An integrated artificial neural network–unscented Kalman filter vehicle sideslip angle estimation based on inertial measurement unit measurements

Vehicle dynamics stability control systems rely on the amount of so-called sideslip angle and yaw rate. As the sideslip angle can be measured directly only with very expensive sensors, its estimation has been widely studied in the literature. Because of the large non-linearities and uncertainties in the dynamics, model-based methods are not a good solution to estimate the sideslip angle. On the contrary, machine learning techniques require large datasets that cover the entire working range for a correct estimation. In this paper, we propose an integrated artificial neural network and unscented Kalman filter observer using only inertial measurement unit measurements, which can work as a standalone sensor. The artificial neural network is trained solely with numerical data obtained with a Vi-Grade model and outputs a pseudo-sideslip angle which is used as input for the unscented Kalman filter. This is based on a kinematic model making the filter completely transparent to model uncertainty. A direct integration with integral damping and integral reset value allows the estimation of the longitudinal velocity of the kinematic model. A modification strategy of the pseudo-sideslip angle is then proposed to improve the convergence of the filter’s output. The algorithm is tested on both numerical data and experimental data. The results show the effectiveness of the solution.