Can Data-Based Risk Acceptance be Trusted for Autonomous Vehicles? Challenges and Insights

Risk acceptance is a critical element for the safety assurance of autonomous vehicles. To argue the system is free from unreasonable risk, organizations use data from accident databases that are publicly available. The common practice in the automotive industry is to compare the performance of the system to that of human drivers. However, the data from these databases are often interpreted in a way that enables organizations to showcase that an autonomous vehicle operation is safe, although they may not be. Additionally, these databases have disparities when it comes to processes and modes of collection of accident data. This leads to the question: How can we trust such risk acceptance? In this session, we’ll detail the challenges we face when we use data-based risk acceptance – and what the repercussions are. We’ll also discuss some potential directions we can explore to reduce these problems.