Karis Courey, Booz Allen Hamilton; Stephen Gerrells, Booz Allen Hamilton; Michael Moniger, Booz Allen Hamilton; Aubrey Shields, Booz Allen Hamilton; Karen Phumisithikul, Booz Allen Hamilton
Keywords: LLMs, Space Domain Awareness, SDA, Space Battle Management, SBM, Multi-Agent Systems, Agentic Architectures, Orbital Data Analysis, Publicly Available Information, PAI, Multimodal Data Fusion, AI, Machine Learning, Anomaly Detection, RAG, NLP
Abstract:
Introduction
With a surge in satellite launches and an increasing risk of collisions and debris generation, effective space domain awareness (SDA) and space battle management (SBM) are crucial for ensuring sustainability of space operations. Current methods often require specialized expertise, hindering accessibility and prompt analysis. This work introduces a conversational platform powered by Large Language Models (LLMs) to democratize access to satellite data and enhance collaborative decision-making in the space domain.
Approach & Results
The platform integrates a Retrieval Augmented Generation (RAG)-based LLM assistant, Luna, with a multi-user chat environment. Luna leverages a comprehensive knowledge base to respond to user queries in natural language. Meta’s No Language Left Behind (NLLB) is integrated for in-context translation of chat messages, fostering inclusivity and cross-language communication. This enables users to collaborate effectively, supporting faster and more informed decision making in critical situations, such as collision prediction, debris impact assessment, and closing kill chains.
Next Phase
With the ever-shifting space landscape, SBM, which requires rapid situational awareness, accurate threat assessment, and the ability to make time-critical decisions, has become a key focus area.
The next research phase focuses on developing networked LLM agents to enhance SBM by providing timely and accurate intelligence, automating data analysis, and facilitating collaborative decision-making among stakeholders. By leveraging multimodal data fusion and agentic architectures, this approach empowers users to effectively navigate the complexities of the evolving space environment and maintain a decisive advantage in space operations.
To achieve this, OrbAI and PubAI are incorporated in the Luna architecture.
OrbAI ingests, analyzes, and performs anomaly detection of orbital data to provide contextual insights for space-based operations. It leverages multi-intelligence tipping, queuing, and sense-making techniques to automate data collection and analysis from space-based sensors.
PubAI synthesizes orbital analysis outputs with Publicly Available Information (PAI), utilizing LLMs to generate actionable insights and recommendations. It aggregates data from diverse sources like news websites, satellite databases, and other PAI to construct a holistic understanding of the space environment.
This novel architecture offers several key advantages over Luna alone:
Enhanced Autonomy and Adaptability: The architecture enhances autonomy and adaptability in responding to dynamic situations and new information. Agents can independently gather, process, and act upon information, enabling faster responses and more flexible decision-making.
Modular and Adaptable Design: The modular design of the architecture allows for scalability and maintainability. New agents or data sources can be seamlessly integrated, such as a space weather agent for forecasting solar activities to improve orbit determination. Its modular design enables integration and compatibility with existing systems, while its flexible structure allows for easy adaptation to evolving requirements.
Multimodal Data Fusion: By combining space-based data with PAI, the architecture provides a comprehensive understanding of the space domain. PubAI interacts with data sources such as Space-Track, while OrbAI focuses on orbital data and anomaly detection.
By harnessing the power of agentic architectures and multimodal data fusion, these agents ensure timely, accurate, and comprehensive information is available to key stakeholders, empowering them to make critical decisions in the dynamic space environment.
Date of Conference: September 16-19, 2025
Track: Machine Learning for SDA Applications