Project Summary
Autonomous vehicles (AVs) have the potential to transform California’s transportation system, but a critical gap remains: agencies and stakeholders lack a principled methodology to assess AV safety and performance in ways that support transparent and trustworthy deployment decisions. Many consequential decisions concern counterfactual deployment conditions where little or no direct testing data are available, such as whether AV service can expand to a new neighborhood, operate under different traffic conditions, or introduce a larger fleet into an existing service area. Blindly extrapolating from existing testing data is risky, while comprehensive real-world testing under every target condition is costly, time-consuming, and potentially disruptive. To address these challenges, this project will develop SAFER-AV, a Simulation-Augmented Framework for Efficient and Reliable Autonomous Vehicle Evaluation. SAFER-AV will combine limited real-world AV data with scalable simulation-based testing to forecast AV safety, operational performance, and traffic-system impacts under target deployment scenarios. The framework will explicitly account for discrepancy between the testing results in simulation and real-world traffic using paired real-world and simulated AV rollouts and quantify uncertainty to identify when existing evidence is sufficient and where additional testing is needed. The project will produce a prototype toolkit and evaluation protocol as deliverables to share with relevant California stakeholders for facilitating transparent, trustworthy, and grounded AV deployment decisions in California.
