, and if these programs keep pace with current technological trends, which I believe they will, it will be absolutely perfect. In recent graduates, we are primarily looking for a desire to constantly learn new things, work in a team, and generally have the courage to be active and think critically about how everyday work contributes to the product itself and to customers.
FEE was a completely different school in 2000
Why did you choose FEL for your studies?
I would probably start by saying that I graduated in 2000, and at that time, FEE was a completely different school than how I perceive the faculty today, so my decision is probably not very transferable to the present day. In any case, at that time, FEE had a reputation as a faculty closely linked to industry, and I knew from my surroundings that companies were very interested in FEL graduates.
How do you remember your studies?
I studied at FEE before the program was divided into bachelor's and master's degrees and before FIT was established, for example. The first years were very general and served more as a demanding filter for students—the so-called comprehensive exam was a nightmare at the time, which I fortunately avoided thanks to my grades. Fortunately, today's students don't know anything about that. I really enjoyed studying aviation because it was taught by enthusiasts and experts with practical experience; it was also very interdisciplinary—from aerodynamics and flight mechanics, through control theory and autopilots, sensors and signal processing, to navigation, satellite and radio communication, and flying on trainers and simulators. The icing on the cake was taking a course at the Military University of Defence, which was basically the basics of medicine for pilots. We dealt with emergency recorders and aviation accidents, including stays in a hyperbaric chamber and a flight illusion simulator. Just as we soldered and created our own hardware, we also programmed in C and C++ and designed autopilots in MATLAB. The program was incredibly broad in scope, and everyone found what they enjoyed most.
Which teachers or moments from your studies do you remember most fondly?
As I indicated in my previous answer, there were so many interesting moments and experiences during my studies that it would take a very long time to list them all here. However, I have the most memories of my supervisors, Doc. Draxler and Doc. Roháč, with whom I also had the amazing opportunity to travel around the US and attend challenging courses led by top experts in the field, for example from Raytheon. This was mainly because I was able to get involved in ongoing research projects while working on my thesis.
Can you recall any stories from your studies that still make you laugh today?
Back then, about 25 years ago, the mathematical logic exam was a nightmare, where passing the written test did not guarantee success in the oral part, which was decisive. The oral part was very difficult, usually lasting up to 15 minutes, and few people passed it on their first attempt. My classmates were all the more shocked when I spent more than an hour and a half on the oral exam and passed it with a grade of 2 on my first attempt. The truth was, however, that my examiner fell asleep while correcting my written exam, and I just sat there patiently waiting for him to wake up, afraid to say anything. Today, I can laugh about that situation, but at the time, it was perhaps my longest and most stressful exam ever.
Artificial intelligence, robotics, and data science are among the most dynamic fields today. What would you recommend to current FEE students who want to succeed in the world of technological innovation?
Nowadays, it is easy to get the impression that we have the answer to any question at our fingertips, very quickly and reliably. Thanks to the availability of AI tools, it is easy to give up on any effort to continue learning and improving. And that would be a terrible mistake. Anyone who wants to succeed must be prepared for continuous lifelong learning—and school should help them with that. Innovating also means not being afraid to go a step further, experiment, and ask for feedback, to do what is necessary and even a little bit more.
If you had to sum up in one sentence what your doctorate and studies at FEE have given you in your professional life, what would it be?
Studying at FEE gave me a strong foundation across various disciplines that I can still build on today, and it helps me learn new things and technologies.
