Computationally Efficient Attitude Estimation with Extended $\mathcal{H}_2$ Filtering

TL;DR

This paper introduces an extended H2 filter for UAV attitude estimation that outperforms the traditional EKF in computational efficiency, memory usage, and accuracy, especially suited for low-cost MEMS sensors.

Contribution

The paper proposes a novel extended H2 filter for attitude estimation, formulated with unit quaternions, offering improvements over EKF in speed, memory, and error.

Findings

Outperforms EKF in computational speed and memory efficiency.

Achieves lower root mean squared error in attitude estimation.

Effective across various UAV flight scenarios.

Abstract

Accurate state estimation using low-cost MEMS (Micro Electro- Mechanical Systems) sensors present on Commercial-off-the-shelf (COTS) drones is a challenging problem. Most UAV systems use a combination of a gyroscope, an accelerometer, and a magnetometer to obtain measurements and estimate attitude. Under this paradigm of sensor fusion, the Extended Kalman Filter (EKF) is the most popular algorithm for attitude estimation in UAVs. In this work, we propose a novel estimation technique called extended H2 filter that can overcome the limitations of the EKF, specifically with respect to computational speed, memory usage, and root mean squared error. We formulate our attitude-estimation algorithm using unit quaternions. The H2 optimal filter gain is designed offline about a nominal operating point by solving a convex optimization problem, and the filter dynamics is implemented using the nonlinear system dynamics. This implementation of this H2 optimal estimator is referred as the extended H2 estimator. The proposed technique is tested on four cases corresponding to long time-scale motion, fast time-scale motion, transition from hover to forward flight for VTOL aircrafts, and an entire flight cycle (from take-off to landing). Its results are compared against that of the EKF in terms of the aforementioned performance metrics.