Stargaze: SpaceX’s Space Situational Awareness System

Javier Roa Vicens, SpaceX; Eric Stoker-Spirt, SpaceX; Henrique Carmargo-Polido, SpaceX; Jason Chu, SpaceX; Curran Muhlberger, SpaceX; Kyle Krol, SpaceX; Alex Bixel, SpaceX; Cameron Ramos, SpaceX; Mehregan Dor, SpaceX; Nadim Khairallah, SpaceX; Ian Ruh, SpaceX; Vince Giralo, SpaceX; Alex Petrov, SpaceX; Michael Hoot, SpaceX; Erik Babcock, SpaceX; David Goldstein, SpaceX
https://doi.org/10.64861/VDBM5828

Keywords: space situational awareness, optical tracking, orbit determination, ephemerides, screening, orbit predictions

Abstract:

SpaceX has developed a new space situational awareness system called Stargaze. The system is fully operational: it is producing conjunction data messages (CDMs) that Starlink satellites act on, and it is freely available for the space community to submit their own ephemerides for screening. The system collects optical observations of resident space objects (RSOs) from the over 25,000 star trackers aboard Starlink satellites. Observations are downlinked to the ground leveraging the Starlink laser mesh that maintains satellites in uninterrupted contact. An orbit-determination service on the ground processes incoming observations to update the in-house catalog of space objects and propagates their orbits for screening. Additionally, Stargaze features an initial orbit determination service capable of rapidly cataloging newly launched objects, recovering objects after losing custody, and generally tracking objects that are not yet in the public catalog. This paper discusses the architecture and implementation of the system and presents key performance metrics.

Object detection on the star tracker is fully automated. Onboard algorithms find spots on the image and remove stars to identify RSOs. An estimation algorithm tracks the motion of these spots on the image and generates tracklets by associating spots across consecutive frames. Once a tracklet is complete after leaving the frame, it is sent to the ground for processing. The majority of the Starlink satellites are in continuous communication with the ground—either directly in view of one of the Starlink gateways or connected via laser to another satellite in contact with a gateway—which allows tracklets to reach the ground within seconds of being generated on the satellites. Stargaze is currently producing over 40 million tracklets per day, routinely observing objects as small as half a meter. Starlink will soon roll out a new generation of star tracker that is expected to detect cubesat-size objects

Tracklets arriving on the ground are first processed by the association service. This service propagates the orbits of the objects in the catalog to the observation time and runs a statistical confidence test to determine which object the tracklet corresponds to. Special logic handles maneuvering RSOs by testing multiple hypotheses compatible with realistic thrusting capabilities. Positively associated tracklets are sent to the orbit determination service to update the orbit of the corresponding object, whereas unassociated tracklets are sent to the initial orbit determination service. Approximately 17 million tracklets are positively associated per day.

The orbit-determination service uses an extended Kalman filter to process tracklets coming from the association service sequentially, without the need for batching them over long orbit arcs. In addition to position and velocity, the filter estimates a drag multiplier that models both the ballistic coefficient of the object and errors in the density model used by the orbit propagator. 95% of orbit updates occur within 5 minutes of the tracklet being generated on the star tracker.

The initial-orbit-determination service combines pairs of uncorrelated tracklets to find candidate orbit solutions. Two tracklets do not completely define all orbital parameters; Stargaze distils tracklet pairs into line-of-sight vectors and their time derivatives, which define a two-dimensional admissible region that constrains where the true orbit lies. Stargaze parameterizes the admissible region in terms of the semimajor axis and the range to the observer. A global optimization algorithm explores the admissible region seeking the global optimum that minimizes observation residuals, and a new object is created if the residuals are below a prescribed threshold. Stargaze can generally catalog tens of payloads within 24 hours of launch.

As soon as the orbit determination service updates an orbit, it is propagated forward in time by the propagation service, using the latest drag estimate from the filter. Predictions are submitted to the screening system for immediate processing and any resulting CDMs are automatically distributed to satellites and operators.

Stargaze can track the entire Starlink fleet, which provides an excellent benchmark for assessing the accuracy of the instantaneous orbit determination solutions, the performance of the association system through maneuvers, and the quality of the predictions exported by the system. The P90 instantaneous position error from Stargaze compared to the onboard navigation solution from Starlink satellites is generally better than 40 meters, and the P90 propagation error is 400 meters after 3 hours.

Date of Conference: September 15-18, 2026
Track: Space Domain Awareness

 

View Paper