The Variational Attitude Estimator in the Presence of Bias in Angular Velocity Measurements

TL;DR

This paper extends the variational attitude estimator to handle biased angular velocity measurements, ensuring almost global convergence of state and bias estimates despite measurement uncertainties.

Contribution

It introduces a generalized variational attitude estimator that accounts for constant bias in angular velocity measurements, improving stability and convergence.

Findings

State estimates converge almost globally to true states.

Bias estimates converge to true bias after state convergence.

Estimator remains stable despite measurement noise and bias.

Abstract

Estimation of rigid body attitude motion is a long-standing problem of interest in several applications. This problem is challenging primarily because rigid body motion is described by nonlinear dynamics and the state space is nonlinear. The extended Kalman filter and its several variants have remained the standard and most commonly used schemes for attitude estimation over the last several decades. These schemes are obtained as approximate solutions to the nonlinear optimal filtering problem. However, these approximate or near optimal solutions may not give stable estimation schemes in general. The variational attitude estimator was introduced recently to fill this gap in stable estimation of arbitrary rigid body attitude motion in the presence of uncertainties in initial state and unknown measurement noise. This estimator is obtained by applying the Lagrange-d'Alembert principle of variational mechanics to a Lagrangian constructed from residuals between measurements and state estimates with a dissipation term that is linear in the angular velocity measurement residual. In this work, the variational attitude estimator is generalized to include angular velocity measurements that have a constant bias in addition to measurement noise. The state estimates converge to true states almost globally over the state space. Further, the bias estimates converge to the true bias once the state estimates converge to the true states.