17.9.2026 | Metzler meets Fraunhofer – Interview
Prof. Dr. Norbert Elkmann is Head of the Robotics Systems Department at the Fraunhofer Institute for Factory Operations and Automation (IFF) and an honorary professor of Assistive Robotics in the Department of Computer Science at Otto von Guericke University in Magdeburg, Germany. His research focuses on assistive and service robots, safe human-robot collaboration, and AI in robotics for manufacturing usages as well as in the medical and healthcare sectors.
Pascal Spano is Head of Research in Metzler’s Capital Markets division. Before joining Metzler, he was managing director of PASST Digital Services GmbH, a fintech company he co-founded in Cologne, Germany. His other professional experience includes stints at UniCredit Group, Credit Suisse Ltd., Deutsche Bank, and ABN Amro. He has been a member of the German Finance Association for about 20 years.
Artificial intelligence is transforming robotics – shifting away from rigidly programmed, task-specific applications towards adaptive, general-purpose agents that operate in open, dynamic environments. Tasks that robots couldn’t perform just a few years ago can now be automated.
Pascal Spano, Head of Research at Metzler Capital Markets, spoke with Prof. Dr. Norbert Elkmann, Head of the Robotics Systems Department at the Fraunhofer Institute for Factory Operations and Automation (IFF), about how AI robotics can be applied safely, reliably, and effectively.
Pascal Spano: Professor Elkmann, where do we stand today with AI in robotics? What has already been implemented and what’s still a long way off?
Prof. Dr. Norbert Elkmann: We are currently witnessing a paradigm shift in robotics. Robots are evolving from programmed systems designed primarily for repetitive tasks into adaptive, learning systems that integrate perception, planning, and execution. AI-supported perception systems, such as camera systems used for bin picking or object recognition, have already found practical application. For us as robotics engineers, the field of AI-driven robotics – i.e. the autonomous planning and execution of robotic actions – is naturally of great interest. This enables us to automate processes that couldn’t be automated until now, e.g. assembly and disassembly, welding, packing and unpacking, etc.
However, there is a gap between what’s shown in the media and what’s already in use in the industry. Conventional industrial robots and cobots don’t even have an interface for direct control of the robot axes by an AI agent. Fully autonomous humanoid robots with human-like abilities – currently the subject of so much media attention – are, in my opinion, still a long way off.
Elon Musk says every household will have a humanoid robot some day. What are your thoughts on this, and what obstacles do we need to overcome?
The vision of a humanoid robot with human-like abilities in every household is fascinating. Yet no one knows if or when this will be possible. The advantage of a humanoid form is that our world was designed for humans: doors, stairs, tools, kitchens, workplaces, and care settings are all based on human proportions. But the hurdles for using robots in the home are particularly high. Households are open, unstructured, and highly variable. In such a setting, a robot must not only see and navigate but also grasp objects with precision, handle textile items, cope with uncertainty, and operate reliably and safely in close proximity to people. Fine motor skills and tactile sensitivity remain difficult to master, e.g. when grasping flexible objects or folding a piece of fabric.
Fine motor skills, in particular, demonstrate just how challenging seemingly simple everyday tasks are for robots. Where are the benefits already more visible today, in other words, in which industries do you currently see the greatest added value from AI-powered robotics?
In the industrial sector, a wide variety of processes – particularly those with high variability that couldn’t previously be automated – will be automated using AI-powered robotics in the near future. This can help address the shortage of skilled labor and increase efficiency and quality in production. However, even these applications require a certain amount of engineering effort during the setup and training phases. In any case, the greatest added value of AI-powered robotics will be seen in the industrial sector in the foreseeable future.
Industrial applications are therefore the obvious starting point despite the ongoing effort required for training and integration. But as soon as humans and robots begin to work side by side, another question arises: How can AI robotics be designed to be safe, trustworthy, and socially acceptable, especially in the context of human-machine interaction?
This question is often overlooked, yet it’s essential for practical purposes. The cases described regarding AI-driven robotics in assembly and disassembly aren’t a concern as long as the robot performs its tasks autonomously. Things get interesting when robots and humans work side by side, perform tasks together, or cross paths. That’s when the issue of safety comes into play, along with all the applicable regulations and standards. In Germany, a zero-accident strategy applies; injuries aren’t tolerated and must be prevented by implementing appropriate safety measures. Incidentally, different requirements apply in the USA and China. Existing safety concepts are based on deterministic robot behavior as defined by the program. Safety-critical components are designed with dual redundancy to achieve the necessary safety level. But AI-based robotics is not deterministic. Because the AI agent is trained, humans can’t verify the robot’s actions in detail. There’s a fundamental contradiction here between the use of AI and current normative safety requirements. In addition to the technical issues surrounding AI-supported robotics and humanoid robots, concepts, solutions, and new regulations and guidelines must also be agreed upon for this area. Added to this are issues surrounding data protection. AI-powered robotics, and humanoid robotics in particular, will only be considered trustworthy and socially acceptable once they comply with our applicable regulations regarding safety and data protection.
Liability issues, safety standards, and regulations limit the scope of application for AI robotics. In what areas would you like to see a shift in thinking or new legal frameworks in Europe?
It’d be virtually impossible to lower safety standards in Germany and Europe – and that’s a good thing. However, we need new methods and modernization of the verification and approval processes. Safety standards and testing procedures must therefore better reflect how learning systems can be safely evaluated in varying environments. Simulations of robot movements running simultaneously are expected to play an important role here. We call this “automated clearance.” In this process, every robot movement is checked for compliance with standards before execution, and the speed is adjusted if necessary. These issues are already being addressed in standards committees.
The issue of liability also remains unresolved. Who is liable if an AI robot makes a mistake and someone gets hurt? The manufacturer, the user, the systems integrator, or the software developer? Here, too, Europe needs clear, practical guidelines. Added to this are differences in how standards are applied in industrial settings, public spaces, and private homes. Against this backdrop, humanoid robots in private homes represent a vision of the future.
So it’s not about lower safety standards, but rather procedures that can realistically test learning systems. Let’s look at the practical side. What examples from your projects most impressively demonstrate how AI robotics provides real benefits?
Humanoid robots are currently impressing audiences with somersaults, sprints, and dance routines. From a technical standpoint, this is impressive, but where is the practical benefit? I see the benefit in automating processes that couldn’t have been automated a few years ago without the use of AI, and in shifting away from repetitive robotics toward flexible, adaptive robotics. This involves AI-supported planning and execution of a wide variety of processes, as well as interaction and task delegation – for example, via speech and gestures – and object and environmental recognition.
Looking ahead, what technological breakthroughs in this field do you think will have the greatest impact on the economy and our daily lives in the next ten years?
On the one hand, specific foundation models will be integrated into robotics, creating new application possibilities with less effort required for implementation, training, and adaptation. Another important topic is the manipulative ability – or dexterity – of robots. Intensive research is being conducted to develop new hands and technologies. However, humans have immense advantages in this regard. We humans can sense the properties of materials, reposition objects in our hands, exert great force relative to the size of our hands, and much more. These dexterity capabilities, which robots don’t yet have, represent a significant limitation in many applications. Skillful two-handed manipulation, hand-eye coordination, and the use of knowledge that comes from experience are all areas where humans still have a huge advantage over robots. Even though technological progress is advancing rapidly, I still doubt that humanoid robots will be standard in private households in the next ten years.
Thank you very much for the interview.
