Computationally Efficient Robust Information Filtering for In-Flight GNSS/SINS Tightly Coupled Navigation with High-Dimensional Observations on Small UAVs

The full operation of BDS-3 enables users to obtain high-performance positioning services, benefiting from the surge in the number of Global Navigation Satellite System (GNSS) observations with multi-constellation multi-frequency signals. This overabundance is beneficial to improve in-flight navigation accuracy for small unmanned aerial vehicles (UAVs). However, it brings about two-fold challenges for conventional airborne GNSS/SINS tightly coupled (TC) systems. On one hand, limited airborne computing resources suffer from the “curse of dimensionality” caused by extremely high-dimensional GNSS observations (i.e., GNSS pseudo-ranges, pseudo-range rates, and time-differenced carrier phases from multi-system and multi-frequency, such as GPS L1/L2 and BDS B1/B2/B3, totaling up to over 100 observables per epoch), leading to increased calculation burden and potential latency. On the other hand, possible outliers can degrade the obtained navigation accuracy. To enhance overall performance, this paper proposes a computationally efficient Kalman filtering framework for tight integration between airborne GNSS and SINS via a high-dimensional robust information filter. The strategy of kinematic and static information filtering is utilized to handle the matrix inversion complexity caused by high-rate and high-dimensional Kalman measurement updates, and the technique of robust adaptive factor is used to resist the adverse effects of GNSS outliers and modeling errors. Both land vehicular and UAV flight tests indicate that the proposed algorithm outperforms its traditional TC counterparts, demonstrating an over 90% improvement in overall computational efficiency without any loss in accuracy, compared with conventional batch or sequential tightly coupled Kalman filtering.