Connor Segal, EO Solutions; Cameron Harris, EO Solutions; Jonathan Kadan, SSC/SYD 85/BI – Odyssey
https://doi.org/10.64861/JVKV9210
Keywords: Space Traffic Management, Uncorrelated Track, UCT, Autonomous, Sensor Orchestration, MACHINA
Abstract:
At the current rate of activity in space, the global space community requires a transparent and responsive approach to Space Traffic Management (STM). In 2024, we presented the MACHINA-SDA agent framework, demonstrating a live system fulfilling open-loop sensing objectives through a collaborative sensor network. This work marks the critical next step: the maturation of MACHINA into a fully autonomous, closed-loop agent for modern STM, validated using the Decisive Mission Analysis Capability (DMAC), a MACHINA-SDA agent deployed within a high-fidelity simulation environment referred to as MACHINA-sim.
Closed-loop autonomy enables critical STM functions: maintaining custody of objects, characterizing unexpected maneuvers, and assessing conjunction risk; all without manual intervention. When an unexpected maneuver occurs and follow-up observations fail to correlate a Resident Space Object’s (RSO) estimated state to an associated tracklet, an Un-Correlated Track (UCT) is created. Origination is a subset of the broader UCT problem; it begins immediately upon UCT creation and adds the goal of defining and attributing the maneuver responsible for the observed state change, rather than only reconnecting the UCT to its original RSO.
A MACHINA-SDA agent produces its own observation data and directly ingests third-party data from the Unified Data Library (UDL), including maneuver detection notifications, missing-object alerts, and UCTs to autonomously trigger follow-up actions. This closed-loop capability, Tactical Persistent Monitoring (TPM), pulls current RSO states and event notifications from the UDL to produce state estimates from the most up-to-date information available. Within MACHINA-sim, the same UDL connection is replicated with a mock UDL initialized from real UDL data at the simulation epoch, into which uncued notifications, alerts, and warnings are injected to directly test MACHINA’s autonomous reactions.
The UCT origination problem is the focus of this paper. UCTs created within MACHINA, or ingested from the UDL, undergo a three-stage process running concurrently with TPM: Track, where newly created UCTs are continuously observed to produce an initial orbit determination (IOD) solution, updated as subsequent observations are collected; Triage, where UCTs and their missing RSO counterparts are prioritized and acted upon to correlate each UCT to its originating RSO; and Graduation, where a UCT is elevated as the newly correlated RSO state following maneuver attribution.
TPM itself comprises two action sequences, each launched by its own event trigger. By default, all RSO targets are observed at a nominal revisit rate set by orbital regime and orbit parameters, continuing until a maneuver detection or a prescribed number of consecutive missed observations occurs. A maneuver detection triggers an increased revisit rate; a missed-observation event triggers a wide area search (WAS) around the target’s estimated position. A maneuver detection occurs when an observation’s residuals indicate a maneuver has likely occurred, but the object remains in the sensor’s field of view and correlates at the next observation. Follow-on collections either reconverge the orbit determination (OD) solution, returning the network to nominal revisit, or the RSO drifts far enough that correlation fails and a UCT is created. Consecutive missed observations instead describe the case where the RSO has already drifted out of the expected frame by the next scheduled collection, so any later observation creates a new UCT.
After creation, TPM’s bookkeeping at the moment of discovery determines what context the UCT carries forward: if exactly one RSO is under maneuver-reaction or custody-recovery tasking, that RSO travels with the UCT as a suspected candidate; if none or more than one is anomalous, no candidate is attributed, since resolving between several plausible originators is a correlation question TPM alone is not positioned to answer. Origination then collects on the UCT at a higher-than-nominal revisit rate to build its own OD solution, and each new observation triggers a single evidence-loop pass reasoning in two stages. When a suspected candidate is known, the loop first tests that hypothesis alone with a one-shot check against a conservative confidence floor, stricter than the multi-candidate gate used elsewhere, since a lone hypothesis has no runner-up to weigh against it. If the candidate clears the floor after at most one forced ephemeris refresh, the loop proceeds toward graduation. If not, the same pass falls through, without waiting for the next observation, to a full-catalog test that refreshes ephemeris for every catalog object within line-of-sight and re-runs the significance test against every visible candidate simultaneously. Only after both checks fail does the UCT exhaust its Triage options and escalate to active tasking, which either re-observes the suspected candidate directly (Path 1) or re-tasks a proximity-and-age-ranked population of catalog RSOs (Path 2), where either can resolve the case through direct re-observation or a missed-observation signal that is itself evidence of the connection.
When a confident match is produced, one further check runs before graduation: if a runner-up candidate cleared the gate closely enough with the winning candidate to represent genuine ambiguity, graduation is held and the UCT is instead routed to a targeted disambiguation state, where a single, precisely-timed re-observation is tasked to maximize the worst-case angular separation among all ambiguous candidates; a narrowly scoped exception to the system’s otherwise evidence-first, non-blocking posture, justified to directly add a preventative measure against cross-tagging. Once a correlation is unambiguous, the UCT’s observation history is reattributed to the confirmed RSO and custody reverts to TPM. This process is demonstrated through a vignette recreated from publicly available observation data of a real-world rendezvous between SS-21 and SJ-25 that occurred in June 2025.
By advancing from open-loop scheduling to a validated, closed-loop agent, MACHINA provides a scalable software architecture for the long-term sustainability of space. MACHINA-sim offers a powerful capability for verifying MACHINA’s autonomous sensor orchestration against the collaborative, dynamic demands of modern space operations, and its demonstrated effectiveness in addressing the UCT origination problem provides direct value to current operational challenges.
Date of Conference: September 15-18, 2026
Track: Astrodynamics