Imagine a machine which ingests ADS real-world driving logs and simulation results and emits a safety evaluation with good traceability to the source data. Further imagine that this machine is automated and forms a part of the CI/CD and safety validation flows. Why would we want such a machine? What would it look like? Which enabling technologies would be needed? How would different stakeholders (e.g. AV developers, fleet operators, insurance firms, government authorities) extract value from it? The presentation will answer these questions, offering a blueprint for ADS safety evaluation that can be utilized by AV developers as well as various third parties.

• What is an automated and data-driven approach to safety and how is it different from existing approaches?
• What does a blueprint of a data-driven safety framework look like? What enabling technologies are needed?
• How can different stakeholders like AV developers, fleet operators, insurance firms, government agencies, etc derive value from a data-driven safety evaluation?
• How should a data-driven safety evaluation framework be deployed?