Evaluating Data Strategies in Training AI Solutions for Space C2

The Phase I research objective is to design a digital training environment for Data-Informed Competitive Endurance (DICE). The DICE training environment will perform modeling and simulation to train an autonomous system for Space C2 to support responsible counterspace campaigns, avoid operational surprise, and deny first-mover advantage. The objective of the autonomous system is to enable increased autonomy on-orbit for blue operational spacecraft. The training environment will comprise observation space, action space, Artificial Intelligence (AI) algorithms, and visualization. The problem statement is to evaluate the relative performance of a data strategy when the digital training of the autonomous system is performed using Space Domain Awareness (SDA) data of varying quality. A data strategy is defined by a use case and a plan to use SDA data to address specific requirements and objectives of that use case. The use case is to avoid operational surprise by red Rendezvous and Proximity Operations (RPO). The environment will be constructed to train for C2 scenarios by building upon open-source software frameworks for astrodynamics, AI algorithms, and visualization. The initial state of the observation space will be populated using SDA data and simulation-over-real data. The sources of SDA data will be Space Surveillance Network (SSN) and commercial data providers. The SDA data will also include simulated on-orbit optical data collected by the blue spacecraft. The simulation-over-real data will insert anomalous objects (e.g., red inspector satellites and abnormal observables) in the observation space to emulate C2 scenarios for blue spacecraft, e.g. RPO. The modeling and simulation will be set up to train for C2 scenarios and exercises to develop a æTheory of Success,Æ which consists of multiple hypotheses (data strategies) to use SDA data for Space C2. Meta-Reinforcement Learning (RL) AI algorithms will be trained to make choices as per General SaltzmanÆs Competitive Endurance theory.Ā