Event-Triggered Model Predictive Control for Autonomous Vehicle with Rear Steering 2022-01-0877

This paper proposes a new nonlinear model predictive control (NMPC) for autonomous vehicle path tracking problem. The vehicle is equipped with active rear steering, allowing independent control of front and rear steering. Traditional NMPC, which runs at fixed sampling rate, has been shown to provide satisfactory control performance in this problem. However, the high throughput of NMPC limits its implementation in production vehicle. To address this issue, we propose a novel event-triggered NMPC formulation, where the NMPC is triggered to run only when the actual states deviate from prediction beyond certain threshold. In other words, the event-triggered NMPC will formulate and solve a constrained optimal control problem only if it is enabled by a trigger event. When NMPC is not triggered, the optimal control sequence computed from last NMPC instance is shifted to determine the control action. Simulation using a bicycle vehicle dynamic model is conducted and the numerical results show significant throughput reduction of event-triggered NMPC while maintaining comparable control performance. Specifically, event-triggered NMPC requires only 50% of computational efforts compared to traditional time-triggered NMPC, thus improving its real-time implementability with production grade electronic control unit.