Model Predictive Control for Evasive Steering of Autonomous Vehicle

Abstract

For an autonomous vehicle, a steering controller is essential. Model Predictive Control (MPC) is frequently used for steering control because the performance is sub-optimal and it can take constraints into account. The usual model embedded in MPC for steering control is vehicle body dynamics without steering system dynamics because the response of steering system is fast enough to ignore in usual maneuvers. However, the dynamics of the steering system is not ignorable when a maneuver requires higher torque than a limit of an actuator of the steering system, such as evasive steering maneuvers. To handle this problem, the model embedded in MPC should consider the limit of the actuator with a simple structure for a low computational load. Therefore, this paper proposes a simple model of steering system dynamics that can provide information on the actuator without adding too many states. The proposed MPC is implemented with the steering model with usual vehicle body dynamics as in linear form. It shows better performance than a regular MPC in evasive maneuvers, which is confirmed in simulation and experiments with a scaled vehicle.