Dual-Robust Filter Boost UAV Navigation Under Extreme Uncertainty

Accurate position estimation is a critical requirement for the reliable operation of unmanned aerial vehicles (UAVs); however, conventional filtering techniques often degrade severely when faced with the simultaneous occurrence of model mismatch and non-Gaussian measurement disturbances. To overcome these limitations, researchers have proposed a novel dual-robust unscented Kalman filter, designated H∞-GMMC-UKF. This advanced framework integrates two distinct robust mechanisms: an H∞-motivated modification to the prior covariance and a generalized maximum mixture correntropy (GMMC) measurement update. This architecture is specifically designed to enhance the resilience of UAV state estimation under the challenging operating conditions where both the internal model and external sensor data are subject to significant uncertainty.

The core innovation of the H∞-GMMC-UKF lies in its complementary design. The H∞-motivated prior-covariance modification serves to improve the filter’s tolerance to prediction errors stemming from inherent model uncertainty. In essence, this modification provides a robust margin on the prediction side, allowing the filter to better handle discrepancies between the mathematical model and the real-world dynamics of the UAV. Concurrently, the GMMC measurement update addresses the issues arising from sensor noise. This component is engineered to reduce the detrimental influence of asymmetric, heavy-tailed, and impulsive measurement errors, which are common in real-world flight scenarios. The combination of these two robust stages ensures that the filter maintains high performance even when faced with complex, multi-faceted uncertainties.

The effectiveness of the proposed filter was rigorously evaluated through a comprehensive testing regimen. The researchers conducted four progressive simulation scenarios, alongside two challenging real-world stress tests using the EuRoC dataset. Each test involved 100 paired Monte Carlo trials, incorporating detailed ablation and sensitivity analyses. The results demonstrated that while the filter remains competitive under benign conditions, its most significant improvements are realized when model and measurement uncertainties coexist. Specifically, in controlled dual-uncertainty simulations, the mean position Root Mean Square Error (RMSE) was successfully reduced by 18.0%. Furthermore, when tested against contaminated EuRoC stress tests, the proposed filter achieved a substantial 57.3% reduction in error.

Detailed analysis confirmed that the two robust stages contribute through distinct and complementary mechanisms. The GMMC update directly attenuates measurements that have been contaminated by outliers or heavy-tailed noise. Simultaneously, the H∞ covariance modification provides an essential prediction-side robustness margin. The sensitivity analysis revealed that the effect of this H∞ modification is most pronounced in the tested adverse measurement-update settings, highlighting its critical role in maintaining stability and accuracy when the measurement update process itself is compromised. These findings collectively demonstrate the strong potential of the H∞-GMMC-UKF architecture for achieving reliable and precise state estimation for UAV applications operating in highly unpredictable and uncertain environments.

Subject of Research: Development and testing of a dual-robust unscented Kalman filter for reliable state estimation in unmanned aerial vehicles facing simultaneous model and measurement uncertainties.

Article Title: Dual-robust unscented Kalman filter for UAV state estimation under simultaneous model and measurement uncertainties

Article References: Liu, J., Zhao, X., Tian, J., Jiao, Z., Wang, Y., & Zou, X. (2026). Dual-robust unscented Kalman filter for UAV state estimation under simultaneous model and measurement uncertainties. Scientific Reports. https://doi.org/10.1038/s41598-026-77396-7

Image Credits: AI Generated

DOI: 10.1038/s41598-026-77396-7

Keywords: UAV, Kalman Filter, State Estimation, Robust Control, H-infinity, Sensor Fusion, Unmanned Aerial Vehicles, Control Systems, Machine Learning, GMMC, Dual-robust, unscented