Yashica Khatri, NorthStar Earth & Space, Inc.; Manuel Sanjurjo-Rivo, NorthStar Earth & Space, Inc.; Shrouti Dutta, NorthStar Earth & Space, Inc.; Laura Pirovano, NorthStar Earth & Space, Inc.; Yann Picard, NorthStar Earth & Space, Inc.; Narendra Gollu, NorthStar Earth & Space, Inc.
https://doi.org/10.64861/IYYO1005
Keywords: conjunction assessment, large uncertainty, Gaussian Mixture Models, Advance probability of Collision
Abstract:
Collision risk assessment immediately after launch and during the early orbit phase (EOP) remains one of the least mature elements of operational space safety. In this period — typically spanning from separation through the first days or weeks of orbit-raising — the available state knowledge of newly deployed spacecraft is sparse, biased, and rapidly evolving. Tracking coverage may be insufficient, estimation pipelines may still be converging, and publicly available orbit products such as TLEs provide limited or no quantification of uncertainty. Under these conditions, classical conjunction assessment methods become unreliable: screening filters lose efficiency, covariance-based uncertainty propagation distorts the true probability geometry, and probability of collision (PoC) formulations that rely on small-error assumptions or planar projections exceed their range of applicability. Nonetheless, operators are expected to provide defensible risk evaluations from the moment a satellite is released, a responsibility that is growing in parallel with the global expansion of launch cadence.
This work proposes a methodology tailored to the challenges of conjunction assessment in the early orbit phase, where the uncertainty can be both large and strongly non-Gaussian. The approach is structured around the pillars of robust state-vector propagation under large uncertainty and computation of collision probability in a manner compatible with nonlinear geometry and non-Gaussian uncertainty representations. In this paper, the overall framework is presented, including the Unscented Transform (UT)/Gaussian Mixture Model (GMM) hybrid propagation architecture and mixture-compatible PoC formulations. The GMM-UT propagation is shown to incorporate nonlinear capture through mixture spread and high-fidelity propagation through sigma points.
Simulation results illustrate a scenario that reflects early-phase uncertainty conditions. The proposed methodology enhances the reliability, interpretability, and operational utility of conjunction risk assessments exactly when classical methods are least dependable, immediately after launch and throughout the early orbit phase.
Date of Conference: September 15-18, 2026
Track: Conjunction/RPO