Why is testing driver-assistance systems so slow?

The question

Driver-assistance systems are largely tested in simulation, because testing them in real cars is slow, expensive and unsafe. But realistic simulations take a long time to compute, and they also need to model how real drivers behave. We asked two questions:

1. How can the virtual validation of driver-assistance systems be sped up?
2. How does modelling human behaviour help their development?

What we did

• Defined the review protocol: research questions, keywords, inclusion and quality criteria.
• Searched Scopus with combinations of six keywords and screened the results.
• Extracted and compared the methods, data and findings of the relevant studies.

Findings

Most research focuses on driver behaviour: take-over situations, fog, distraction, warnings. Far fewer studies tackle the computing time itself. The promising ones use simplified models (reduced-order models, 2D instead of 3D) and smarter sampling of critical scenarios. Our conclusion: research should focus more on reducing simulation time, which would make adding human behaviour affordable too.