Risk Monitoring and Mitigation for Automated Vehicles: A Model Predictive Control Perspective

Despite recent advances in algorithms and technology, self-driving vehicles are still susceptible to errors that can have severe consequences. As a result, effective risk monitoring and mitigation measures for autonomous driving systems are in high demand. To overcome this issue, several specifications and standards have been developed. However, a theoretical framework for dealing with autonomous vehicle hazards has rarely been presented. This study suggests a risk modeling method inspired by ideas from control theory and introduces a Model Predictive Control (MPC) Framework to deal with risks in general. Two application examples are presented. The first example shows how MPC parameters may affect the aggressiveness of the response. In the second example, our proposed risk monitoring and mitigation module is integrated into a vision-based Adaptive Cruise Control (ACC) system. Simulation results indicate a significant improvement in collision avoidance rate (from 0% to 47% in edge scenarios) during the Euro NCAP ACC Car-to-Car tests with a stationary target, which demonstrates the utility of our approach for addressing various types of hazards faced by autonomous vehicles.