D-Tracking: Digital Twin Enabled Trajectory Tracking System of Autonomous Vehicles

The precision of trajectory tracking significantly influences the driving safety of autonomous vehicles. Therefore, it is crucial to accurately estimate and use control algorithms to reduce the tracking error between the real vehicle and the expected trajectory. However, the discrete expected trajectory points generated by traditional planners result in an estimation bias of tracking error. In this paper, we propose a digital twin (DT) function to calculate the projection point of the real vehicle on the expected trajectory, serving as the position of the DT vehicle, resulting in a more precise estimation of tracking error between the real vehicle and the expected trajectory. We then use the lateral and longitudinal control algorithms of autonomous vehicles to enable the vehicles to track the expected trajectory with the expected speed and acceleration by reducing the tracking error. For lateral control, a Model Predictive Control (MPC) algorithm and a Linear Quadratic Regulator (LQR) algorithm are improved by incorporating a feedforward Proportional-Integral-Derivative (PID). For longitudinal control, a real data-based PID algorithm is designed for longitudinal trajectory tracking. Experimental results based on real vehicle experiments demonstrate significant improvements in both lateral and longitudinal control accuracy, as well as driving stability. Compared to state-of-the-art model-based lateral control algorithms Double-layer MPC and Baidu's Apollo LQR, the lateral error is reduced by more than 60%. Compared to the model-based longitudinal PID algorithm, the longitudinal speed error is reduced by at least 38%. The oscillation amplitude of the wheel angle and yaw angle is also obviously decreased.