Nikita Haupt

• Ensuring safety of Highly Automated Vehicles (HAVs) presents substantial challenges to the prevailing safety assurance approaches within the automotive industry. The current state-of-the-art automotive Hazard Analysis and Risk Assessment (HARA) process relies on worst-case operational assumptions. This results in a robust system design guaranteeing safety, but at the cost of reduced performance and availability, a high reliance on the human driver for control takeover, and a static representation of the environment. As driving automation increases, the involvement of the human driver in driving-related tasks decreases, shifting the responsibility for driving and safety from the human to the system. Consequently, it becomes crucial for the system to be aware of its own functional capabilities and the current operational situation. This can be achieved by shifting certain aspects of safety assurance, such as risk assessment, to runtime. To this end, we present a novel framework, Safety-rules-based Runtime Risk Assessment for Automated Vehicles (STATUS), that demonstrates and implements the shift of risk assessment from design time to runtime. We introduce a systematic approach to elaborated-HARA (elHARA) that incorporates the dynamic entities of the environment during the analysis of hazardous events. Subsequently, we use ontology formalism to represent the information obtained from elHARA and create a corresponding set of safety rules. These ontologies and rules enable the creation of a knowledge base containing risk-relevant entities used for risk inference at runtime. To evaluate the effectiveness of the proposed framework, we apply it to an exemplary Highway Pilot Assist function for a set of simulated highway scenarios. We show how elHARA is performed and facilitates comprehensive situation analysis, resulting in the determination of suitable safety integrity values. Furthermore, we demonstrate how the knowledge base, created using the ontology models and predefined safety rules, enables the system to reason about and infer risk at runtime.