Safety Goals & Challenges for Self-Driving Technologies
Thursday, October 08, 2020: 1:00 PM - 1:45 PM
It is a common theme among the self-driving and driver-assistance engineering communities: "We are going to make driving safer!" However, there is little (if any) consensus on what that statement actually means. Today, technology developers and government agencies (like the NHTSA) talk about driving safety as a function of "miles are driven" - but this perspective does not allow direct comparison of performance to other modes of transport or industries who have established practices for designing and testing safety-critical systems. In this session, we will review the current state of self-driving safety by looking at the implications of data published each year in "disengagement reports." We will then transform that data into the more commonly used framework of System Reliability Analysis methods to directly compare the performance of self-driving vehicles today with human drivers and other modes of transport such as commercial aviation - which have a known high standard of reliability. We will discuss the question of "how safe is safe enough?" for self-driving cars and provide examples from other industries for reference. With this perspective, we will then discuss the challenges of closing the gaps between the current performance of self-driving systems and other human-operated vehicles. And while significant technical challenges remain, there are also societal, legal, and regulatory challenges that must be addressed before self-driving technologies can begin to improve the safety of our roads.
Analyst,Engineering/Technical,Government/Policy/Regulatory,Journalist,Legal,Research & Development