The future of AI takes off in Winterthur

In recent years, a thriving ecosystem for drones and robotics has developed in the greater Zurich region. Universities are driving technological progress through research, creating spin-offs and providing local tech firms with skilled professionals. However, there is a problem: “Many systems perform excellently under laboratory conditions, but we are still a long way from deploying autonomous drones operationally in airspace,” explains Michel Guillaume, Head of the Centre for Aviation (ZAV) of the School of Engineering. Drones are already in use in many sectors today, be this for surveying, in agriculture or by emergency services. With physical AI, however, a new generation of artificial intelligence is now emerging: systems that perceive their environment, make decisions and act in the real world. These systems confront us with new challenges, as important questions remain unanswered: How can it be ensured that they behave reliably and in an explainable manner in all situations? How can this be tested safely and in a reproducible manner? How can AI-based systems be validated in real-world use and how can they be certified? “In aviation, safety comes first. It takes ten to fifteen years before a new commercial aircraft type is approved and enters into service for an airline,” notes Martin Jajcay from the ZAV. “Autonomous systems such as drones offer promising areas of application, for example in disaster relief operations or logistics in remote locations. For such uses, however, the operational concepts, and in some cases also social acceptance, are still lacking.”

One challenge, for instance, is air traffic management, which becomes many times more complex when unmanned autonomous flying objects operate alongside conventional aircraft. “We simply still know too little about how systems based on probabilistic calculations behave when faced with extreme situations,” says Hella Bolck, a lecturer at the Centre for Artificial Intelligence (CAI) of the School of Engineering. “The systems need to be optimised in such a way that they function in all weather conditions and alongside other airspace users.” A key prerequisite for this is testing, testing, testing.

The ZAV and CAI, together with the University of Zurich and the Zurich University of the Arts, have launched the LINA (Shared Large-scale Infrastructure for the Development and Safe Testing of Autonomous Systems) project. The aim is to build a versatile research and testing infrastructure for autonomous systems and physical AI, especially at the interface between aviation and AI. The initiative is being supported by the Digitalization Initiative of the Zurich Higher Education Institutions (DIZH). Industry partners such as Skyguide and Matternet are also contributing their expertise. The first site is already at an advanced stage: At the Hegmatten glider airfield in Winterthur, autonomous systems at any stage of maturity can be tested on an area spanning 700 by 30 metres.

For Guillaume, the test site represents a long-overdue opportunity for academic groups in Switzerland conducting research in this field as well as for manufacturers of autonomous systems. “It is crucial to bring together all the expertise required for the development, testing and authorisation of these systems in Europe. Without nearby testing infrastructure, there is also a risk that companies will relocate, for example to the US or Canada.” The further advancement of these systems not only involves technological questions, but also regulatory considerations. The test site should therefore also be understood as a scientific instrument that systematically generates the specialist knowledge required to shape new regulations. “We need the expertise of specialists in AI and aviation to take all aspects of integrating AI into account,” says Guillaume. “Thanks to its broad interdisciplinary set-up, the ZHAW is ideally suited to this task.”

The aim of the project is not to develop its own drones. Instead, LINA is positioning itself as a national point of contact for testing and for all questions relating to the development of physical AI and robotic systems. For a system to become operational, many elements are required: in addition to a sound design, functioning AI components and flight testing, this also includes, for example, noise testing, environmental testing and integration with other systems in both air and ground operations. “We need efficient processes that are accepted by the authorities in order to enable us to test systems more quickly under real-life conditions and also to bring them to market,” explains Guillaume in summarising the purpose of LINA. In order to become a competence centre for all matters relating to autonomous systems, the project leads work closely with manufacturers, other universities and authorities, and also engage in PR work. Guillaume stresses that the importance of social acceptance should not be underestimated. Even the best technology is of no use if the drones are so loud that they lead to noise complaints. In addition, the team maintains an international network. As a testing centre in the heart of Europe it offers a great deal of potential to attract researchers and manufacturers from other countries.

The test site at Winterthur Hegmatten is therefore only the beginning for the project team. For tests beyond visual line of sight, it is in any case only suitable for drones weighing less than 25 kilograms, says Bolck. “Testing beyond visual line of sight is, however, essential for autonomous drones. For example, they have to be able to land safely if the connection is lost or the GPS signal fails. For this, we require larger testing areas.” Together with the University of Zurich and the Swiss Federal Laboratories for Materials Science and Technology (Empa), the ZHAW is planning a nationwide network of test sites covering at least 10 square kilometres. The application has already been submitted, and there is no doubt that the demand is there – enquiries from parties interested in testing have been received on an almost daily basis. Guillaume is convinced: “LINA is making a decisive contribution to the next stage of development in AI: systems that not only process data, but act robustly and reliably in the real world. This is applied research at its best, which is precisely our strength.”