From FI to Boston and Stockholm: connection between theory and practice in formal methods for robotics

Klára Petrovičová for fi.muni.cz

How does a love of mathematics lead to research in robotics? Jana Tůmová has become an expert in formal methods for robotics thanks to her active involvement in the lab at Faculty of Informatics (FI) and international internships. In an interview with FI, she explains how she was attracted to the structures and principles of theoretical computer science and how this knowledge has been applied to research on parallel systems and distributed processes. Later, Jana Tůmová focused on multi-robot systems and safety, in collaboration with researchers at MIT and in Stockholm. It was her collaboration with leading researchers abroad and work with different types of robots that led her to lead her own research group.

What was your path to Computer Science and FI?

I always liked mathematics, which I also wanted to study. My surroundings discouraged me from doing so and recommended me to go study computer science, so I would not worry about getting a job. So I went into Computer Science, but I still ended up taking courses related to mathematics, mainly those in theoretical Computer Science.

What interested you most about theoretical Computer Science at the time?

Apart from the fact that it was very close to mathematics, I really enjoyed its structures, principles, and proofs. I liked that it all worked and that there was a clear reason for it.

Were you immediately determined to go for a Ph.D.?

No. I rather thought I was going to go into industry somewhere. But I was fortunate that we were a small group who met in lectures and tutorials, and some of the group members said they were going to go to work in a lab. I loved the idea of getting a sniff of research and till this day I consider that opportunity a big plus for FI. Undergraduates and Masters students have the opportunity to engage in research work in the labs. This then makes the decision whether to go for a Ph.D. or not much easier. When I fell in love with research in this process, I knew I wanted to go for a Ph.D..

What were you researching in the lab at the time?

I was in the lab of Luboš Brim, Ivana Černá, and Jiří Barnat and we were researching parallel and distributed systems. At that time it was mainly model checking Markov chains and Markov decision processes with linear temporal properties. At that time it was quite new because they were working mainly with branching-time logic, but we went into the linear one.

Was your research then put in practice?

The whole introduction to practice was basically the reason I went abroad, because we were approached by Professor Calin Belta, who was doing robotics and control theory, and he was looking for someone who knew LTL and Markov decision processes. At the time he was looking for someone who was working in those areas and whenever he typed them into Google, our group's work popped up. So he eventually emailed us, that is, Jiří Barnat, and said he would like to come and visit and talk about these topics in the context of robotics. During his visit to our lab, he asked me if I wanted to come to Boston for an internship. I didn't have to think about it and agreed right away. That's when everything changed.

The topics we were exploring at FI were related to robotics rather indirectly, but a lot of the related techniques that evolved from there were a good fit.

Did you go to Boston during your Ph.D.?

Yes, I went during my freshman year for 4 months. The collaboration was successful, so we extended it for about 2 more months and then applied for a grant at FI that paid for my stay in Boston for another six months. During that time, Calin Belta took me to MIT, where we started another collaboration, and so I was at MIT for another 4 months. The collaboration went well too, and so they took me with them to Singapore. There we were part of a smaller project to apply our research on an autonomous car, which at that time in 2013 was basically a golf cart that had autonomous software. We were able to get formal methods on it, which was a very compelling moment for me to stay in the academy. Despite the fact that we're doing theoretically interesting things, we can get feedback from application to concrete things.

Didn't you then want to stay abroad for your Ph.D.?

No, I wanted to go back to the Czech Republic, which had several reasons. One was family ties, because my boyfriend stayed in the Czech Republic. Another reason was that I had support here for things I needed, such as technical knowledge in the field, which helped me bring interesting results to other areas. This faculty gave me great knowledge and a way of thinking about things that I still use today, I wouldn't have had that opportunity elsewhere. Here I was at home and went on trips that gave me a lot of contacts that I still use now, whereas if I had gone abroad for my Ph.D. I would have been in a lab and not had the opportunity to try as many things as I did here from my position.

What was your journey like after your Ph.D.?

I had a vision of staying here for my postgraduate degree, but then I was approached by a person, professor Dimos Dimarogonas, from Stockholm, who was looking for someone with my expertise. What I had done in the context of robotic systems and planning for one robot, he wanted to use for multiple robots. So my knowledge could be used, but at the same time there was an aspect of learning something new — multi-robot systems.

In the beginning I had a vision of going to Stockholm for a year at most and then getting a job as an assistant professor somewhere. But that didn't quite work out (laughs). We wanted to stay in Europe where the system is set up so that you need postgraduate experience for at least 2–3 years before you are ready to lead your own research group. Although I didn't think so at the time, I know now that this postdoctoral experience is really important. One can look around a bit more because one doesn't feel the pressure of a dissertation, and at the same time learn new things that one doesn't learn during a Ph.D., such as a grant writing, supervising students and projects, writing project reports, and giving talks. One then starts to realize what an assistant professor position is going to be all about. That is exactly how I proceeded, only in a different faculty and department. At that time, I left my graduate supervisor and started building my own group.

Were you teaching any courses at that time?

Postgraduate students are not usually given major lectures because there is a danger that they will leave and there would be no one to take over, so I just gave a few lectures in different subjects. When I started as an assistant professor, I got a big course in Artificial Intelligence right away. In the beginning, we had 350 students, and a year later we had 450, and the next year we had 550 because the topic of Artificial Intelligence started to be interesting to everybody and the course started to grow.

What topics did your group focus on?

Primarily formal methods for robotics, so I still benefit from the knowledge I picked up here and the experience I gained abroad and during graduate school. But I have found my own angle on the subject, which I have tried to use to set myself apart from other perspectives. What we mainly focus on is risk aware planning using formal methods, where we take into account that the world is very uncertain and that it is hard to have 100% guarantees. The second topic is the comparison between provable safety and perceived safety, i.e. the safety of the system that we can prove using formal methods and, in addition, the safety of the system that is perceived by the people who interact with the system. This work was done in collaboration with my colleague Iolanda Leite, who works in the field of social robotics and is working on the perceived safety part, we are working on provable safety. We sat down together one day and talked about safety and found that we had a completely different perspective on it. We agreed that it would be good to be able to translate what we can prove through the system so that a person can trust it to exactly the level that they should, no more and no less.

What kind of robots do you work with, for example?

My department involves robotics, perception in the sense of seeing and semantically interpreting what the robot sees, and learning. We have robots that "live" underwater, robots that fly, humanoids that don't look quite like humans, but maybe ones that sit in an airport and tell you which way to go. We have little mobile robots that can drive in all directions, then we have little cars that can't turn their wheels in all directions. We're building a space robotics lab, so we have a platform where robots will work in microgravity and float on the surface of the air. At the heart of our work are robots that physically interact with the outside world. We have to transfer all the intelligence to a physical system that has to grab something, carry something somewhere. Everything always works in our simulations, but then the robot doesn't do what we want it to do (laughs).

Do you build your own robots?

We don't build them ourselves, but often the robot needs to be adapted. For example, to add a sensor or an actuator. We make these small adjustments, but we usually buy robots as a complete system, especially those that interact with humans. We want to rely on those and we want them to look good. We don't do state of the art research on robotic capabilities on these robots, but they typically have physical limitations and we customize them depending on where we want to deploy them.

Do you offer any courses where students learn with robots?

We have a whole master's program called Systems, Control and Robotics, where the first course is Introduction to Robotics. Students learn about how robotics and the whole system work. They also have required project courses that require working with a robot. For example, a colleague of mine hands out a robot called Crazyflie, which is a tiny drone, and has them play with them, while students learn computer vision, perception, and control.

Are these students learning formal methods as well?

Yes, they learn those in my course that I put on there (laughs). Of course I try to teach about the science that we do. Formal methods in robotics are not a mainstream topic, the important thing is to get the robot moving and the guarantee quite often comes later. We try to think about these formal methods from the beginning when we design the system.

What other courses do you teach?

Artificial Intelligence, Planning and Control of Robotic Systems, I supervise undergraduate and graduate theses, and a course called The Sustainable Systems and Control Engineer, where students from the program meet and discuss different topics so that they grow up to be conscious engineers who know that there are things like ethics, gender issues, and sexual harassment. As part of this course, students also interview alumni to learn how they can work and where they should go next. In this course, students don't learn much new, but in return it makes them think about things they haven't thought about and gives them perspective.

What do you see as the differences between universities in the Czech Republic and Sweden?

In Sweden, there is much more focus on topics such as sustainability, global warming, recycling, gender equality, and diversity, compared to the Czech Republic. The fact that these topics are addressed in Sweden means that they are also reflected in the university's policies and are very important to the university. In Stockholm there is a huge opportunity for industry especially in fields such as Machine Learning. Most of the theses in Sweden are done in industry and there is a selection process for them. Besides, at FI, for example, student involvement in labs is much higher, especially for Bachelor and Master students.

What has allowed me to build a research career is that there is a huge support for women's careers in Sweden, and the children are shared equally between partners. Maternity leave is 450 days, of which 60 are reserved for the partner. I was at work half the day and half the day with the children, and so was my husband. This allowed me never to fall too far out of the work pace and lose the basic momentum. This would not be possible in the Czech Republic and if I could change anything here it would be this. An opportunity for women to do it this way if they want to. I see very few opportunities like that here and I see that dads on paternity leave are not the norm, etc.

Where do you see yourself in five years?

Ideally where I am. I've spent 6–7 years building my research group to the point where I'm happy with it and I can say that this is my vision, this is my group, this is who I am as a scientist. I'm at the point where I'm fresh back from maternity leave and only now do I feel like it's all working as it should. In the next 5 years, I want to keep the size of my group first and foremost, focus on the science and make something nice that works, and I know why.