VeriCAV

Self-driving cars will face millions of simulated traffic situations—including rare crashes and erratic pedestrians—before they are allowed onto real roads, and this project is building the testing framework to make that possible. Current validation methods rely on a limited set of scripted scenarios, which cannot cover the near-infinite variety of real-world driving. VeriCAV aims to replace that approach with an automated system that generates optimal test cases from the entire space of possible situations. A “test oracle” will evaluate the automated driving system’s performance during each simulation run, flag failures and compile coverage statistics without requiring constant human supervision. The project will also create realistic digital agents—other vehicles and pedestrians—that behave unpredictably, forcing the system to react. If successful, the framework could give developers and regulators genuine confidence that an automated driving system is safe before deployment. The project will verify its own accuracy by running a real automated driving system in simulation and then comparing those results with physical tests using the same system in a real vehicle. This matters because the gap between simulation and reality has been a persistent barrier to certifying self-driving technology for public roads.