Enrique De Alba, EO Solutions; Alexander Cabello, EO Solutions; Garrett Fitzgerald, EO Solutions; Zach Gazak, Space Systems Command A&AS; Justin Fletcher, Space Systems Command A&AS
Keywords: Large Language Models, Space Domain Awareness, Command and Control, Human-Machine Teaming, Natural Language Processing, Sensor Networks, On-premises AI
Abstract:
Space Domain Awareness (SDA) systems require operators to manage increasingly complex sensor networks through specialized scripting languages and legacy interfaces. While automation has advanced sensor scheduling and data fusion capabilities, the human-computer interface remains a critical bottleneck that limits operational efficiency. We present an on-premises natural language processing (NLP) system that leverages a large language model (LLM) to enable natural language command and control of the MACHINA (Mission-driven Autonomous Collaborative Heterogeneous Intelligent Network Architecture) framework for space operations. Our approach uses a state-based graph architecture using a 24B parameter open-source model, supporting three core capabilities: autonomous objective definition that translates natural language into structured commands, sensor-target visibility assessment, and preliminary system performance monitoring. Through trajectory-based evaluation, we demonstrate that our agent achieves high task completion rates (>90%) while maintaining process fidelity and error recovery mechanisms. Our analysis shows that reliable agent-based control emerges with larger models (24B+) compared to simpler structured output tasks achievable with 7B models. Our findings suggest that natural language processing systems are able to replace specialized scripting languages and dedicated visual interfaces for some tasks. More broadly, our work validates a template for deploying LLM-based agents in defense environments with potential to enhance operational workflows while preserving human oversight.
Date of Conference: September 16-19, 2025
Track: Machine Learning for SDA Applications