Tiffany Phan, University of Texas at Austin; Moriba Jah, University of Texas at Austin
Keywords: machine learning, natural language processing, socio-technical configuration analysis, data fusion, zero-shot classification, retrieval-augmented generation
Abstract:
This research aims to develop a socio-technical configuration of anthropogenic space objects (ASOs) in Earth’s orbit to enhance space domain awareness (SDA). As space becomes increasingly congested with both governmental and private sector activities, understanding the complex relationships between ASOs is critical for predicting and managing threats in Earth’s orbit. Traditional SDA methods focus on tracking the physical characteristics of objects (e.g., orbits, trajectories, and velocities) using radar and optical observations, but while these methods are crucial for collision avoidance, they do not fully address the broader socio-political context in which space operations take place. To overcome this limitation and better enhance our understanding of the space environment, the focus of this work is to apply natural language processing (NLP) techniques including zero-shot classification and retrieval-augmented generation to extract and combine information from diverse data sources about space objects and their socio-political connections.
Through several ventures including national space programs, commercial investments, and international collaborations, ASOs in orbit are deeply intertwined with global political, economic, and military dynamics. As such, a complex network of relationships spanning these interconnected domains govern the use and management of Earth’s orbit beyond the technical aspects of these objects. Political alliances, military objectives, international agreements, and economic ties can provide additional insight to the environment in which these ASOs operate. By incorporating socio-political data, this research seeks to illuminate the broader context in which space objects are deployed and operated. This socio-technical configuration approach allows for a more nuanced analysis of the space environment, bridging the gap between the technical characteristics of hard data and the social characteristics of soft data.
To achieve this, this research employes various NLP techniques, primarily zero-shot classification and retrieval-augmented generation (RAG), to extract and combine these hard and soft characteristics from various data sources. Zero-shot classification models are used to categorize space objects based on information such as their launch state, their mission purpose, and their operational status. While less accurate than other models, zero-shot models do not require domain-specific training data, but do allow for real-time classification of new space objects from publicly available information. RAG models help extract data and synthesize information from various sources, including news articles, research papers, and official government publications, increasing the reliability and functionality of our proposed framework. These NLP techniques help extract and combine information from both hard and soft data sources.
This integration of both hard and soft data helps construct a dynamic graphical network that represents the interconnectedness of space objects. This network captures not only the physical proximity and potential collisions between objects in orbit, but also the socio-political relationships that influence space activities. For example, satellites owned by different countries may have political, military, or economic connections that could suggest potential points of conflict or collaboration. Two satellites launched by countries with strained political relations may pose a higher risk of becoming involved in a conflict than two satellites from politically neutral nations. Similarly, commercial satellite operations influenced by certain economic interests may indicate potential vulnerabilities. By analyzing these relationships, this research aims to capture the complex nuances of the space environment, thereby enhancing the reliability and accuracy of SDA efforts.
By integrating socio-political factors into the hard data analysis, we provide a more comprehensive tool for understanding space activities and their potential threats. Socio-technical configuration analysis can help identify vulnerable points in space infrastructure, including dependencies on critical satellites for communication, navigation, or reconnaissance. The graphical network constructed from this configuration can also serve as a predictive model, helping policymakers anticipate future events and take proactive measures to mitigate risks in space. Application of this research can expand to conflict prevention, diplomatic negotiations, and space traffic management; the ability to identify political, military, or economic motives behind the launch and operation of ASOs could facilitate diplomatic efforts to resolve conflicts before they escalate. It will advance our understanding of the space domain and contribute to the development of tools to improve decision-making in space security and sustainability forums.
In conclusion, this study proposes a novel socio-technical approach to SDA, combining technical data with socio-political insights to create a more holistic representation of space object interactions, thus enabling a deeper understanding of space activities and improving predictions of potential threats to ASOs in an increasingly complex and crowded space environment. The application of NLP techniques from zero-shot classification to retrieval-augmented generation enables the extraction of relevant information from diverse sources, creating a comprehensive graphical network that maps both the physical and socio-political dimensions of space activity. By linking technical data with socio-political insights, the research aims to improve the ability to predict and mitigate potential threats in Earth’s orbit, contributing to the safer and more sustainable use of space.
Date of Conference: September 16-19, 2025
Track: Space Domain Awareness