Towards a Formal Model for Safe and Scalable Automated Vehicle Decision-Making: A Brief Survey on Responsibility-Sensitive Safety

The promise and potential for a future of automated vehicles (AVs) remains great, with safety and societal transformations that may rival the original introduction of the automobile. Yet an inability for industry and governments to define what it means for an AV to drive safely has tempered enthusiasm and risks causing a “winter of AV” just like the one that affected Artificial Intelligence technologies decades ago, which is only now being overcome. Towards this end, the Responsibility-Sensitive Safety (RSS) model was introduced as an open and transparent white-box, an interpretable and scalable formal model that defines minimum safety requirements based on reasonable assumptions of others, balancing safety and usefulness for automated driving vehicles. The first publication of RSS in 2017 has inspired a global wealth of new and adjacent research, ranging from proposed enhancements to the original model and to the analysis of proposed parameters and novel concepts on how to consider perceptual uncertainty inherent in perception systems. Therefore, in this article, we do a thorough review of the leading contributions to the state-of-the-art thinking around RSS and summarize the contributions or critiques for consideration.