Robust vehicle state estimation method for synergistic trajectory planning and tracking control

Abstract

In the field of autonomous driving, a trade-off exists between the physical consistency of signals required for trajectory planning and the low-latency response demanded by tracking control. To address this problem, a Robust Adaptive Extended Kalman Filter method based on Limited Memory and Maximum Correntropy (LM-MC-AEKF) is proposed. First, a Finite Impulse Response (FIR) structure is constructed by employing a fixed-length sliding window mechanism, aiming to reduce phase lag and meet the requirements for real-time tracking. Second, the Maximum Correntropy Criterion (MCC) based on a Gaussian kernel is utilized to replace the traditional Minimum Mean Square Error (MMSE) criterion, adaptively suppressing non-Gaussian sensor outliers to provide statistically robust and physically feasible state inputs for the planning layer. Furthermore, an adaptive update architecture for noise covariance driven by MCC-weighted innovation is designed to compensate for model parameter mismatches online. Joint simulations based on CarSim and Simulink indicate that the proposed method reduces the longitudinal velocity estimation error by approximately 95% compared to traditional filters, and compresses the yaw rate phase lag from 30.5° to 3.8°. Simulation results indicate that this method yields improved data support for upper-level trajectory planning, while mitigating the risk of lower-level control instability caused by perception delays, thereby improving the overall robustness of the system within the simulation boundaries.