AI for Satellite Collision Avoidance – Go/No Go Decision-Making

Abstract

The proliferation of manmade objects in orbit for commercial, defense, and research purposes has resulted in frequent collision alerts. While much of the conjunction assessment and collision avoidance pipeline is automated, the decision on whether or not to carry out an avoidance maneuver is typically left to satellite operators. This work explores the potential of automating this go/no go decision using artificial intelligence (AI). A set of high-risk Conjunction Data Messages (CDMs) pertaining to 200 events is provided to analysts who are tasked with manually classifying them with a go/no go for maneuver. These events have a probability of collision (PoC) close to the critical threshold of 1E-4, treading the line between a typical go and no go decision. The classification results from the analysts are used as training data for AI models. Feature engineering and hyperparameter optimization is done to ensure the models are well equipped to learn from this data. The models’ abilities to identify relevant patterns from the CDMs and correctly classify unseen CDMs into the two classes is assessed. This work discusses the characteristics and processing of the CDM population used, the results from the developed models, and the insights gained on the practicability of applying AI to decision-making in this context.