Real-time Recursive Risk Assessment Framework for Autonomous Vehicle Operations

Abstract

Existing risk assessment (RA) methodology used for autonomous vehicle (AV) development and validation is insufficient for future AV operations. Existing frameworks operate based on processes such as hazard analysis and risk assessment (HARA) where risk is defined based on functional hazardous event severity and the likelihood of occurrence. This is a static process performed during the development stage and relies on prior lessons learnt and know-how. A drawback of this is the omission of potential complex environments that could occur during real-time – especially with more stringent safety requirements for AV operating at higher automation levels. Therefore, there is a need for an additional framework to further enhance the safety levels of the AV, focusing on real-time instead of static risk assessment during development. In this paper, a novel real-time recursive RA framework (ReRAF) addresses the gap by creating a novel risk representation, predictive risk number (PRN), and eventual safety levels (SLs) in the temporal and spatial domain. This approach focuses on risk assessment based on AV collision to the detected hazardous object and controllability of the AV. A dynamic recursive RA continuously captures potentially hazardous events in real-time and compares them with past occurrences to predict future safety actions. ReRAF provides a continuous improvement on the RA and acts as an additional safety layer for AV operations.