A Monday morning in Berlin in 2050. Students on their way to class click their destination on their smartphones. The traffic app sends their location to the next autonomous shuttle-on-demand. The bus comes and picks them up. Via radio it informs the other vehicles in the area that it is ready to start up again. The autonomous cars brake gently and allow the bus to merge into traffic. The next traffic light already knows that the shuttle is approaching and turns green.

Students who ride their bikes can also reach the lecture hall in a more relaxed frame of mind: their smartwatch signals to them in advance of the traffic lights whether it is worth putting in a sprint. No? So it’s better to shift down a gear because it will be red when they reach the intersection anyway. No more nerve-wracking honking. No hard braking. No verbal outbursts on the two-wheeler or behind the windshield, and there will probably be fewer traffic jams. Rush hour traffic, a steady, calm flow... that is how it could be in the not-so-distant future.

Daniel Göhring, a Junior Professor of Mobile Robotics and Autonomous Vehicles in the Department of Mathematics and Computer Science at Freie Universität Berlin, and his team are researching how corresponding “cooperative maneuvers” between autonomous vehicles could work. His work is part of the joint project “AI-based system for networked mobility” – KIS’M for short – funded by the German Federal Ministry for Digital Affairs and Transport with 9.53 million euros. In addition to Freie Universität Berlin, the Berlin Senate Department for the Environment, Urban Mobility, Consumer Protection, and Climate Action; Technische Universität Berlin; the Berlin Transport Authority (BVG); Fraunhofer FOKUS; and the German Aerospace Center are all involved.

Communication between Vehicles Is Important

Vehicle-to-vehicle communication is the slogan that one day could ensure smooth traffic in a big city like Berlin. In the medium term there would be a traffic mix in which autonomous cars share the road with conventional vehicles, i.e., vehicles driven by people. On the one hand, this involves less complex processes, such as merging an autonomous shuttle bus into the flow of traffic. “We are in the process of developing solutions for how the bus can tell the autonomous vehicles in its immediate vicinity that it now wants to enter the flow of traffic. The cars are then made to brake,” explains Göhring. Emergency vehicles from the police and fire department would also be able to influence autonomous vehicles by radio to form a rescue lane. This is no easy task, even for some flesh-and-blood motorists.

“The second focus is on communication with traffic lights. This should apply to both buses and bicyclists, who will be informed whether they can still make the next green phase. Our project partners are developing an app for this,” says Nicolai Steinke, a member of the KIS’M project. A third focus is “passenger support.” If there is no longer a driver on the bus, you can no longer ask anyone: Does the bus also go to Schloßstraße? Where do I have to change trains if I want to go to Ku’damm? It is also necessary to find a technical solution for how people with disabilities can get in and out of the vehicle. “As technicians, we would of course prefer to accompany everything and everyone with cameras. Of course, that would be a major intervention. That is why we are trying to involve citizens in the project. We want to find solutions that meet both the needs of traffic safety and data protection,” Göhring emphasizes. Citizen surveys and other forms of user-centered technology development are therefore being developed as part of KIS’M. “We want to implement the results in our algorithms.” Is that to avoid ending with a high camera density such as that already in use in China? “Cameras have a bad image, but you have to be able to understand where the person is going. This could also be tracked anonymously. Or by the person triggering the tracking themselves,” Steinke points out.

Autonomous Vehicle on the Streets of Berlin since 2011

In concrete terms, the research team is first developing detection algorithms. They will then simulate the relevant traffic situations, and finally hit the road using the university’s own autonomous vehicle, which has been allowed to drive freely on public roads in Berlin since 2011. “With a person on board, of course,” Göhring emphasizes. In the KIS’M project, autonomous shuttle buses run by the Berlin public transportation system will communicate with Freie Universität’s autonomous vehicle. What already works in simulation will need to prove itself in reality. And all this in real time because the “vehicle brain” cannot afford to pause for thought. In the worst case, that could have fatal consequences.

Speaking of which: it is not yet clear whether autonomous driving will make traffic safer. Scientists still hope that will prove to be the case. Computers strictly execute the rules given to them. However, for most of the world’s problems, such as recognizing people, it is not so easy to create a solution formula by hand. Göhring points out that to improve the safety of autonomous vehicles, the scientists have to resort to methods such as machine learning, which cannot be proven, but can only be empirically proven. “Because if something has worked in a million cases, that is not proof – it’s statistics.”

Göhring also notes that people make surprisingly few serious mistakes when driving: there are only 1.67 fatal accidents for every billion kilometers driven on Europe’s highways. He says that for fully autonomous traffic, the road traffic regulations could be simplified. Traffic lights and right-of-way rules would still be needed, of course. And speed limits: when using autonomous vehicles, more than 130 kilometers per hour on highways is not feasible, for safety reasons, as well as for energy-saving reasons.

Humans Best at Predicting Behavior of Others

However, many special rules would be less relevant. “If everyone really drives autonomously, pedestrians could generally be given the main ‘priority,’” argues Nicolai Steinke. They are the most difficult factor to detect using sensors and are therefore a virtually unpredictable factor. “Is the elderly lady still crossing the street or not? How do I spot the kid running out from behind the car? What about at night? How do animals react? So far, AI experts have had a hard time answering questions like this,” Daniel Göhring admits. Humans are best at predicting the behavior of others. Currently, it is much easier for them than artificial intelligence to predict how their peers will behave. The main tasks of the research team at Freie Universität are therefore to improve the quality in the recognition of people and all other participants in traffic and to predict their behavior with subsequent path planning.

---

This article originally appeared in German on May 8, 2022, in the Tagesspiegel newspaper supplement published by Freie Universität Berlin.