<p>Accurate vehicle state estimation is essential for vehicle motion control and safety systems, as it enables precise maneuvering and stability, especially in dynamic driving conditions. A new approach for vehicle state estimation is introduced, which integrates the unscented Kalman filter (UKF) with the estimation of the gated recurrent unit (GRU) neural network to enhance the accuracy and robustness of vehicle motion control systems. The hybrid method combines the output of the network with the model-based UKF to estimate the vehicle states. The estimation network processes sensor inputs and time-series data through a GRU neural network to predict the longitudinal velocity, side slip angle, and their associated uncertainties. Subsequently, the UKF updates the system’s state based on these predictions and dynamically adjusts the reliance on both the UKF and the GRU network estimates in response to varying driving conditions, optimizing overall performance. Extensive simulation test and real-vehicle experiment validate the proposed UKF-GRU method, demonstrating significant improvements in estimation accuracy and robustness over the traditional UKF and purely data-driven approaches.</p>