Gaussian Process Sequential Filtering for Small Body SLAM with Silhouette-Based Measurements

A Gaussian Process Sequential Filter (GPSF) is developed for Simultaneous Localization and Mapping (SLAM). The small bodies are modeled using a Gaussian Process (GP), where multiple basis nodes and radii are used to predict the entire shape, greatly reducing computation time for onboard, autonomous implementation. The GPSF processes silhouette-based measurements and associates the data to the estimated GP model. The GPSF simultaneously estimates the body’s shape, orientation, and spin, along with the spacecraft’s relative position and velocity. The formulation of the GPSF is provided, including the root-solved analytic partials, measurement underweighting techniques, and covariance inflation methods. The performance of the GPSF is highlighted with a Monte Carlo study about multiple small bodies, including Lutetia, Eros, Toutatis, and Bennu.