Matthew Popplewell, Advanced Space; Matthew Givens, Advanced Space; Frank Centinello, Advanced Space; Michael Caudill, Advanced Space; Benjamin Tatman, Advanced Space; Nathan Parrish Ré, Advanced Space
Keywords: orbit determination, space debris, astrodynamics, optical, radar
Abstract:
With the renewed interest in space activity over the last two decades and the consequential surge in launches, the debris population has grown significantly. The increased reliance on space assets for defense, commercial, and civil applications has bolstered this growth and will be a driving factor for continued debris population growth. Awareness of the debris population is crucial not only for current space operations but also for deterring the growth of future space debris populations. However, the smallest objects capable of regular tracking with current industry and Space Surveillance Network (SSN) capabilities are approximately 10 cm in LEO, which leaves a significant portion of the debris population undetected and untracked. The European Space Agency estimates only 36,860 space objects, less than 1% of all debris, are regularly maintained in the SSN catalog.
The Intelligence Advanced Research Projects Activity’s (IARPA) Space Debris Identification and Tracking (SINTRA) program is one of several ongoing efforts to better track and characterize the small debris population. Recent advances in sensor hardware and observation processing technologies (such as machine learning) are enabling the detection of small debris in Low Earth Orbit (LEO) via multiple measurement phenomenologies. However, due to constraints such as restricted observer-object geometries, high relative velocities, inconsistent lighting conditions, and the limited number of available observational platforms, these small debris objects are often only detected over a short time span. The resultant measurement sets can be sparse and span a small section of the orbit geometry. These “Too Short Arcs” are insufficient to generate an accurate initial orbit estimate, even when using measurements with little to no noise. Incorporating realistic measurement noise further exacerbates the problem.
Recent literature has explored the use of probabilistic methods, or admissible regions (AR), to initialize navigation filters for the Too Short Arc problem. DeMars and Jah introduce the Gaussian Mixture Model, Unscented Kalman Filter (GMM-UKF), in which the AR distribution serves as an initial guess for an UKF, which refines the admissible set as additional observations become available [1]. The authors show promising results for a geostationary (GEO) target in a simplified dynamical model.
Building on these prior efforts, we first provide a discussion of the state of the art in orbit determination with Too Short Arcs. Then, a new technique that incorporates measurement uncertainty into the probabilistic solutions is presented. The AR is constructed using radius of periapsis and apoapsis constraints, derived from the observed visual magnitude of the object. The resultant distribution serves as an initial guess for an UKF which then iteratively improves upon the GMM. Attempting to apply the baseline GMM-UKF to LEO objects revealed that many of GMM component weights quickly go to zero and do not contribute to the solution. Thus, a key contribution in this work is the incorporation of GMM splitting and pruning to reduce degeneracy within the filter. The filter output is a GMM that represents the set of admissible orbit states. This set is subsequently used for measurement and track correlation, which relates tracks across observation gaps and enables a traditional batch filter to iterate to a more precise orbit determination solution.
The primary contribution of this work is the application of the introduced pipeline to real optical and radar data. Much of the literature applies the same concepts to simplified simulated scenarios. As part of the SINTRA program, the authors have access to optical collects of resident space objects from ExoAnalytic Solutions, MIT Lincoln Lab, and the VADeR observatory at CU Boulder, and radar collects from LeoLabs, MIT Lincoln Lab, and the ALTAIR sensor. The full orbit determination pipeline, from raw measurement to final CCSDS Orbit Data Message, is applied to the real sensor collects. The results are presented in this study and a short discussion follows. Overall, the combined AR-UKF method enables the tracking of space debris despite limited observability and nonlinear dynamics. Applied to real observations of LEO objects, the filter demonstrates robust initialization and refinement of admissible orbit states, supporting track correlation and subsequent orbit determination.
[1] K. J. DeMars and M. K. Jah, “Probabilistic Initial Orbit Determination Using Gaussian Mixture Models,” J. Guid. Control Dyn., vol. 36, no. 5, pp. 1324–1335, Sep. 2013, doi: 10.2514/1.59844.
Date of Conference: September 16-19, 2025
Track: Space Debris