A Conceptual Framework of Dynamic Risk Management for Autonomous Vehicles

Abstract

Hazard Analysis and Risk Assessment (HARA) of ISO 26262 evaluates system malfunctions resulting in hazardous events to determine Automotive Safety Integrity Levels (ASILs). To ensure system safety, these ASILs are used to identify corresponding safety measures like fault prevention, tolerance, or mitigation, which act as safety goals to be implemented during the design time. The entire process of HARA is based on worst-case assumptions and the very premise that a human driver is available to take control of the vehicle and is thus responsible for its safe behavior. However, Autonomous Vehicles (AVs) are safety-critical systems operating in uncertain and dynamic environments envisioned to function without human attention and intervention. These worst-case assumptions prompt intensive safety measures – which are not only unnecessary for certain situations but also result in limited performance of the vehicle. Therefore, in order to ensure safety in uncertain environments and in the absence of a human driver, Dynamic Risk Management (DRM) is necessary. To this end, we propose a DRM framework that takes a risk-dependent decision for a safe reconfiguration of the system at runtime. We use data from the on-board sensors to perform risk assessment and trigger safety-actions when necessary. This aids the vehicle in making decisions that ensure its safe behavior at all possible times.