By Dr Jürgen Dickmann, DVN Senior Adviser
Note: On 17 – 18 November, our DVN conference will take place in Stuttgart and for the first time, host special sessions on dual use and on “The Road to Type Approval: Mastering End-to-End AI Systems”.
In the runup to the DVN conference, we are presenting Key ADAS/AV players , and sharing their results, perspectives in this field. The first in this series was our report on a preview drive with Wayve’s latest vehicle; now comes the Bosch interview, and further instalments will follow every fortnight until the conference begins
Dr. Jürgen Dickmann for DVN: Hi, there! What is FKIE all about?
Prof. Wolfgang Koch: Our institute has a long history. At the latest since Russia’s attack on Ukraine, we have again become a strongly defence-oriented institute. One early guiding question was how to create what we would today perhaps call a combat cloud: how to network different radar systems so that a recognised air picture can be generated and courses of action can be offered for how to manage it.
FKIE stands for Research, AI and Ergonomics: research for AI and people. The institute has been very successful, with many developments in data fusion, digital radar, command and control, and human–machine interfaces.
DVN: How do you structure topics that are classified in such a way that they can still be used for doctoral research?
Prof. Koch: That is an important question, and it is also one the Ministry of Defence regularly asks. We have to clarify what may be published. At the same time, it is essential that we do publish, because many of the underlying topics are broadly relevant: sensor data processing, sensor resource management, platform control IT, robustness and safety, and systems thinking.
International cooperation also requires a degree of openness, for example within NATO and with partner nations. Because AI methods play a major role and the field is still evolving, we often have to find a careful balance between secrecy and scientific publication.
DVN: How do you attract talent?
Prof. Koch: People with a strong interest in mathematics are often intrinsically motivated. Fraunhofer offers advantages because we are embedded in the international scientific community, while also giving staff the opportunity to learn project management and many practical skills that are typically not taught at universities.
DVN: When employees leave, where do they tend to go? Do they move to the automotive industry for software and automated driving systems, or to defence companies such as Hensoldt or Rheinmetall?
Prof. Koch: Primarily to our co-operation partners: Rheinmetall, Hensoldt and Rohde & Schwarz. Only very few go to automotive companies such as Audi or Mercedes-Benz.
DVN: What technologies do you use?
Prof. Koch: The institute uses mathematical methods to turn sensor data into robust situational knowledge and to optimise data acquisition by intelligently controlling sensors and platforms. We understand AI as applied mathematics, aiming to prepare complex data for human decision support.
Where conventional approaches reach their limits with imprecise sensor data, the current revolution in data-driven AI models provides significant progress by extracting knowledge directly from training data.
DVN: Is everything AI-based, or do you still use classical, model-based methods?
Prof. Koch: We pursue a hybrid approach. Data-driven AI can deliver breakthroughs where physical models reach their limits, for example in highly complex signal processing or in chemical sensing (‘electronic noses’) where efficient classification is key.
However, where expert knowledge is already precisely describable in mathematical form, it is usually better to use it rather than re-learning models purely from data.
DVN: Do you also work with approaches that could be described as end-to-end learning, or only with task-specific AI models?
Prof. Koch: We use a combination of classical mathematical methods and data-driven AI. A particular challenge in the defence sector is the limited availability of training data. To avoid performance degradation due to outdated models, continuous retraining and rigorous data management are essential.
In networked systems, it is also important to minimise redundancy and to make efficient use of available sensor data.
DVN: In the automotive world, homologation, validation and type approval of AI end-to-end systems are discussed intensively. A recurring point is that AI is a black box and cannot be validated deterministically. How do you address this?
Prof. Koch: Data-driven models can be transformative in complex chemical or electromagnetic signals, but the integration of expert knowledge remains essential where processes are already mathematically well understood.
A central challenge, especially in defence, is the shortage of training data and the need for continuous retraining to prevent performance decay. The experience from large drone programmes shows that regulatory approval and algorithm certification must be considered from the outset if operational deployment in networked systems is the goal. One major German drone programme failed for these reasons.
DVN: What solutions do you see for convincing an approval authority to approve a system whose function cannot be proven deterministically?
Prof. Koch: Software engineering standards (such as the V-model) need to be extended by end-to-end data governance. This includes traceability and accountability for training data, as well as safeguards against data contamination.
Given AI’s capabilities, we need a responsible approval strategy. A legal framework can enable use as long as responsibilities are clearly defined. I believe elements of this thinking are also applicable in the automotive domain.
DVN: Where do you see technology trends that could be useful in automotive, and vice versa? Is dual-use realistic, and where?
Prof. Koch: In the late 1980s, military technology was often the innovation driver for civil industry; later, the trend reversed. We are now seeing another shift: large investments are increasingly flowing into military applications of AI and automation.
The focus is on intelligent automation, the synergy of AI and robotics. This is essential to mitigate acute personnel shortages and to protect personnel on an increasingly transparent battlefield. Current conflicts, such as in Ukraine, underline the need for uncrewed systems (drones) in all domains. Industrial scalability, for instance by drawing on methods from civil automotive production, is also becoming more central to strategic planning.
DVN: So you would expect the automotive industry to turn to defence and learn from it?
Prof. Koch: One way to look at innovation is through the available hardware. In civil automotive applications, even at suppliers, complex tracking algorithms have often been constrained by limited compute, bus systems and cost-driven sensor quality.
Defence applications typically have access to more powerful hardware resources. While I have supervised many automotive PhDs focused on sensor data fusion, there remains a gap in depth: the civil sector often optimises relatively simple algorithms for minimal hardware, whereas military robotics can combine high-grade sensors with complex fusion methods to achieve top-end performance.
DVN: What effects are you seeing of trade wars, embargoes, and current military conflicts on the technology mix and on Europe’s knowledge and technology landscape? Does this affect your work?
Prof. Koch: Very much so. A recent research focus was a year-long study on AI sovereignty in the defence sector. Under the leadership of General Endler (BMVg), an expert commission, involving well-known academics such as Prof. Niggemann (University of Hamburg) and Prof. Roser (University of Munich), developed concepts for strategic autonomy in AI.
The results highlight how complex this objective is. Given global dependencies and interconnections, full AI sovereignty is seen as barely achievable. The strategic goal therefore shifts towards targeted partial sovereignty, to safeguard national freedom of action and technological control in critical areas.
DVN: Why can’t Germany and the EU become sovereign? What prevents us? Does the same apply to the automotive industry?
Prof. Koch: High technology today is characterised by a bipolar dominance of the US and China, because hardly any other actors have the resources for full technological autonomy. For the armed forces, this creates a critical issue, for example when considering the use of generative AI in mission planning.
Sovereign capability requires that we understand how algorithms work, control the data basis, and can maintain and update systems independently. Otherwise, we risk using systems whose internal logic and data history are not transparent.
The gap is also visible in robotics: Chinese drone technology is highly mature, while some European alternatives have shown significant deficits in operational reality, including in the context of the war in Ukraine. The core challenge is therefore to move from a buyer role towards deeper technological self-reliance and long-term control of complex systems. I observe similar dynamics in the automotive industry: key software stack providers are, in many cases, not European, which creates substantial risks if trade conflicts escalate.
DVN: Where do you see FKIE in five years’ time?
Prof. Koch: Fraunhofer FKIE consistently follows its founding approach: combining sensing, communications and algorithms to create situational awareness. In modern defence architectures, this information advantage is a prerequisite for decision and effect superiority.
Over the last decades, the institute has grown substantially, from around 80 employees to more than 650 today, including through the integration of Fraunhofer INT and the establishment of a defence technology hub in Saxony. This scale enables close interdisciplinary collaboration across ergonomics, communications, cyber security and sensing, and the synergies are a decisive advantage.
As the institute grows, leadership requirements also change. In my role as Chief Scientist, I focus on scientific representation and strategic coherence, which also includes actively discontinuing fields that do not fit our profile. Large structures require less micro-management and give departments significant autonomy. That autonomy depends on proactive information-sharing and mutual trust. Despite administrative complexity, the institute is likely to continue expanding and to consolidate its position as a leading actor in defence-related research and development.
DVN: Thank you for this insightful look into the world of defence and potential dual-use opportunities. I hope we will be able to welcome you or your colleagues in the dual-use special session at the DVN conference.
