9. statistical validation

Why is statistical validation needed for L3 automated driving?

Real-world driving alone would require too much time and mileage to demonstrate very low accident rates. Statistical validation uses evidence and models to estimate safety risk.

What is allowed with BMW Personal Pilot L3?What are the approximate human safety benchmarks on German highways?

Permitted activities include reading, taking notes, using mobile devices, watching videos, and video calls. The example applies to traffic jams up to 60 km/h.

What is missing from ISO 26262 and ISO 21448?

They do not provide a complete method to quantify and propagate general uncertainty into a vehicle-level risk estimate.

(?) What are IFVs and SPVs?

Influence Factor Variables (IFVs) describe environmental/scenario conditions. Safety Performance Variables (SPVs) describe safety-relevant system performance. Statistical validation models how IFVs affect SPVs and ultimately risk.

What is factor screening?

Factor screening identifies which potential influence factors significantly affect system performance or risk. The lecture uses Design of Experiments, regression, and hypothesis testing.

Why are Bayesian Networks used in statistical validation?

They model conditional dependencies between environmental factors, system performance, and risk, allowing different sources of knowledge and data to be combined.

What is empirical factor modeling?

It estimates the joint probability distribution of the IFVs from real-world data, including dependencies between variables.

What is empirical performance modeling?

It models the distribution of SPVs depending on the IFVs, using evidence from field and proving-ground tests, typically through regression-based methods.

What is residual risk analysis?

Residual risk is the risk remaining after safety measures have been applied. It is estimated using stochastic/Monte Carlo simulation, potentially with variance-reduction methods for rare events.

What is the purpose of sensitivity analysis?

It identifies which input variables have the strongest influence on the resulting risk, so the most safety-relevant aspects of the system can be improved.