Filippo Fonseca, Yale University; Owen Pinhasi, Yale University; Zachary Zitzewitz, Yale University
Keywords: sda, space-domain-awareness, uct, algorithm, algorithmic-development, uct-processing, benchmarking, performance, satellite, satellites, orbit, orbital-analysis, tracking
Abstract:
The rapid evolution of space activities—coupled with the contested nature of orbital environments—has amplified the importance of Space Domain Awareness (SDA). Central to this field is the processing of Uncorrelated Tracks (UCTs), which represent detected objects with incomplete or ambiguous orbital data. These tracks often arise from noisy or sparse sensor readings and are further complicated by adversarial tactics like camouflage, concealment, deception, and maneuver (CCDM). UCT processing is critical for identifying potential threats, mitigating collision risks, and ensuring safe and strategic operations in space. That being said, there exists an imminent danger that deplorably few have tackled: the SDA community currently lacks standardized frameworks for algorithmic evaluation (particularly in the realm of UCT processing), limiting progress in this vital area.
Thus, the research addresses these gaps by, as a primary resource, proposing a comprehensive benchmarking framework for UCT processing algorithms. This framework introduces novel datasets and metrics that enable consistent evaluation across varying operational scenarios, including cases of noisy, sparse, or adversarially manipulated data. The framework is designed to facilitate algorithmic innovation and collaboration within the SDA community by providing clear, reproducible methods for testing and comparison.
The methodological foundation of this work includes a thorough review of existing UCT processing literature and input from the Space Domain Awareness Track Assessment and Processing Laboratory (SDA TAP Lab). Existing UCT algorithms often fail to address discrepancies in input parameters or perform poorly under real-world conditions. Benchmarking datasets aim to simulate these challenges by introducing thousands of unique batches of UCTs with varying degrees of data sparsity and noise. These datasets enable researchers to assess algorithms using universal performance metrics, such as root mean square error and probabilistic correctness ratios.
Building upon these benchmarks, a novel UCT processing pipeline and algorithm is presented that integrates physics-informed modeling with modern computational techniques, such as multiple hypothesis tracking (MHT) and neural networks. Notably, this system is meticulously designed to work in conjunction with the standardization and performance benchmarks being proposed. Furthermore, the algorithm leverages classical physics principles—from Kepler’s Laws to Lambert solvers—to generate precise and informed initial orbital hypotheses, which are then refined using data-driven methods. Additionally, physics-informed neural networks contribute to the model’s robustness by incorporating hard constraints. The pipeline also employs recurrent neural networks and transformers for maneuver detection, enabling accurate trajectory predictions even under sparse observation conditions. The proposed system architecture follows an end-to-end approach, starting with raw sensor data ingestion and preprocessing. Adaptive filtering techniques address sensor noise, while compressive sensing and interpolation bridge observational gaps. Sensor fusion algorithms, such as Kalman filters, unify data from diverse sources into a cohesive framework. Object discrimination is achieved through probabilistic clustering and observational priors, with MHT algorithms resolving ambiguities in track associations. The final output includes a probabilistic object catalog with trajectory predictions and maneuver likelihoods, providing actionable insights for both human decision-makers and autonomous systems.
The expected outcomes of this research include the establishment of a replicable benchmarking framework, the development of improved UCT processing algorithms, and significant contributions to the SDA community. By standardizing evaluation criteria, the research to be presented will unleash more accurate and granular comparisons of existing and future algorithms. The novel algorithm developed in this study demonstrates the utility of the proposed framework, achieving consistently low error rates in UCT categorization and processing.
As touched upon previously, this research holds transformative potential for the SDA community. The introduction of benchmarking practices accelerates innovation, while the novel algorithm minimizes human intervention in UCT processing. These advancements align with the SDA community’s goals of enhancing space object tracking capabilities and operational readiness in contested environments; on top of that, focused and strategic collaboration with the U.S. Space Force’s SDA TAP LAB brings practical applicability, access to training data, real-life knowhow, and implementation expertise to the Yale SPRC’s contributions (bridging the gap between theoretical advancements and real-world implementation). Looking ahead, the aim remains to catalyze a movement within the SDA community toward universal adoption of standardized benchmarking and performance evaluation practices. Though particularly pertinent within the context and scope of UCT processing algorithms (as is the focus of the technical paper), such a principle applies to a plethora of aspects within the SSA/SDA space—from data fusion to collision avoidance algorithms and systems, the applications are vast, and the possibilities bright. Setting a precedent for rigorous and reproducible research in this rapidly evolving field is not a quote-on-quote “nice-to-have” ambition, but a fundamental necessity that must be addressed with the utmost urgency and conviction.
Date of Conference: September 16-19, 2025
Track: Astrodynamics