Towards Automated Satellite Conjunction Management with Bayesian Deep Learning

Abstract

After decades of space travel, low Earth orbit is a junkyard of discarded rocket bod-ies, dead satellites, and millions of pieces of debris from collisions and explosions.Objects in high enough altitudes do not re-enter and burn up in the atmosphere, butstay in orbit around Earth for a long time. With a speed of 28,000 km/h, collisionsin these orbits can generate fragments and potentially trigger a cascade of morecollisions known as the Kessler syndrome. This could pose a planetary challenge,because the phenomenon could escalate to the point of hindering future spaceoperations and damaging satellite infrastructure critical for space and Earth scienceapplications. As commercial entities place mega-constellations of satellites in orbit,the burden on operators conducting collision avoidance manoeuvres will increase.For this reason, development of automated tools that predict potential collisionevents (conjunctions) is critical. We introduce a Bayesian deep learning approachto this problem, and develop recurrent neural network architectures (LSTMs) thatwork with time series of conjunction data messages (CDMs), a standard data formatused by the space community. We show that our method can be used to modelall CDM features simultaneously, including the time of arrival of future CDMs,providing predictions of conjunction event evolution with associated uncertainties.