Augmentation of an extended Kalman filter with a neural network

This paper shows how a neural network can augment a Kalman filter by estimating initial conditions and unknown system parameters. The neural network training is done off-line, using an approach similar to multiple Kalman filters. After off-line training, real-time operation can take place using the neural network without the computational requirements of multiple Kalman filters. An example shows how the general regression neural network (GRNN) augments a Kalman filter for terminal guidance of an interceptor missile