TRACER: A Robust and Autonomous Framework for Angles-Only Orbit Determination

Orbit determination from optical observations remains a challenging problem due to the absence of direct range measurements and the presence of sparse, noisy, and irregularly sampled data. This work presents TRACER (Tracking, Recognition, and Analysis for Celestial Ephemerides Retrieval), a robust and fully automated framework for angles-only orbit determination. The proposed approach integrates probabilistic and deterministic strategies within a unified, decision-driven architecture. In particular, statistical ranging is employed for short-arc regimes to explore admissible solutions, while deterministic methods, including modified Gauss and Väisälä techniques, are used for longer arcs and refinement. Candidate solutions are evaluated through a unified scoring function that combines observational consistency with physically motivated penalties. A key contribution of TRACER is the introduction of a randomized subset-selection outer loop, which repeatedly solves the orbit determination problem on different observation subsets and validates solutions against the full dataset, enhancing robustness in challenging scenarios. Additional mechanisms for adaptive subarc selection, recovery from failure, and progressive data assimilation further improve reliability. The resulting framework enables fully autonomous orbit determination without manual intervention, bridging the gap between individual algorithms and operational pipelines for real-world astrometric data processing.