SafeADArchitect: End-to-End Architecture for Safe and Risk-Aware Automated Driving in Urban Environments

Automated driving is still waiting for its major breakthrough and mainstream adoption. One roadblock for mass deployment in crowded and unstructured environments such as urban traffic is the high degree of uncertainty that significantly reduces the vehicle's usability using today's safety solutions. To tackle this challenge, this paper presents SafeADArchitect, an end-to-end architecture for automated vehicles that ensures usability and safety at the same time. It covers major influencing factors to driving safety from the driving platform, sensor limitations or behavior uncertainties through occlusions, inefficiencies of AI-based perception or planning approaches via adequate safety modules and by mapping possible errors to associated driving risks. Hence, by knowing the overall acceptable risk, reasonable driving decisions can be taken that balance usability and safety.