Precautionary Safety for Autonomous Driving Systems: Adapting Driving Policies to Satisfy Quantitative Risk Norms

Abstract

Road safety has been studied for more than 80 years with the objective to prevent traffic accidents, injuries, and fatalities. Proper road design, standards, traffic rules, driver education and law enforcement are continuously improved to reduce the crash risk and enable mitigating actions. While significant advancements have been made, humans still frequently do mistakes, sometimes with severe consequences. In this paper we introduce the notion of Precautionary Safety and propose a methodology such that autonomous vehicles can adjust their trajectory planning to their capabilities, external conditions, and knowledge on human mistakes in order to satisfy overall requirements on accident-, injury- and fatality rates. More precisely, we describe how to make adjustments to existing driving policies so to satisfy real-world safety requirements, rather than only obeying to the law and traffic rules. As an illustrative example, the methods are applied to accident scenarios between vehicles and jaywalking pedestrians.