Driving Maneuver Estimation for Naturalist Driving Data with State Space Model Predictive Control

In recent years, naturalistic driving data (NDD) has played an essential role in understanding driver behavior and has become a critical resource for the development of more advanced driver assistance systems (ADAS) and autonomous vehicles (AVs). However, accurately estimating driving maneuvers from NDD is a challenging task due to the complexity and variability of real-world driving behavior. In this research, we propose a state space model predictive control (SSMPC) approach to estimate driving maneuvers from NDD datasets, including NGSIM, HighD, ZTD, Mirror, and Dajiang were collected from the US, Germany, Japan, and China, and encompass diverse driving scenarios. The SSMPC models the underlying dynamics of the vehicle's acceleration and steering behaviors and employs model predictive control (MPC) to predict the optimized maneuver based on the observed driving data. Additionally, we compare the estimated tracks to observation tracks with the proposed dynamic time warping (DTW) algorithm, mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). Results show that the SSMPC effectively estimates driving behaviors for NDD datasets and performs with high accuracy and robustness. Our approach has potential applications in ADAS and autonomous vehicles, where accurate and robust maneuver estimation is crucial for safe and reliable operation.