Advancing Safety Validation & Testing of Autonomous Vehicles Using Advanced Simulation Techniques

Even as extensive efforts are put in to the research and development of self-driving vehicles, many safety concerns persist. On-road testing presents unrealistic costs for manufacturers, leading to a development cycle that moves into deployment before sufficient validation has occurred. Additionally, there are a multiplicity of driving scenarios that combine in various ways to define what safe driving behaviors look like, and it is not clear what finite set of scenarios would constitute a valid safety test. In my doctoral dissertation, we thoroughly examine virtual simulation as a tool to overcome these obstacles toward the validation of self-driving vehicles. There are two primary threads that tie all of the work described in this dissertation together: diagnostic clarity and realism. Both of these concepts are explored through original research and novel approaches, expanding the capability of virtual simulation as a tool for validation of real-world self-driving cars.