In the era of autonomous vehicles, state estimation is crucial for planning and control. While various sensors like GNSS, IMU, cameras, and wheel encoders provide essential data, each has limitations. Sensor fusion, especially combining GNSS and IMU, addresses these challenges. Recently, data-based techniques using machine learning tools have emerged to enhance tuning processes. However, if an architecture that has a practical impact is utilized, the training of the neural net results in a complicated task. This paper explores different state estimation architectures in the era of data-based techniques and proposes an algorithm generating reference tuning values. The presented methods are tested with real vehicle measurements.