Smart ADAS Verification and Validation Methodology (SAVVY)

Self-driving cars need millions of test miles to prove they are safe, but physical prototypes are scarce and road testing is slow. This project builds a faster, cheaper way to verify and validate Advanced Driver Assistance Systems (ADAS) using computer simulations and hardware emulators. The problem is that modern ADAS features—like automated braking or lane-keeping—must handle an enormous number of possible driving scenarios. Testing each one on real roads with real cars is impractical. The consortium, led by AVL with partners including Warwick University and Horiba MIRA, will create a simulation-based verification process that runs on Field Programmable Gate Arrays (FPGAs) using deep learning and convolutional neural networks. This allows engineers to test ADAS software in a repeatable, controlled environment without needing a physical vehicle. If successful, the methodology could cut the time and cost of bringing new driver-assistance features to market. It would also make those features safer, because engineers could test edge cases—like a child running into the road at dusk—that are too rare or dangerous to replicate on public roads. The project builds on three existing Innovate UK feasibility studies, combining their insights into a single, practical V&V framework for the automotive industry.