Timothy Olson, Slingshot Aerospace; Clarice Reid, Slingshot Aerospace; Jason Stauch, Slingshot Aerospace; Conner Grey, Slingshot Aerospace; Dylan Kesler, Slingshot Aerospace; Belinda Marchand, Slingshot Aerospace
Keywords: Conjunction Assessment, Space Situational Awareness, Space Domain Awareness, Space Traffic Control, Machine Learning, Predictive Analytics, Sensor Tasking, Probability Dilution, Space Operations Optimization
Abstract:
Satellite operators regularly assess conjunction risks and weigh trade-offs between mission operations and conjunction avoidance actions in the increasingly congested orbital environment. As a conjunction event approaches, operators typically receive a sequence of conjunction data messages (CDMs) from the US Space Force’s 18th Space Defense Squadron. The CDMs identify the objects involved in the conjunction, report a measure of risk and margins for the event, and estimate uncertainty in the form of state covariance matrices. Operators then have a choice of attempting to acquire additional data to refine conjunction information, or use the information in the CDMs to determine whether to take action. The methods presented in this paper aim to facilitate operators’ decision-making processes by identifying which conjunction events are likely to benefit from additional data acquisition well in advance of when they must commit to maneuvers.
A decision to maneuver for collision avoidance implies costs and risks to operators, including fuel consumption, increased wear on spacecraft subsystems, and overall reduced operational lifetime, in addition to interruptions in mission execution. It is therefore essential to obtain reliable information about each conjunction in order to make data-driven decisions regarding upcoming conjunction events. In particular, an informed decision requires a sufficiently small combined covariance estimate at the maneuver commit point (MCP). Otherwise if the covariance is too large at the MCP, the metrics used to evaluate the risk of a collision including the miss distance and probability of collision (Pc) may not be reliable indicators of the true risk. Without remedy, the operator may be left without enough detail to make an informed decision, as discussed extensively in the “dilution region” literature.
There are several approaches to reduce covariance. One approach is simply to wait until the conjunction is closer in time, and allow additional observations to be collected. As new observations are processed, the propagation duration between the most recent observation and the time of closest approach (TCA) will shorten, thereby decreasing the covariance expansion. However, variability in tracking cadence may lead to insufficient covariance reduction by the MCP. Thus, an important second option is to decrease the uncertainty by sourcing additional observations that would not have otherwise been collected. These are usually on the secondary resident space object (RSO) because it tends to be the object with larger covariance, and often the primary has precise ephemeris information available.
Collecting additional observations prior to the MCP is not always feasible for every potential secondary object in a large candidate set, especially as the number of objects in orbit continues to expand. It is therefore paramount to identify conjunctions that would benefit from additional tracking as early as possible, so that additional observations may be collected, fused, and delivered in time to enable informed decisions. This scaling challenge motivates the primary contribution of this paper – a machine learning based model that accounts for the temporal evolution of CDM sequences and predicts, in advance, the conjunctions for which the covariance will not decline enough via routine tracking prior to the MCP. The model also incorporates important contextual information about the objects involved in the conjunction, including physical characteristics and pattern of life details which enable it to outperform baseline covariance-only approaches. Our model has demonstrated reliable performance, identifying high-covariance CDMs up to five days before TCA, which equips the operators with sufficient advance notice to request observations and obtain refined state estimates with enough time to incorporate the enhanced information into their collision avoidance maneuver decision. By providing a predictive prioritization of upcoming conjunctions for the purpose of data acquisition, this approach could be integrated into tasking or planning systems and scaled to support conjunction assessment processes, and improve trust and responsible decision-making for the growing spaceflight industry.
Date of Conference: September 17-20, 2024
Track: Conjunction/RPO