A new weight updating method for INS/GPS integration architectures based on neural networks

Inertial navigation system (INS) and global position system (GPS) technologies have been widely utilized in many positioning and navigation applications. Each system has its own unique characteristics and limitations. Therefore, the integration of the two systems offers a number of advantages and overcomes each system's inadequacies. INS/GPS integration is usually implemented using Kalman filters. However, Kalman filters perform adequately only under certain predefined dynamic models and suffer from several problems related to observability and immunity to noise effects. An INS/GPS integration method based on artificial neural networks (ANNs) to fuse INS measurements and differential global positioning system (DGPS) measurements has been recently suggested. Although able to provide high performance INS/DGPS integration with accurate prediction of position components during GPS outages, the conventional methods of updating the ANN weights limit the real-time capabilities. This paper offers a new weight updating criterion to improve the limitation of traditional weight updating methods with the utilization of two different architectures; the position update architecture and position and velocity update architecture.