Enhanced inertial navigation with low-cost sensors using deep neural networks

Abstract

In this paper, enhancing accuracy of inertial navigation using only inertial measurement units (IMUs) is addressed. Classical inertial navigation algorithms integrate inertial measurements for state estimation, which result in fast deterioration of accuracy due to accumulation of errors. A method of correcting accumulated errors with motion clues extracted from inertial measurements is proposed. In detail, direction of gravity and accelerometer noises are inferred, which are used in corrections of rotations and translations respectively. Proposed method is applied on real inertial datasets, and overall enhancements in inertial navigation were observed.