I recently had the pleasure of sitting down with Anneli Björkman from Uppsala University’s UppTalk series for a conversation that turned out to be one of the most engaging interviews I’ve had in a long time. You know those rare moments when you’re so absorbed in a discussion that time seems to evaporate? This was one of those experiences.
The interview lasted 32 minutes, but I genuinely thought we were still in the opening stages when Anneli started wrapping things up. That’s the sign of a truly great conversation when you lose the sense of time because you’re completely immersed in the exchange of ideas.
What really excites me and what we spent a significant portion of the interview discussing is our focus application are which is multi-robot systems. Rather than perfecting a single type of robot, we focus on understanding how different types of robots can work together toward common goals. When should they collaborate? When should they communicate? How can robots of different sizes, shapes, and abilities form an intelligent network. In the lab, we have quite a collection: high-precision robotic arms, smaller wheeled platforms with less precise arms, quadruped robots that look like dogs, robot cars, and even spider-shaped robot a new addition to our robot colleagues. Each has different capabilities, and the magic happens when they learn to work together.
One of the most fascinating parts of our conversation was exploring how biological ecosystems inspire our work. During my research visits to marine stations in Kristineberg and Askö, I realized that oceans and forests aren’t filled with identical organisms all doing the same thing. Instead, there’s incredible diversity – different species, different sizes, different capabilities – all coexisting and thriving together. This is exactly what we aim for in multi-robot systems. Just as in nature where diversity creates resilience and efficiency, we don’t want homogeneous robot populations. We want heterogeneous teams where each robot brings unique strengths to the collaboration.
We dove deep into the various forms of artificial intelligence being used in robotics. As I mentioned in the interview, the term “AI” is often used too broadly in the media. In our lab, we primarily use reinforcement learning, and specifically multi-agent reinforcement learning for our multi-robot systems. And to me, the beauty of reinforcement learning is that we don’t explicitly program every action. Instead, we define rewards and penalties, and the robots learn through trial and error. I demonstrated this with a few examples from a simple one to more complicated one. The real exciting thing for me is the emerging behaviours and my lab is currently working on very exciting scenarios that we can finally observe and record these interesting behaviours that emerges from interactions between different robotics systems.
A huge thank you to Anneli Björkman and the entire UppTalk team – including colleagues Maja Garde Lindholm and Johan Åhrgren – for creating such a welcoming and intellectually stimulating environment. The recording studio at Blåhus provided the perfect space for this conversation. These platforms for making research accessible to alumni and the general public are invaluable.
Whether you’re a robotics enthusiast, a student considering a career in technology, or simply curious about how artificial intelligence and physical systems are converging to shape our future, I hope you’ll take 32 minutes (that will feel like 5!) to watch the interview.
Didem Gürdür Broo 