Model Predictive Control for Evasive Steering of an Autonomous Vehicle

Abstract

Model Predictive Control (MPC) is frequently used for the steering controllers of autonomous vehicles, and simple vehicle dynamics is usually used for the reference model. However, simple vehicle dynamics is not sufficient for evasive steering maneuvers that require very fast steering actions. The objective of this study is to design an MPC that works for such maneuvers. We considered the steering system model and vehicle model in the MPC design. The proposed steering model provides information about the disturbances, the bandwidth of the actuator, and the motor voltage while maintaining a simple structure. The controller computes less demanding desired steering angles that do not violate the voltage constraints at the cost of tracking error. Otherwise, the MPC would result in failure of the desired evasive maneuver due to voltage saturation causing failure of the steering angle tracking. A simulation and experimental results confirmed that evasive maneuvers are successfully performed with the proposed steering controller within the allowable voltage range, and it outperformed a regular MPC based on a simple vehicle model.