Driver interference and risk in semiautonomous braking under uncertainty

Abstract

Emergency maneuvering systems can take control of a vehicle in high-risk situations caused by distracted, fatigued, or careless drivers, which can reduce the frequency and severity of collisions. But in order to override the user's control the vehicle must reason with uncertain information: sensing provides noisy and partial input, vehicle dynamics models are never perfectly calibrated, and other agents (vehicles, pedestrians) may behave unpredictably. In the context of collision imminent braking (CIB), we explore the tradeoffs between risk and interference with normal driving behavior that are inherent in the presence of uncertainty. Specifically, control policies that take a conservative approach to uncertainty are more likely to brake unnecessarily. We compare several control policies on different scenarios using Monte-Carlo hardware-in-the-loop simulations in order to quantify their behavioral characteristics in terms of collision risk and driver interference. We demonstrate that a relatively high degree of safety can be achieved at a relatively low degree of interference, but “idealized” behavior is unattainable in the presence of uncertainty.