Autonomous Vehicle Safety Reasoning Utilizing Anticipatory Theory

Today we have autonomous vehicles already on select road-ways and regions of this country operating in and around humans and human operated vehicles. The companies developing and testing these systems have experienced varied degrees of success and failure with regard to safe operations within this public space. There have been safety incidents that have made national headlines (when human fatalities have occurred) and their also exist a litany of other physical incidents, usually with human operated systems, that have not grabbed the headlines. Some of the select communities where these autonomous systems have been operationally tested have revoked access to their roadways (kicked out) some of these companies. As a result of these incidents recent data suggests that the public trust in autonomous vehicles is eroding [1]. This situation is couponed by the fact that there are no established safety standards, measures or technological methods to help local, state or national entities to ensure that these systems are operating under any level of safety scrutiny. This situation has accelerated the need for innovative research within the domain of autonomous vehicle safety approaches.

This paper describes a new methodology for automated driving to address these safety issues that entails the creation of a new computational process we call the Safety Reasoning System (SRS). This system will monitor and adjust the actions of an autonomous vehicle operating in highly cluttered scenarios with a focus on traffic intersections (specifically T-intersections). The SRS works in probabilistic space and models the world into propositions informed by both current and projected data sets. By inferencing on the relationships between data sets we are able to form anticipated safety propositions on the likely effects of the autonomous vehicles projected actions. Thus, potentially reducing the occurrence of catastrophic outcomes.