Augmented Virtual Filter for Multiple IMU Navigation

Abstract

Navigation plays a vital role in autonomous surface and underwater platforms’ ability to complete their tasks. Most navigation systems employ a fusion between inertial sensors and other external sensors, such as global navigation satellite systems, when available, or a Doppler velocity log. In recent years, there has been increased interest in using multiple inertial measurement units to improve navigation accuracy and robustness. State of the art examples include the virtual inertial measurement unit (VIMU) and the federated extended Kalman filter (FEKF). However, each approach has its drawbacks. The VIMU does not improve sensor biases, which are significant sources of error in low-cost inertial sensors. While the FEKF improves accuracy, it models uncertainty propagation empirically. If not modeled correctly, this can cause the global solution to diverge. To cope with those shortcomings, we derive a novel filter structure as an extension of the VIMU approach for multiple inertial sensors data fusion: the augmented virtual filter (AVF). In addition, to cope with the multiple equal bias variance estimation difficulty, we developed the bias variance redistribution algorithm. Our filter design enables bias estimation for each inertial sensor in the system. This improves its accuracy and allows the use of a varying number of inertial sensors during a single run. We show that our AVF performs better than other state of the art, multiple inertial sensor filters using real data recorded during sea experiments. Lastly, an observability analysis is performed to demonstrate the system’s properties.

Topics & Concepts