Observability Analysis of Heading Aided INS for a Maneuvering AUV

Abstract

Underwater navigation of autonomous underwater vehicles (AUVs) is a challenging task that requires the fusion of multiple sensors used as aiding to the vehicle inertial navigation system. In this paper, we focus on the problem of fusing heading measurements under different maneuvering conditions. We analyze the observability of the heading measurement fusion problem and derive the observable and unobservable subspaces as a function of the AUV's maneuver. Using this analysis, we are able to predict the convergence of the error states for different maneuvers and thus identify the quality of the navigation solution at different mobility patterns. As expected, the results show that the down misalignment error and gyro z-axis bias are always observable. In addition, our results showed that for some maneuvers, the rest of the gyro biases also become observable. Our numerical simulation and sea experiments verify our analytic solutions.

A Learning-Based Approach for Bias Elimination in Low-Cost Gyroscopes

“…In 2010, Iozan, and Collin, reported achieving 1-degree accuracy in determining true North orientation with a low-cost MEMS gyro, while accounting for small errors such as g-sensitivity and cross-axis coupling [31,32]. Among other works over the past decade, Ali (2011) presented a second-order divided difference filter (DDF) [33], Du (2016) introduced a disturbance observer-based Kalman filter (DOBKF) [34], and Klein (2018) demonstrated how kinematic constraints, when coupled with appropriate observability analysis, can effectively enhance unobserved states [35][36][37].…”

A novel approach for aiding unscented Kalman filter for bridging GNSS outages in integrated navigation systems

“…The inaccurate description of the system noises, measurement errors, and uncertainty in the dynamic models lead to unreliable estimates and degradation in accuracy, especially during GNSS outages when KF operates in prediction mode based on the predefined state error models, which are not necessarily correct. In addition, there are several significant drawbacks of KF, such as sensor dependency and observability problems (Hong et al, 2005;Klein & Diamant, 2018;Tang et al, 2008).…”