Deep Fusion of a Skewed Redundant Magnetic and Inertial Sensor for Heading State Estimation in a Saturated Indoor Environment

Abstract

Robust attitude and heading estimation in an indoor environment with respect to a known reference are essential components for various robotic applications. Affordable Attitude and Heading Reference Systems (AHRS) are typically using low-cost solid-state MEMS-based sensors. The precision of heading estimation on such a system is typically degraded due to the encountered drift from the gyro measurements and distortions of the Earth’s magnetic field sensing. This paper presents a novel approach for robust indoor heading estimation based on skewed redundant inertial and magnetic sensors. Recurrent Neural Network-based (RNN) fusion is used to perform robust heading estimation with the ability to compensate for the external magnetic field anomalies. We use our previously described correlation-based filter model for preprocessing the data and for empowering perturbation mitigation. Our experimental results show that the proposed scheme is able to successfully mitigate the anomalies in the saturated indoor environment and achieve a Root-Mean-Square Error of less than [Formula: see text] for long-term use.