Reducing Computational Cost in the Invariant Unscented Kalman Filtering For Attitude Estimation

Jean-Philippe Condomines (1) , Gautier Hattenberger (1)

1 ENAC - Ecole Nationale de l'Aviation Civile

This article proposes a new formulation to derive the invariant unscented Kalman filter (IUKF) algorithm for attitude estimation problem, where both state and sigma-point are considered as a transformation group parametrization of the filter. The detailed IUKF equations are presented in this article. The filter equations relie on the same ideas as the usual Unscented Kalman Filter (UKF), but it uses a geometrically adapted correction term based on an invariant output error. The specific interest of the proposed formulation is that only the invariant state estimation errors between the predicted state and each sigma point must be known to determine the predicted outputs errors. As we have already computed the set of invariant state errors during the prediction step, the computation cost to find the covariance matrix of the invariant state estimation in the update step is greatly reduced.

Mots clés: unscented Kalman filters (UKF); invariant fil- tering; attitude estimation; extended Kalman filter (EKF)

Domaines: Automatique / Robotique

Jean-Philippe Condomines, Gautier Hattenberger. Reducing Computational Cost in the Invariant Unscented Kalman Filtering For Attitude Estimation. 2019. ⟨hal-02106315v2⟩