"We need artificial intelligence that can be licensed"
VDE: You mentioned in the run-up to an aspect that I would like to expand on. You said that you could learn something from the automotive industry. Can the automotive industry also learn something from the rail industry?
Edel: I had an interesting conversation with a tier-one supplier from the automotive industry about supplying sensors for our assistance systems. We wanted the sensors to be even smarter and better adapted to our needs. My conversation partner then asked me: "How many streetcars are sold worldwide each year?" I replied that there are around 600. He then said: "Even with 600,000 units sold, we can't find any companies in the automotive sector to support us in further development." We have also had similar discussions in connection with fuel cells. In the automotive sector, it only becomes interesting when sales reach millions. We would therefore have to transfer more developments from the automotive sector that are relevant to us to our area.
In the case of automated, assisted driving, however, the sensor technology in the automotive sector is not suitable for the specific requirements of rail transportation. The braking distances differ considerably due to the different tires (rubber on the road, steel on the rails). Although such sensors are available in military technology or aerospace, for example, they are too expensive for a high-speed train. We are currently pursuing a hybrid approach that combines intelligence in both the vehicle and the infrastructure. Can such a solution also be used in the automotive sector? We regularly have this discussion with representatives of the automotive industry, that it's not just about making vehicles more intelligent, but also the infrastructure. The railroads already have an intelligent infrastructure. We can therefore learn from each other.
VDE: You have convinced me with your arguments on the subject of artificial intelligence. But how do your customers react to the topic of AI?
Edel: Customers in our B2B business are, for example, operators of rail vehicles and infrastructure. They are very open to testing new technologies together with manufacturers and as part of collaboration platforms such as Europe's Rail. European projects like this allow us to build and test prototypes. There is little skepticism on the customer side and a lot of courage to innovate. One example of an innovative technology is the European Train Control System, which has developed a common safety system for European rail transport. The idea has been well received and the infrastructure is gradually being upgraded to enable a system solution.
There are similar challenges with AI. If we want to drive autonomously, we need route data in the vehicle. Nowadays, a vehicle doesn't need to know anything about the route, but receives information via the signaling. But for autonomous driving, we need a route atlas that does not yet exist. Information about roadworks, the condition of the route, etc. must be available at all times. This is a challenge that requires digital solutions, such as cloud-based applications.
The question arises as to who is responsible for this, who may have a business model and who bears the costs. Discussing this will probably be more difficult than finding customers for prototype testing.
VDE: This brings us to the question of the challenges involved in integrating artificial intelligence: what areas of tension do you see?
Edel: Siemens Mobility and the entire industry are ambitiously pursuing the development of a licensable artificial intelligence, known as Safe Artificial Intelligence (Safe AI). In order to obtain approval, rail vehicles currently still require a driver, similar to autonomous shuttle buses on the road. This driver can intervene in an emergency and stop the vehicle. However, the question is whether approval is possible without a human backup.
To this end, we have launched the Safe-TrAIn project, which brings together relevant stakeholders such as customers, manufacturers, the German Federal Railway Authority and other approval bodies. Our aim is to develop an AI-based solution that can be approved for rail transport. This is our biggest challenge, not the system's capabilities, as these are already very advanced.