Satellite Collision and Risk Assessment using Machine learning (SCRAM)

In response to the 2023 NASA SBIR solicitation subtopic Z8.13, ldquo;Space Debris Prevention for Small Spacecraftrdquo;, Advanced Space, LLC proposes to develop Machine Learning (ML) techniques to reduce and subsequently remove the ldquo;human-in-the-looprdquo; bottleneck exhibited by the Collision Avoidance (COLA) Concept of Operations (ConOps). The proposed solution is named SCRAM or Satellite Collision and Risk Assessment using Machine learning. SCRAM features a trade study of Recurrent and Transformer Neural Networks (NNs) to develop autonomous risk analysis for spacecraft collision avoidance. These new ML applications in astrodynamics provide early predictions of future collision risk trends and validation of collision avoidance maneuvers. The autonomous conjunction assessment highlights specific information for early collision risk prediction, while the dynamic space debris catalog builds on historical Conjunction Data Messages (CDMs) to incorporate uncertainty in real time. ML models can be inferenced orders of magnitude faster than traditional methods, significantly reducing both the computational and human hours required to perform collision avoidance operations. By identifying conjunction events early and automating the validation of collision avoidance maneuvers, the strain on COLA operators is reduced. SCRAM will be developed with the goal of future implementation into current COLA ConOps for space agencies such as the NASA Conjunction Assessment Risk Analysis (CARA) team. The framework created by the innovation has similar applications for mega-constellations and private SDA providers.