A fault tolerant attitude estimation architecture with GLR-based disturbance rejection

This paper presents an architecture for fault tolerant attitude estimation systematically handling common attitude sensor faults and perturbations. The solution is posed as a filtering architecture composed of four distinct parts: The first stage serves to estimate the sensor outputs and provide fault sensitive residuals. The second stage is a fault detection and rejection stage based on $\chi^{2}$ - and Generalized Likelihood Ratio (GLR) tests. The third stage consolidates the measurements and activates or deactivates the attitude correction. The fourth and final stage is an Invariant Extended Kalman filter (IEKF) with a saturated bias model and local decoupling of the inclination and heading estimation. A series of flight tests with a multi-rotor drone subject to severe magnetic perturbations and inertial accelerations shows the resilience of the attitude estimation to sensor faults as compared to classical approaches.