Whether as autonomous vehicles, bolted to the factory floor, or in human form, robots are taking on a growing range of tasks, creating new opportunities and questions for society. In Freiburg, Professor Abhinav Valada is working to make these machines more flexible. In this interview, the computer scientist and robotics expert talks about recent advances in robotics, the challenges of reliable deployment, and the societal implications of automation.

Professor Valada, when manufacturers show humanoid robots folding laundry or tidying up the kitchen, it looks impressive. How close are we to having them in our homes?

These demonstrations reflect substantial progress in robot capabilities. I expect many more deployments over the next five years, particularly for repetitive tasks in factories, warehouses, and other semi-structured environments. Nevertheless, a robot that reliably undertakes every household task must have a broad range of skills and great adaptability.

What are the big technical challenges?

Flexible robots need to work reliably in changing environments and handle situations they have never encountered before. Safety is particularly important because robots act directly in the physical world, often close to people and objects. Over long tasks, even small errors can affect later decisions. Robots therefore need to recognise uncertainty, detect when an action has not had its intended effect, and recover safely. Depending on the situation, they may need to slow down, ask for help, or stop. Our research at the Robot Learning Lab in Freiburg aims to connect these capabilities so that robots can learn from experience while preserving what they already know.

You mean that they can find their way in situations they weren't explicitly trained for?

Exactly. Many robot learning systems are developed using curated demonstrations and data collected for particular tasks. Generative AI helps robots draw on broader knowledge from images, videos, and language. But data showing robots acting in the physical world remain much scarcer. We want robots to apply their knowledge to unfamiliar objects and situations as well. One promising direction is world models that predict how an action will change the environment. 

Could you walk us through an example?

Take autonomous driving as an example: commercial robotaxi fleets demonstrate considerable progress within defined operating conditions. Extending these services to new regions can introduce different weather, road layouts, construction zones, and traffic habits. Our research addresses how systems can recognise unfamiliar situations, predict what might happen next, and take uncertainty into account when making decisions. Similar questions arise in factories, farms, and homes, although each application has its own physical and safety requirements.

One particular focus of your research is less technical and more ethical in nature: you investigate what "fair" robotic systems should look like. What does that mean?

At its core, fairness means avoiding unjustified differences in how people are perceived and treated. For example, we need to examine whether a robot detects people reliably across different appearances, ages, and mobility aids. We must also look at its behaviour: whether it systematically gives some people less space or makes them wait longer. Fairness does not always mean identical behaviour. A wheelchair user may need more space, for example. 

Your expertise would also be highly welcome at technology companies. What keeps you in academia?

Above all, the freedom to pursue a long-term research agenda while staying close to practical problems. We collaborate with many companies, while also investigating questions that may only become practically relevant in five to ten years. Then there is the educational mission. I find it very rewarding to help young researchers become independent scientists. The people we train carry ideas into companies and universities around the world, extending the impact of our work far beyond an individual project.

You're a board member of the Zuse School ELIZA. ELIZA brings together the German units of ELLIS, a European research network for machine learning. How do you assess Europe's position in the international AI landscape?

Europe would benefit from further investment in computing resources, growth funding, and support for scaling technology companies. In robotics, it has strong foundations in perception, control, robot learning, and trustworthy AI, alongside established automotive, manufacturing, sensor, and mechanical engineering industries. Physical AI brings these strengths together because progress depends on real machines, data, and engineering expertise as well as computing power. 

AI and robotics are changing how many tasks are performed. How do you see their long-term impact on work?

I expect the effects to vary across sectors and occupations. In ageing societies and sectors with labour shortages, robotics can improve productivity and take over dangerous or physically demanding work. But people will experience these changes differently, so continuing education, social protection, and worker participation in deployment decisions will be important. Researchers, industry, and policymakers all have a role in helping the benefits reach as many people as possible.

Interview: Steffen Ermisch (20 October 2026)