Model-based dependability assessment of fault-tolerant architectures for autonomous driving systems

The design of SAE Level 4 autonomous driving systems requires architectures that can tolerate hardware and software faults to maintain dependable operation. To achieve fail-operational behavior, these systems rely on redundancy and fault-tolerance mechanisms that guarantee continuous and resilient operation over time. This study presents a quantitative dependability assessment of four architectural solutions for an SAE Level 4 Highway Pilot: Triple Modular Redundancy, Channel-Wise Doer/Checker/Fallback, Layer-Wise Doer/Checker/Fallback, and AD-EYE. The analysis employs Stochastic Activity Networks (SAN) to model both independent and common-cause failure scenarios, evaluating probabilistic safety and reliability metrics over transient and asymptotic system behavior, along with a temporal indicator represented by the mean time to failure. A comprehensive sensitivity analysis investigates the influence of critical model parameters on these measures to assess the robustness of the results and highlight the main architectural trade-offs. The results provide quantitative evidence of each architecture's strengths and limitations under different operating conditions, supporting the design of dependable and fail-operational autonomous systems.