Relative Visual-Inertial Odometry for Fixed-Wing Aircraft in GPS-Denied Environments

This paper introduces an odometry-like front-end estimator for GPS-denied fixed-wing flight using a monocular camera. This development is a critical component for enabling fixed-wing aircraft to use a novel methodology called relative navigation. Relative navigation allows a single vehicle to gracefully accommodate intermittent GPS or geo-registered image updates. It can also enable communication and a computationally feasible cooperative, batch-based navigation approach for numerous vehicles. This paper presents a visual-odometry estimator that is appropriate for fixed-wing flight characteristics and sensing requirements. We propose a modification to the multi-state-constraint Kalman filter which, instead of running the filter in a single inertial/global frame, regularly resets the navigation frame to a new local origin. The vehicle operates based on the most recent keyframe and produces a marginalized change in pose (odometry) and covariance output whenever the keyframe is updated. This allows the front end to operate at the IMU rate to accommodate guidance and control inputs while sharing marginalized keyframe-to-keyframe odometry estimates with a back-end, graph optimization at a much lower rate than traditional batch navigation schemes. Results from testing the proposed method in a high-fidelity simulation of the filter is also presented.