Developing a simulator to aid in the design of a safety interlock for self-driving cars

Abstract

Self-driving cars have tremendous potential to be safer than human drivers, but are complex systems for which evaluating safety is challenging: using a statistical approach requires self-driving cars to have clocked on the order of billions of miles of driving to present convincing evidence. Thus there is potential in exploring a new design architecture for self-driving cars in which a small, trusted module of code cooperates with the main controller to ensure safety while being easily verifiable; we call this the safety Interlock. This thesis focuses on the scenario of an ego car driving in a single, straight lane behind a lead car that may suddenly brake. We first propose and prove, using formal verification, an algorithm for Interlock to prevent collision by maintaining a safe separation distance that allows the ego car to stop in time. We then present a simulation program developed using the Processing programming language, which provides visual confirmation of the efficacy of the Interlock algorithm, and is designed to be extensible to more complex road scenarios.