Generative AI for autonomous driving: Luxoft’s new tool uses AI in automotive testing

When was the last time you had a near miss in your car? Can you describe what the situation on the roads was like at the time? Sure you can. But can you create a machine-readable description of the same traffic situation in an automotive-grade format?! That’s a whole different ball game. Creating such a description requires deep domain-specific knowledge, but this description is exactly what’s needed for virtual validation of an autonomous driving system (ADS). Luckily, generative AI — with the right software application — can now be used to develop simulation-executable scenarios from a natural language description. Our new software application uses generative AI to turn your description of a road incident into a scenario used for virtual validation

To achieve homologation for an ADS, car manufacturers must ensure that their systems function effectively across diverse road environments (including specific predefined conditions deduced from the operational design domain (ODD) specification of the ADS). This necessitates significant efforts in testing, validation and verification — often referred to as the “billion-mile challenge” in autonomous driving (AD).

Not all of those test miles can be driven in the real world: Effort, cost, time and environmental reasons all set a natural limit. So, a scalable virtual validation approach is needed to achieve the trustworthy testing levels, and this virtual validation approach requires detailed descriptions of the diverse road environments.

The devil lies in the details when it comes to creating a consistent and machine-readable description of a traffic scenario. A seemingly simple traffic situation depends on so many individual granularities and parameters that creating a concise description is already very complicated — and creating a description for a complex traffic scenario is even more so.

Consider a conversation with your best friend explaining a dangerous situation you faced on the freeway earlier that day. You would probably tell them where, when and how it happened — in general. Plus, you’d include your version of what the traffic was like and how the other driver(s) behaved. This conversation would automatically imply details that were not explicitly mentioned, like how many lanes, motion static elements and motion dynamic vehicles, and the weather conditions. Yet, even if you were able to take all these conditions into account, the description would still be a far cry from a simulation executable example — the actual creation of such scenario descriptions is tedious and repetitive. And, to create autonomous vehicle solutions, carmakers need to test against hundreds of thousands of scenarios (specified in textual form) and identify crucial combinations or gaps to ensure a sophisticated enough scenario coverage of the ODD¹.

On top of that, interdisciplinary teams have to work together for scenario specification — testing and debugging to avoid ambiguities. These activities are time consuming, will likely cause delays in product development, and require deep domain knowledge. All this at a time when well-qualified experts are difficult to find

To address the process of scenario specification and enable alignment across teams and organizations, the industry has accepted standards such a ASAM OpenDRIVE^®, ASAM OpenSCENARIO^® DSL and ASAM OpenSCENARIO^® XML — all developed by ASAM e.V.’s cross-industry initiative. And to manage the lack of qualified experts, large language models (LLMs) and generative AI are now mature enough to be used as a universal, scalable extension of labor forces².

Such an AI does not exist as an out-of-the-box solution. Developing and training such an AI still requires some effort combined with the necessary domain knowledge. The good news is, it’s a one-time effort: Once the system is set up and trained, it can provide you with a constant flow of scenario descriptions — while the system still continues to learn.