Enhancing Ground Vehicle Dead Reckoning through ABS Sensor and Adjusted Kinematic Model Using Historical Data

Abstract

This study aims to enhance the dead reckoning performance of vehicles and reduce the dependency on observation sensors for state estimation. Observation sensors such as GPS and LiDAR are prone to errors caused by external factors, including signal blockage, computational burden, and interference from reflective objects. These errors critically affect the reliability and safety of state estimation. Therefore, this study proposes an approach to improve the prediction model's accuracy and enable robust state estimation, even in the presence of observation sensor errors. To achieve this, the study focuses on improving the accuracy of longitudinal and lateral state estimation through two approaches. First, for longitudinal state estimation, ABS sensor data was utilized to estimate vehicle speed. By applying an Adaptive Moving Average Filter (AMAF), the filter dynamically adjusts the window size according to speed variations, effectively mitigating noise and ensuring stable speed estimation. Second, for lateral state estimation, a modified kinematic model was developed to address discrepancies between the traditional kinematic model and real vehicle behavior. A polynomial regression-based correction function was introduced using speed and steering angle as inputs, significantly improving yaw rate prediction accuracy. The modified model was validated through Carsim simulations and real-world driving experiments. Additionally, the improved prediction model was integrated into a Robust Extended Kalman Filter (EKF) framework for sensor fusion. By leveraging Mahalanobis distance and residual-based scaling factors, the measurement noise covariance was dynamically adjusted, enabling effective detection of sensor anomalies and enhancing the robustness of state estimation. This study provides a robust solution for accurate and reliable state estimation, even under observation sensor failure or error conditions. By implementing these methods, the research demonstrates the feasibility of achieving precise localization for fixed-route vehicles while improving the reliability and safety of autonomous driving systems.