Sequential Asteroid-Relative Visual Simultaneous Localization and Mapping Using Inverse Incremental Pose Augmentation

Optical navigation and mapping in the vicinity of a small celestial body can be cast as a form of the well-studied simultaneous localization and mapping (SLAM) problem of robotics. While incremental batch solutions have become the standard approach to landmark-based SLAM in many robotics applications, this work investigates a novel sequential filtering approach with an emphasis on computational efficiency that may be motivated by the unique challenges of the spaceflight domain. This new algorithm, dubbed inverse incremental pose augmentation (IIPA) SLAM, can generate an accurate solution for the state and landmarks at all image times while incorporating arbitrary nonlinear dynamics models with constant time complexity until it encounters a loop closure. A new inverse depth measurement model is derived, and the necessary dynamic models are presented to estimate the spacecraft state, asteroid-fixed landmarks, and other parameters of interest before applying IIPA-SLAM to both simulated and real data. An efficient Schmidt-based relocalization routine is also derived that mitigates the computational impact of loop closures.