Quaternion-Based Robust Attitude Estimation Using an Adaptive Unscented Kalman Filter

Abstract

This paper presents the Quaternion-based Robust Adaptive Unscented Kalman Filter (QRAUKF) for attitude estimation. The proposed methodology modifies and extends the standard UKF equations to consistently accommodate the non-Euclidean algebra of unit quaternions and to add robustness to fast and slow variations in the measurement uncertainty. To deal with slow time-varying perturbations in the sensors, an adaptive strategy based on covariance matching that tunes the measurement covariance matrix online is used. Additionally, an outlier detector algorithm is adopted to identify abrupt changes in the UKF innovation, thus rejecting fast perturbations. Adaptation and outlier detection make the proposed algorithm robust to fast and slow perturbations such as external magnetic field interference and linear accelerations. Comparative experimental results that use an industrial manipulator robot as ground truth suggest that our method overcomes a trusted commercial solution and other widely used open source algorithms found in the literature.

A Novel Fuzzy-Adaptive Extended Kalman Filter for Real-Time Attitude Estimation of Mobile Robots

“…Recent studies have proposed the use of unscented KF (UKF) over EKF [52,53], and stated that UKF-based approaches better deal with the high-order nonlinear terms of large attitude errors. Attitude estimation has been solved with computationally efficient geometric UKF [53], where a new formulation of the UKF algorithm was proposed in [52] to maintain fast and slow variations in the measurement uncertainty. The latter algorithm was augmented with both an adaptive strategy to tune the covariance matrices on-the-fly and an outlier detector to reject the effects of external disturbances.…”

Fast AHRS Filter for Accelerometer, Magnetometer, and Gyroscope Combination with Separated Sensor Corrections

“…It is a classical approach [ 10 ] used in the human motion domain with a large number of variants, see e.g., [ 11 ] for their comparison. Various extensions focus on different aspects of the filtering differing such as the effect of nonlinear models that is addressed by the more computationally costly unscented Kalman Filter [ 12 ]. However, the limiting factor of Kalman-based filters is the underlying assumption of Gaussian distributed disturbance.…”

“…This has been addressed by extensions proposing adaptive estimates of the covariance matrices, see e.g., [ 13 ] for an excellent review. The approaches are based on various approaches monitoring the evolution of estimation errors using fuzzy rules [ 13 ], moving average covariance estimation [ 12 ], Markov chains [ 14 ], or segmented moving average covariance estimators [ 15 ] to name a few examples. The performance benefit of these methods usually comes with a significant computational price, making it unsuitable for low cost, low-power applications.…”

A Framework of Cubature-Hᵢ/H∞-Fault Detection and Robust H-Infinity Kalman Filter of Ship SINS/GNSS Integrated System

“…For example, adaption methods based on innovation and residual [27]- [29] have been used to adapt the measurement covariance noise matrices. Covariance is another parameter used for adapting both the process and measurement noise [30], [31]. However, all examples in the literature are based on innovation sequences, which may cause measurement noise or other covariances to be negative for the subtraction.…”