Asbach discusses the refinements needed in the application of AI in testing, the essentials of an effective simulation program and how automation specialists are addressing some of today’s biggest headaches
What are the key elements of a safe test strategy for ADAS and AVs?
First, a validated simulation environment. Second, the selection of the ‘right’ test scenarios. Third, authorization in combination with a certain infrastructure.
Silicon Valley has arguably led the way in AV development and testing. From a safety perspective, do you agree with the aggressive approach taken by developers there?
This is a matter of the safety culture in Europe, especially the safety understanding of the public. I personally prefer a slower approach, because I think the risk of being stigmatized is too high if something goes wrong. It would make the wide application of autonomous vehicles more difficult if we needed to ‘reconvince’ the public.
AI is changing the way driverless vehicles are tested. What’s your opinion on AI?
The application of AI in autonomous vehicles is difficult as the behavior is not always determined. The use of AI for scenario mining or other data-based approaches to generate (parts of) a test specification is very straightforward and will help increase the testing possibilities. Some weeks ago, a colleague of mine finished his master’s thesis. He tested the influence of AI methods on generating realistic scenarios from a traffic simulation. The interesting result is that right now, a configuration calibrated by an expert delivers significantly more realistic scenarios than the tested AI methods. Therefore, maybe the development of AI application needs some more time, but the potential is there.
Is simulation the answer for efficient and safe CAV development and testing?
Yes. Most of the testing can be executed in simulation environments, especially a vehicle’s interaction with the infrastructure.
What simulation tools are the most and least effective here and why?
The most important thing is not to forget classic test approaches. First of all, the sensors have to be tested extensively. This can be done via a hardware-in-the-loop testbed. They need to be checked under different conditions and we need to be sure that they recognize every object under every (or most) conditions.
In a second step, a sensor model has to be created and a digital model of the vehicle has to be integrated in a simulation environment. This setup needs to be validated, of course. Afterward, it can be used to test the correct behavior of the vehicle with different traffic, roads, infrastructure and weather.
Not all road infrastructures are available, so it is impossible to validate simulation results at a very detailed level. Therefore, spot checks can be implemented to gain trust in the simulation and to be the final test run in specific operation domains.
How can industry players work together to ensure the best autonomous vehicle testing strategy?
Usually by joining research projects with a clear focus on application. However, sometimes the focus is too far in the future and the results are hard to apply.
What milestones still need to be reached to achieve widespread adoption of automation across the transport sector?
In Germany, a new law has opened the door for Level 4 authorization. This law needs to be deployed across the rest of Europe in a similar way. Right now we are performing the first iteration of the process, including the aforementioned simulation approach. We aim to show a way through the process to make application easier.
How realistic are industry predictions for the introduction of full autonomy? Do you believe we will ever reach full autonomy?
In my opinion, it will not be soon. We will see Level 4 introduction, particularly within freight traffic; also for certain connections of public transportation. All other transportation sectors will take longer to reach this point.
In his role at the DLR, Asbach is responsible for the testing of autonomous and connected transportation systems, particularly simulation-based analysis for the automotive and railway sectors. He is also key account manager for the Test Bed Lower Saxony for automated and connected mobility, including requirements engineering.