Velocity-Aided IMU-Based Tilt and Attitude Estimation

Abstract

This paper addresses the problem of estimating the tilt and more generally the attitude of a rigid body that is subject to high accelerations and equipped with inertial measurement units (IMU) and a sensor providing the body velocity (expressed in the reference frame attached to the body). In the absence of a magnetometer, tilt estimation is proposed through a two-step observer: a global-exponentially stable pre-estimation given to a manifold-constrained complementary filter. In the presence of a magnetometer, the presented observer allows to reconstruct the full attitude and tune how much the estimation of the tilt is influenced by the magnetometer, depending on the level of confidence given to the measurements of the magnetometer. All state estimators are proposed with proofs of almost global asymptotic stability and local exponential convergence. Finally, these estimators are compared with state-of-the-art solutions in clean and noisy simulations, allowing recommended solutions to be drawn for each case.