“We need to promote entrepreneurship in a targeted manner. There is still often a lack of willingness to take risks."

How can we use the behavior of bee colonies to make our agriculture more sustainable? And what can we learn from the swarming behavior of animals for robotics and artificial intelligence? Prof. Dr. Tim Landgraf, head of an interdisciplinary research group at the Dahlem Center for Machine Learning and Robotics at the FU Berlin, works at the fascinating intersection of biology, AI, and robotics.

In this interview, he talks about his research on swarms as “biological computers,” explains how the robot “Polly” controls bee flights in a targeted manner, and why it often takes many hypotheses, data, and patience to turn a biological idea into a functioning AI system. He also provides exciting insights into Berlin's AI ecosystem, reveals which startups particularly impress him, and why universities need to be more courageous when it comes to entrepreneurship.

Prof. Landgraf, what originally led you to research at the interface between biology, robotics, and machine learning—and what brought you to the Dahlem Center for Machine Learning and Robotics?

Social interactions are the basis of all human knowledge and action. I am fascinated by how swarms process information without central control. With my background in AI and robotics, I study these “biological computers” in animal groups.

At the Dahlem Center, I find ideal conditions for combining data-driven methods, robotics, basic research in biology and medicine, and AI method development – true interdisciplinarity instead of mere lip service.

Your research group deals with fascinating topics such as swarm behavior, insect robotics, and bio-inspired AI. Can you tell us about a current project that you are particularly excited about—and why?

Since human social interactions are difficult to track, we mostly use animal models, such as honeybee colonies or fish swarms, to investigate more closely how information is represented and processed, or how the group arrives at decisions. With high-resolution cameras and AI, we can observe all individuals in the swarm, for example, several thousand bees in a beehive. A deeper understanding of the fundamentals enables interesting applications. One highlight is our robot “Polly” – a kind of cinema for bees. It simulates the flight to food sources for individual foragers, making them believe they have found nectar outside. They pass this information on to other foragers in the hive via the bee dance, which triggers targeted flights to real (and by us adjustable) locations. This allows us to control where the bee colony sends its workers, thereby making pollination of agricultural land more efficient, reducing the use of bee colonies, and minimizing transport stress – all of which contribute to sustainable agriculture.

How do you manage to translate biological principles into algorithms and robots in such a way that robust and adaptive systems actually emerge? What are the biggest challenges here?

We usually start with a precise analysis of the biological model – often supported by AI to enable longer, faster, and more objective observation. This often requires several cycles of hypotheses, experiments, and data analysis. In addition to gaining knowledge, these cycles often give rise to interesting research branches: new digital models, AI systems trained on new data, and sometimes bio-inspired physical systems. We have achieved a successful transfer in a swarm charging system for electric vehicles, for example. Just as bees can transfer honey to each other (they have an extra stomach just for this purpose), electric cars could also transfer small “energy snacks” to each other while driving in the future.

However, applications like these are just the tip of the iceberg – the basic research behind them is often laborious, time-consuming, and, compared to other fields of research, rather niche. The main limiting factors are always the same: computing capacity, funding, and, when it comes to personal data, such as in medicine, data protection – a structural problem.

Berlin is considered a growing research location for artificial intelligence and robotics. How do you experience the cooperation between scientific institutions such as the FU Berlin and non-university research centers such as BIFOLD or the Fraunhofer Institutes?

Berlin is small enough that everyone knows each other – and large enough that you can find almost any specialist knowledge within cycling distance. I work with partners at all three Berlin universities, the Charité, and various Leibniz and Fraunhofer institutes. Together, we are building a competence center for open-source hardware, developing components for intelligent healthcare, and researching new AI methods. Even though my research is fundamentally oriented, exciting collaborations with industry and public administration arise regularly – a strong sign of the potential we could leverage in Berlin and Germany.

Where do you currently see the most exciting approaches in Berlin when it comes to translating AI research into industrial applications – for example in robotics, logistics or agriculture?

There are many exciting startups riding the LLM wave, such as Qdrant, Parloa, and VIA Health. Here are a few examples that I particularly like, partly because they are taking an unusual path or are bold in their industry:

• Peregrine – a startup founded by former colleagues at DCMLR that uses AI in dashcams to optimize vehicle fleets – or rather, vehicle swarms. Great founders and a great vision!
• Inceptive – develops models in the field of molecular biology and relies heavily on real data instead of simulations. In biology, everything is somehow connected. Deriving good predictions from the data will make all the difference here. Great team, very capable people.
• 7Learnings – uses AI for dynamic pricing in retail, taking into account the complex market environment. Here, too, everything is connected, and this team has what it takes to create real value from it.
• Korsch AG – a long-established Berlin-based company that uses AI to make its tablet presses even more flexible and responsive – important innovations in a conservative industry!

In recent years, more and more scientifically based AI start-ups have emerged in Europe, including in Germany – keyword UNITE, the central hub for innovation and entrepreneurship in Berlin-Brandenburg. What role do universities like yours play in this, and what structures do you think are needed to promote this transfer even better?

We need to specifically promote entrepreneurial spirit. There is still often a lack of willingness to take risks – partly because this is hardly ever covered in education. I am currently offering a seminar called “How to Start Up” and wonder why I didn't do this earlier. We have technically brilliant students, but little early guidance. The state could specifically promote start-ups from courses, as part of a “hands-on” education, so to speak. To do this, we need IP-light contracts, more entrepreneurship courses starting at the bachelor's level, and open makerspaces where prototypes don't fail because of fire safety regulations. The willingness to take risks in the German investment culture is another stumbling block. Many startups also solve the same problems over and over again – structures like UNITE help with rapid development and are worth their weight in gold, but need to be scaled up further. I think we could also finance shared infrastructure and structurally integrate close partnerships between universities and startups into public contracts.

How do you assess the availability of qualified ML and robotics talent in Berlin – both on the research side and on the business side? Is there a shortage or is the dynamic positive?

Skilled workers are rare, but talent is increasingly drawn directly to industry – even more so than in the past. While machine learning is relatively easy to access thanks to rentable computing resources and freely available learning materials, robotics remains hardware-intensive and therefore more complex. This is also noticeable when it comes to recruiting young talent.

Looking ahead to the next five to ten years: Where do you see the greatest potential for your research – and how can Berlin contribute to remaining internationally competitive?

Understanding collective behavior is becoming more important in many areas – for example, in health (behavior as a disease factor), mobility (traffic jams as emergent phenomena), and politics (AI is changing communication – we need to better understand social dynamics). It will be exciting when these previously separate areas come together in modeling the world and enable a broad understanding of how each part contributes to the whole – mega exciting!

Berlin – and Europe as a whole – must reduce bureaucracy, create space for innovation, invest more courageously, rethink educational pathways, and enable access to (all of our!) data for research and development. Otherwise, we will lose touch.

Thank you very much for the interview.