Targeted on-line modeling for an extended Kalman filter using artificial neural networks

The authors compare implementation techniques of an extended Kalman filter that is augmented by an artificial neural network that trains online. The purpose of the neural network is to model mismodeled dynamics of the system that are used in the process of the extended Kalman filter. The authors compare using a neural network that augments the entire model to a neural network that targets the dynamics of specific system states. The idea is to show that targeting specific states will reduce computations while maintaining a high degree of effectiveness