A Computationally Efficient Moving Horizon Estimator for Ultra-Wideband Localization on Small Quadrotors

Abstract

We present a computationally efficient moving horizon estimator that allows for real-time localization using Ultra-Wideband measurements on small quadrotors. The estimator uses only a single iteration of a simple gradient descent method to optimize the state estimate based on past measurements, while using random sample consensus to reject outliers. We compare our algorithm to a state-of-the-art Extended Kalman Filter and show its advantages when dealing with heavy-tailed noise, which is frequently encountered in Ultra-Wideband ranging. Furthermore, we analyze the algorithm's performance when reducing the number of beacons for measurements and we implement the code on a 30 g Crazyflie drone, to show its ability to run on computationally limited devices.