The strange braking maneuvers repeatedly caused angry customers: the vehicles stopped for no apparent reason at Google-sister company Waymo in Phoenix. Hundreds of minivans are now on the streets of the US city without a driver.
The company’s software and AI experts found the source of the problem: birds. The cars mistook them for obstacles and braked abruptly whenever one approached the vehicle. Waymo improved the dataset and algorithms-and the complaints stopped. The taxi service is now running so satisfactorily that the company was able to expand its service to San Francisco a few days ago.
The example shows that self-driving cars behave differently than humans in many cases. The neural networks are trained according to the precautionary principle: keep a distance, accelerate more slowly and stay longer at the intersection. In a way, the systems do what people learn in driving school but often quickly forget in everyday traffic: they drive extremely defensively.
This will have consequences for everyday traffic, as simulations by the software company PTV show. Even a 20% share of autonomous vehicles in traffic is enough to significantly disrupt traffic flow. With a 50% share, traffic congestion doubles.” That doesn’t bode well for the cities,” says Jochen Lohmiller, who managed the project at PTV.
Researchers are therefore relying on so-called digital twins, which are intended to prevent gridlock. The streets are recorded by cameras, radar, lidar, and other sensors and processed by artificial intelligence (AI). A digital image of the traffic is created in real time, which allows more efficient route planning and increases traffic throughput.
Level 3 is ready for series production
The topic is being pushed with force onto the agenda of traffic planners because autonomous driving is no longer a vision of the future. You can already drive the new S-Class in traffic jams at speeds of up to 60 kilometers per hour (km/h) without having to put your hands on the steering wheel. The so-called Level 3 is ready for series production. Under certain conditions, the driver relinquishes control of the vehicle but must be able to intervene again at any time.
Realizing fully autonomous driving at Level 4 is technologically more demanding. But here, too, progress is unmistakable, as numerous test projects around the world show. At the highest level, 5, the steering wheel and driver are completely superfluous. It will probably take many years before this level of autonomy becomes part of everyday life.
Autonomous vehicles avoid accidents, reduce emissions and create new freedoms. The robotic cars will also improve the flow of traffic, but only when significantly more than 50 percent of the vehicles are self-driving. The networked cars can then coordinate with each other, calculate and maintain optimal speeds and distances.
According to traffic experts, nothing is better for the flow of traffic than evenly moving vehicles. The problem lies more in the interaction between autonomously driving cars and those that are driven by people. “People simply react too differently,” says Professor Knoll. “At the traffic light, for example, there is an accordion effect.” Drivers overlook the fact that the line of cars is moving in front of them; the gaps between the vehicles become larger and fewer cars come through the green traffic light.
More traffic jams, more waiting times
It will be a long time before self-driving cars dominate urban traffic. Statistically, even if all newly registered cars and trucks could drive at Level 5 starting today, it would take 9.8 years for all conventional vehicles to be phased out.
What happens in the transitional period when careless pedestrians, aggressive drivers, and careful robotaxis share the road? PTV got to the bottom of this question. The company, with 900 employees, was until recently a subsidiary of VW-parent company Porsche SE, is the world market leader in software for traffic systems.
In an experiment conducted last year by the PTV team, the experts observed the rush-hour traffic in the early morning: Which vehicles are on the road? How long do the traffic light phases last? How many cars come through the intersection when the light turns green?
PTV also reproduced empirical values from self-driving vehicles and their traffic behavior, collected by leading providers such as Waymo, Cruise, or Aptiv. The team fed the data into their model and created several scenarios. According to one version, self-driving cars account for 20% of all traffic. The most visible consequence is that the flow of traffic slows down significantly – the average speed falls from 23.5 to 20.9 km/h.
Self-driving cars are like sand in the gearbox for everyday traffic. For example, PTV determined an average speed of 58 km/h for human drivers in Cologne – faster than permitted in a built-up area. The autonomous vehicles, on the other hand, follow the regulation exactly at 49 km/h.
The traffic flow is disrupted the most when the share of self-driving vehicles is 50 percent. Then they hardly make any headway, especially on single-lane roads, braking too carefully and keeping a large safe distance. The number of stops goes up from an average of 1.8 to 3.3—an increase of 80 percent.
At the crossings, this has a major impact on the queues. If the average traffic jam is 55 meters long without autonomous vehicles, it increases to 114 meters with 50 percent autonomous vehicles.” The cities have to think about it,” says project manager Lohmiller. “With the new technology, everything doesn’t automatically get better.”
One solution could be dedicated lanes for autonomous cars. In the cities, however, there is often not enough space for this. At some point, these lanes will no longer be sufficient due to the increasing number of robotic vehicles. The even more radical idea of banning human-driven vehicles is unlikely to be politically feasible in the foreseeable future.
Fewer accidents, more efficiency
The scientists observed a lot in five years: accidents with serious injuries or lowered cars that raced at 100 km/h through an intersection where only 50 km/h is actually allowed. Due to data protection, the researchers did not report the incidents to the police. But the data is valuable in other ways.
A digital analysis of traffic in real time could help to avoid accidents by reporting that a car has come to a standstill in a lane of the motorway. Or the authorities could set up speed traps at speeders’ focal points to prevent accidents.
Above all, however, the digital twins should prevent traffic jams. On motorways, the emergency lanes are opened to traffic at peak times, but not in bad weather. With radar or infrared-based sensors and the simulation of the digital twin, the motorway operators could observe the traffic much more precisely and make more tailored decisions.
Digital twins can also be made of city traffic, as the pilot project by the Technical University of Munich showed. This allows the traffic for the next 60 minutes to be predicted with a high degree of accuracy. The planners could activate traffic control systems, optimize traffic light phases, or redirect robotic vehicles.
It would even be conceivable to centrally control the autonomous cars. One would ensure that the vehicles travel at an even speed and thus ensure a “high traffic throughput,” as Professor Knoll puts it. “This is currently being implemented in China.”
Traffic data as a platform
Concern about privacy isn’t even the biggest issue. The strongest argument from critics is that the cameras in cars already provide enough data to map the traffic. However, Knoll does not accept this. The cameras are only mounted at the driver’s height and would not look behind. There is no overview; the data sets are incomplete and not accessible. “It’s like wanting to abolish street lighting with the argument that car headlights do exist,” says Knoll.
The police, municipalities, and road operators could react more quickly to accidents or traffic jams. Fewer roads would have to be built to handle the same traffic. “The traffic area could be used much better,” says Knoll.