Formal Methods and Safety for Automated Vehicles: Modeling, Abstractions, and Synthesis of Tactical Planners

Abstract: One goal of developing automated road vehicles is to completely free people from driving tasks. Automated vehicles with no human driver must handle all traffic situations that human drivers are expected to handle, possibly more. Though human drivers cause a lot of traffic accidents, they still have a very low accident and failure rate that automated vehicles must match. Tactical planners are responsible for making discrete decisions for the coming seconds or minutes. As with all subsystems in an automated vehicle, these planners need to be supported with a credible and convincing argument of their correctness. The planners interact with other road users in a feedback loop, so their correctness depends on their behavior in relation to other drivers and road users over time. One way to ascertain their correctness is to test the vehicles in real traffic. But to be sufficiently certain that a tactical planner is safe, it has to be tested on 255 million miles with no accidents. Formal methods can, in contrast to testing, mathematically prove that given requirements are fulfilled. Hence, these methods are a promising alternative for making credible arguments for tactical planners’ correctness. The topic of this thesis is the use of formal methods in the automotive industry to design safe tactical planners. What is interesting is both how automotive systems can be modeled in formal frameworks, and how formal methods can be used practically within the automotive development process. The main findings of this thesis are that it is viable to formally express desired properties of tactical planners, and to use formal methods to prove their correctness. However, the difficulty to anticipate and inspect the interaction of several desired properties is found to be an obstacle. Model Checking, Reactive Synthesis, and Supervisory Control Theory have been used in the design and development process of tactical planners, and these methods have their benefits, depending on the application. To be feasible and useful, these methods need to operate on both a high and a low level of abstraction, and this thesis contributes an automatic abstraction method that bridges this divide. It is also found that artifacts from formal methods tools may be used to convincingly argue that a realization of a tactical planner is safe, and that such an argument puts formal requirements on the vehicle’s other subsystems and its surroundings.