Novel Approach to Improve Performance of Inertial Navigation System Via Neural Network

Abstract

Inertial Navigation Systems (INS) serve as a critical component in nautical, aerial and land-based navigational systems, especially within Global Navigation Satellite System (GNSS) unavailable environments. In recent years with the development of autonomous transportation, it has gained even more popularity. The main drawback of INS's is its ‘drift error’ that increases with on-going travel. This paper proposes a method with which to navigate, by using data from low grade INS sensors (accelerometers and gyroscopes) on-board a moving vehicle by employing Machine Learning (ML) techniques, specifically neural networks. In most cases, GNSS is available, and therefore can be used as an accurate input for the training and optimizing of the ML algorithms. After training, ML can be used in GNSS unavailable environments and urban areas, to improve the performance of the INS. This paper also shows the output results of the machine-learning algorithms compared to the results of the traditional method of using a Kalman Filter.