Relative Navigation and Mapping using Visual Point Clouds & Neural View Synthesis

In response to the 2025 NASA SBIR Phase I solicitation subtopic H9.03, Advanced Space, LLC (Advanced Space) and partner The University of Colorado, Boulder (CU Boulder) propose to develop through trade studies an RPOD-focused, monocular optical navigation and mapping pipeline for unknown, uncooperative space objects that is robust to the harsh lighting conditions of space. The first algorithm in the pipeline is CloudNav, a new relative pose estimation solution that uses visual point cloud triangulation and matching, which was developed for use in asteroid-relative navigation and mapping. The team at CU Boulder invented the CloudNav algorithm with funding from the NASA JPL SURP Program. Our intent is to leverage this preexisting relationship between CU Boulder, JPL, and Advanced Space to collaborate on the maturation of CloudNav. These same techniques will be applied to RPOD datasets to generate relative pose observables that then can be passed to a specialized Extended Kalman Filter (EKF) implementation to process dynamical constraints, additional parameters, and possibly additional measurements to enhance state observability. The combination of CloudNav measurements with the high-heritage, mature EKF represents a combination of algorithms for relative navigation that is likely amenable to real-time, onboard implementation in future phases of this effort. Finally, with a robust relative navigation solution in combination with the input camera images, the Advanced Space team proposes to postprocess these outputs in a set of identified neural view synthesis (NVS) algorithms to obtain dense, relightable, photorealistic object reconstructions that could aid investigators in interpreting target object characteristics and even material properties. Advanced Space has already invested in an IRAD project to demonstrate that publicly available NVS algorithms can be trained to produce strikingly accurate reconstructions of asteroid shapes given a set of input images and poses.