A Robust Physics-Informed Neural Network for Vehicle Sideslip Estimation

This paper presents a novel vehicle sideslip angle estimation framework that integrates an Unscented Kalman Filter (UKF) with a Physics-Informed Neural Network (PINN). Generally, labelled data from optical sideslip sensors is scarce and expensive to obtain. Instead, unlabelled measurements from standard vehicle sensors span much broader ranges of operating conditions, but they cannot be used in a traditional neural network.Our method leverages unlabelled data through a physics-based loss, enabling a cost-effective and comprehensive data acquisition strategy, and improving robustness and sideslip estimation performance. The proposed architecture is evaluated against conventional model-based and data-driven estimators, showing significant improvements across multiple performance metrics.