An Error Model of a Complementary Filter for use in Bayesian Estimation - The CF-EKF Filter

Abstract

It is well known that stand-alone inertial navigation systems (INS) have their errors diverging with time. Consequently an upper bound on the duration of INS systems precludes their use in low-cost micro unmanned aerial vehicles. The traditional approach for solving this matter is to resort to aiding devices, e.g., global navigation satellite system (GNSS) receivers, sighting devices, etc. Two philosophies have been extensively applied to perform data fusion: extended Kalman filtering (EKF) and complementary filtering (CF). Previous work in the literature showed that the computationally less expensive CF can be robustly applied to attitude estimation using low-cost sensors and achieve performance that is comparable to that of a full EKF. However, performance is degraded by vehicle manoeuvres and no measurement on estimate uncertainties is given. Furthermore, a large number of sensors makes it impracticable for optimal tuning of the CF. The present work lays the foundation for sensor filtering that employs the CF for attitude estimation by means of a magnetometer as an external aid, and an EKF for additional sensors integration. The main feature of this architecture is the possibility of deployment in a distributed multi-platform system and implementation of fault isolation by running the CF stage in a separate low-throughput reliable machine for stand-alone degraded mode operation. A case study is performed on synthetic data from inertial, magnetic and GNSS sensors.