BMW expert: "There won't be any acceptance problems with autonomous driving"
2025AD: Dr Wisselmann, you can be considered a pioneer of automated driving, having worked on driverless cars since 2004. What was your first encounter with the technology? Why did it fascinate you?
Dirk Wisselmann: One job of a researcher in the automobile industry is to "transform" future mobility visions into hardware so that the vision can be physically experienced as a basis for an internal evaluation. We picked up the topic of autonomous driving "“ in a BMW specific manner - in the year 2004 driven by a technical fascination. The result was the "BMW-TrackTrainer" "“ a car that was driving autonomously on the ideal line of racetracks with racing speed. We used this technology on many European racetracks within our driving training.
2025AD: How has the significance of automated driving developed over the years at large car manufacturers like BMW?
Dirk Wisselmann: The increasing automation of vehicle functions is inextricably tied to the development of cars. Therefore the introduction of advanced driver assistance systems (ADAS), like parking distance control (PDC) or the radar based speed and distance control (ACC), in the nineties of the last century was a logic step. The enormous increase of the performance of these ADAS in the last years will now make the next step realizable "“ to give the driver the possibility to completely delegate the driving task to the car if he wants to do so. Fully automated driving also offers the chance to realize completely new mobility concepts with a high potential to optimize the traffic situation especially in urban areas.
2025AD: Surveys show that many consumers remain skeptical when it comes to driverless cars. How can OEMs tackle this issue?
Wisselmann: If you ask "žnormal" drivers, if they want to use a self-driving car, you are very often facing a high degree of skepticism, especially in Germany. The people are scared or irritated by the loss of control, the aspects of safety, the individual benefits and so on. In my experience, these negative arguments disappear once the drivers experience an autonomous car in real life. Then they are fascinated that the cars are already working very well, they develop trust in the new functionality very quickly and they acknowledge the positive aspects of comfort and safety. Therefore, I am sure that we will not have any acceptance problems with autonomous driving. The prerequisite is of course that the autonomous car is driving on a very high technical level. The reverse conclusion is also valid: a "bad" autonomous car confirms all prejudices and worsens the acceptance situation.
2025AD: Most consumers haven't seen an autonomous vehicle on the road yet, let alone sat in one themselves. Some of your competitors like Volvo include average consumers in their testing efforts, hoping to increase acceptance. What is BMW's stance on this?
Wisselmann: The evaluation of new functionalities by normal customers to the earliest possible date is an integral part of the standard development process of BMW. The first evaluation usually starts in the driving simulator. In the next step, BMW invites customers to test drives on test and proving grounds and on public roads at a later stage of the development. All these activities are ongoing in an intensified mode today for the development of our highly and fully autonomous cars. An even broader field operational test would have to be discussed with respect to communication aspects.
2025AD: Once they use them, people tend to quickly trust automated cars, which can create risky situations if they suddenly need to take back control. Did you have similar experience with test drivers? What can be done to avoid this "complacency effect" "“ or should we skip Level 2 and maybe even 3 straightaway?
Wisselmann: Theoretical discussions of the human-machine-interface of so called level-2 functions, for example the lane and steering control assistant, are very often critical, because these functions take away the "workload" from the driver and they degrade him only to monitor the driving task. In reality, the circumstances are different. A broad investigation of the accident behavior of one million BMW cars equipped with level 1 and 2 functions over the period from 2014 to 2017 has shown that these cars had between 15 and 60 percent less accidents than cars without any ADAS. This enormous increase in safety shows that all theoretical secondary effects like "overtrusting" are by far overcompensated. For this reason, we see the introduction of level 3 functions as an important first step. In specific driving situations the car requests the takeover of the human driver, for instance to drive through a complex working zone. Intense studies in driving simulators have shown that the driver has enough time to execute this takeover process safely. In the medium term, these level 3 functions will be replaced by level 4. Then, the driver will be able to sleep on a long distance trip on highways because the car is able to handle safely all driving situations by itself.
2025AD: We all hear a lot about all types of routes and trips that fully automated vehicles have mastered. So far so good. But maybe it would also be very fruitful be to talk about daily situations that automated vehicles in general still have difficulties with. Can you shed some light on this?
Wisselmann: On a long term view, the biggest challenge for automated cars will be heavy weather conditions. I'am not talking about a "normal" rain shower but conditions where human drivers also operate on the limits of their perceptibility. Examples are rain storms in autumn, or heavy snow fall with snow-capped roads in the winter time. In these situations, the performance of an experienced human driver is still above the machine capabilities. The complexity of the driving situation itself is generally not a problem of automation. Here we are facing problems that are more practical. That means we need enough training and testing material for our KI-algorithms to interpret very complex driving situations correctly.
2025AD: What is BMW's overall strategy when it comes to testing autonomous vehicles? What are the different cases and scenarios that need to be covered?
Wisselmann: The testing of autonomous cars is completely different from the present testing of ADAS. For an automated emergency braking system for example you can quite easily define the potential accident situations and investigate these in the simulation or in reality with prototype cars. With extensive driving tests on public roads with experienced test drivers it is guaranteed that no faulty activations appear. The increase in road safety mentioned above shows that this procedure works well for today's ADAS. For autonomous driving, it is clear that it is not sufficient only to look at these "specific use cases". Here the car manufacturer has to ensure that the car behaves safely in all future driving situations. Theoretical investigations have shown that "“ depending on the accident type and the driving environment "“ between 240 million and 6.6 billion testing kilometers would have to be driven to be sure that the autonomous driving is safer than the average driver. This extent is not achievable in reality.
Simulations play an important role in driverless car testing. (Photo: BMW)
2025AD: Simulation seems to be seen as the magic tool to let cars learn. But where are the limits? What cases can be simulated and what cases need to be tested in real-world driving situations?
Wisselmann: The validation of autonomous cars therefore requires the usage of all today's methods in a new combination:
- As a first step, prototype cars with safety drivers are testing the autonomous driving in normal- and specific driving situations. Here no safety critical driving situations can be investigated because this would be too dangerous for the test drivers and the other participants on the road. The testing kilometers are in the range of double-digit millions.
- On the basis of this broad situation catalog of normal and specific driving situations, a stochastic simulation is performed to vary all situation parameters to reach a "100%"-coverage of all future driving situations.
- Safety critical driving situations are then safely investigated on test tracks and in simulation studies. These situations are derived from the existing test catalog of ADAS as well as from field data and from the accident statistics and accident analysis.
Overall, this mixed approach based on a mix of experiments and simulations ensures that we fulfill our enhanced safety targets in real life traffic later on.
2025AD: When it comes to real-world testing, where on the planet do you test the different driving scenarios or road and weather conditions?
Wisselmann: It is necessary to test autonomous cars in every market. Every country is different with respect to traffic rules, traffic signs, driving style of the human drivers, weather conditions and so on. This leads to an enormous effort "“ but we are used to enormous efforts from the development of ADAS.