Machine Learning Argument of Latitude Error Model for LEO Satellite Orbit and Covariance Correction
A new ML model corrects atmospheric drag errors in LEO satellites, improving orbit prediction accuracy.
Deep Dive
Researchers Alex Moody, Penina Axelrad, and Rebecca Russell developed a machine learning model that corrects errors in the 'argument of latitude' for Low Earth Orbit (LEO) satellites. It uses a time-conditioned neural network or Gaussian Process trained on public ephemeris data to predict error as a Gaussian distribution. This extends the usable time horizon for satellite-based Position, Navigation, and Timing (PNT) services by maintaining more accurate uncertainty estimates without modifying existing propagators.
Why It Matters
Enables more reliable satellite navigation as an alternative to GPS, crucial for autonomous systems and global connectivity.