Identifiability-Aware Virtual Force Sensing and Self-Calibration for Quadrotors
Journal
IEEE Access
ISSN
21693536
Date Issued
2026
Author(s)
Quito, Angelica
Andaluz, Victor H.
Brandao, Alexandre Santos
Type
Article
Abstract
Many aerial tasks, such as flying in wind or carrying a payload, require the external force on the airframe, yet quadrotors rarely carry a force sensor. We turn the onboard inertial measurement unit into a virtual force sensor: the external specific force is read directly from the accelerometer residual. The obstacle is calibration, since the reading needs the vehicle mass, which is unidentifiable near hover, where a mass error is indistinguishable from a vertical force; a standard extended Kalman filter then corrupts the mass. We make the sensor identifiability-aware and self-calibrating. An online identifiability measure, the Fisher information on the mass, gates the calibration and freezes the mass when it cannot be seen; and when the flight lacks the excitation to recalibrate, the controller injects a minimal probe that the same measure shapes. In simulation the sensor reads the external force with a per-axis correlation of 0.93 under continuous wind; the gate holds the mass calibration to 3 % through a windy hover where an ungated filter drifts to 7 %; the probe recovers the mass from a wrong prior that hover alone cannot identify; and feeding the sensed force to a model-predictive controller reduces the tracking error by 34 to 80 % under wind, most where the wind dominates the dynamics. The sensor degrades gracefully under injected accelerometer and thrust errors and runs in real time at 100 Hz. © 2013 IEEE.