Data-driven robust UAV position estimation in GPS signal-challenged environment

TL;DR

The paper tackles real-time UAV position estimation in GPS-signal-challenged environments using IMU and GPS/barometer data. It introduces a data-driven robust estimation framework based on the robust extended Kalman filter (REKF), where the true transition density lies within an ambiguity set defined by conditional KL divergence with a learnable tolerance $c$. By selecting $\hat{c}$ from a finite set via data-driven prediction error, the approach yields a REKF recursion that remains robust to environmental uncertainties, demonstrated on a quadcopter with training/validation under GPS denial. Results show REKF outperforms the standard EKF during GPS-denied periods and rapidly realigns trajectories when GPS data are restored, offering improved reliability for low-cost UAV navigation in challenging environments.

Abstract

In this paper, we consider a position estimation problem for an unmanned aerial vehicle (UAV) equipped with both proprioceptive sensors, i.e. IMU, and exteroceptive sensors, i.e. GPS and a barometer. We propose a data-driven position estimation approach based on a robust estimator which takes into account that the UAV model is affected by uncertainties and thus it belongs to an ambiguity set. We propose an approach to learn this ambiguity set from the data.

Figures (5)

• Figure 1: Left. Test site in which the flight test has been performed. Right. The vehicle used for the flight test.
• Figure 2: The vehicle started at the black circle, moving counterclockwise, following the black path.
• Figure 3: Three dimensional trajectory plot for flight test in the ideal environment: Reference trajectory (black dashed line); Predicted position using EKF (blue line).
• Figure 4: Training with $\bar{S}=8s$ and validation with straight line with different GPS denial time $S$. Three dimensional trajectory plots: Reference trajectory (black dashed line); Predicted position using REKF (red line); Predicted position using EKF (blue line); GPS-barometer denied segment (between cyan crosses).
• Figure 5: Training with $\bar{S}=8s$ and Validation with turn with different GPS denial time $S$. Three dimensional trajectory plots: Reference trajectory (black dashed line); Predicted position using REKF (red line); Predicted position using EKF (blue line); GPS-barometer denied segment (between cyan crosses).