Cars don’t yet automatically get us from A to B. The tech group AVL is teaming up with startups to change that.
The car is becoming a mobile living room: future fantasy or soon reality? Getty Images / mathisworks
Where are the autonomous cars now? We’ve been talking about it for years, but no vehicle in Germany rolls through our streets without human intervention. We are currently in the third of five possible evolutionary stages on the path to self-driving cars. The ADAC calls this “highly automated driving”. A few registered cars can transport us independently through slow-moving traffic on the motorway.
But only from level four onwards is the occupant allowed to hand over complete control to the vehicle. While driving, we could read a book in the car or even sleep.
But there are a few hurdles to overcome before then. “This is an area in which we are still relatively new in terms of homologation,” says Max Nestoriuc. He is head of development for driver assistance systems and autonomous driving at AVL. The Stuttgart company develops the necessary technologies as a partner for the automotive industry. In addition, AVL works with the responsible authorities, for example to develop the regulations necessary for Level 4 systems. “There is a first version of the legislation in Germany and one in the EU, which has not yet been used for homologation. “There are still no 100% clear guidelines,” says Nestoriuc. That still involves a lot of interpretation of the available legislation and guidelines, and research into norms and standards in order to generate a blueprint.
Max Nestoriuc works at AVL together with a team of 300 experts to develop new technologies for the automotive industry AVL
In order for autonomous cars to be allowed to drive at all, they must be safer on the road than human drivers. But for appropriate approval, many tests and kilometers driven are necessary to prove this safety.
Requirement: One billion kilometers of driving distance
Nestoriuc calculates: If the autonomous vehicle is to be five times better than human accident statistics, for example, then one fatal accident should only happen every ten million kilometers. In order to be able to prove this statistically, you would have to drive ten times as much: i.e. 100 million kilometers. Further security again requires a factor of ten: “I would ultimately have to drive a billion kilometers with the autonomous vehicle to be released,” said the AVL expert. This applies to every software iteration that you want to release for homologation.
“It would take roughly ten to 20 years to implement this,” estimates Nestoriuc. So completely different approaches and methods are needed to bring such a system into circulation. “That’s why you have to discuss this with the authorities and TÜV during the development process.”
Test autonomous driving with simulations
One approach to mastering the high requirements is simulation: the driving systems are tested completely in virtual worlds. Up to 80 percent of the kilometers required for approval can be tested in this way, says Nestoriuc. The remaining percentage is distributed via a so-called “multi-pillar approach” to tests with real hardware in simulations, on test tracks – and, of course, real field tests, which account for the fewest kilometers.
Up to 80 percent of the necessary tests for autonomous vehicles can be carried out using simulations AVL
But simulations come with their own challenges. “The main problem is: How do I represent reality as best as possible? How is the credibility of the simulation?” So the comparison between what I do in the simulation and what happens in real life. During the tests, it must be proven for each of the 15 to 20 sensors – such as radar, lidar or ultrasound – that the models work in the simulation and reflect reality, says Nestoriuc. Added to this is the complexity of the real world, which is difficult to simulate, such as the behavior of pedestrians, different movement sequences or faces. And: Object recognition that was trained with images from the real world cannot easily be transferred to 3D avatars and a virtual world.
There are many problems, but also many possible solutions. For some, AVL relies on startups – and has launched the Creator’s Expedition program. The goal: cooperation with startups. The question: “How can a business case be generated from which both benefit?” “Startups gain customers, AVL gains the startups’ technology,” says Nestoriuc.
Startups deliver necessary technologies
An example of this collaboration is the startup Deepscenario, also from Stuttgart. The founders work with drones to detect objects and their movements from the air. In co-op with AVL, this technology is used to analyze the behavior of vehicles entering and exiting motorways. AVL then embeds the data into its systems so that the car learns how a person would behave – or what the respective safety needs are, depending on the situation on the road. There are other co-ops with the startups Aves Reality, NXRT and Leftshift One, among others.
The startup Deepscenario uses drones to detect objects from a bird’s eye view. This allows scenarios for autonomous vehicles to be trained in Deepscenario
The importance of the data from the drone also shows: A car can only use its sensors to detect what is happening around the vehicle. The discussion about autonomous driving is therefore often about infrastructure data, such as an intelligent camera that monitors traffic above a traffic light and could pass this data on to an autonomous vehicle. This brings with it new problems: Is the data prepared in such a way that it can be interpreted correctly by every vehicle? How do the systems communicate? Via Wifi? Via 5G? And if an accident happens at the traffic lights, who is to blame: the autonomous car or the infrastructure manufacturer that provided the data?
That’s one of the reasons why the federal government isn’t investing billions in the sector, says Nestoriuc. It is simply still too uncertain which standards will become established. But yes, autonomous driving will still come. The regulations are developed, the technologies are refined, the tests are carried out virtually and in reality. And those who get in early now have the chance to define the standards.
Read too
“Fully autonomous driving will remain a dream for decades”