NASA Conjunction Assessment Risk Analysis (CARA) Compendium for Artificial Intelligence and Machine Learning for Satellite Collision Avoidance

To ensure the continued and sustainable use of Low Earth orbit (LEO) and beyond, reliable and effective collision prediction between space objects is critical as the orbital environment faces unprecedented growth. Space Situational Awareness (SSA) for commercial and government missions now includes rapidly expanding satellite populations ranging from small, potentially less agile CubeSats to large constellations, each presenting unique conjunction assessment challenges and collision risks in the orbital regimes. With space object catalogs expected to increase tremendously and conjunction events becoming more frequent and complex, traditional risk assessment methods may face significant scalability limitations. NASA’s Conjunction Assessment Risk Analysis (CARA) program has conducted comprehensive research utilizing over 450,000 Conjunction Data Messages (CDM) from 2015-2018 to evaluate whether artificial intelligence and machine learning capabilities could enhance or augment collision avoidance decision making by performing conjunction assessment with increased speed or certainty using CDM data available early in event timelines. In essence, can these approaches enable rapid, reliable decision-making for collision avoidance within the critical 7-day identification window, incorporating expected or predicted information from anticipated new observations that could provide a bounded solution space? The research employed multiple Artificial Intelligence/Machine Learning (AI/ML) methodologies including supervised learning, unsupervised clustering techniques, Fuzzy Inference Systems, Deep Neural Networks, and Long Short-Term Memory models to identify risk-associated patterns during a conjunction as well as to analyze the temporal evolution of defined risk parameters. Statistical and information theory parameters were investigated beyond traditional probability of collision calculations, developing adaptive models for varying CDM availability, and implementing sophisticated time-series analysis approaches. The study identified fundamental challenges with the operational use of AI/ML for conjunction risk assessment, including data scarcity, the stochastic nature of orbital mechanics, model interpretability requirements, and the critical need for explainable AI approaches that can meet the high-reliability standards essential for space operations decision-making.