Threat-Analysis Oriented Digital Twinning of ML-Powered Future Autonomous Weapon Systems

<p>Warfare is undergoing a rapid transformation with the integration of artificial intelligence (AI) and machine learning (ML) into autonomous weapon systems (AWS) for perception, decision support, and control. As these systems become more software-defined, their cyber attack surface expands across sensing, communications, autonomy logic, and human-machine interfaces. As human oversight diminishes, ensuring the cybersecurity, resilience, and reliability of these systems becomes critical to mission success. This thesis investigates how a digital twin-driven threat modeling framework that integrates system-centric analysis with adversary-informed methodologies can support structured cybersecurity vulnerability evaluation and defensive strategy development associated with ML-powered AWS. First, the study analyzes the architectures, operational roles, and cyber-relevant vulnerabilities of the Army’s Robotic Combat Vehicle (RCV) and the Air Force’s Skyborg autonomy-core effort. From that analysis, a traceable set of security and assurance requirements spanning perception integrity, communications, human command and control interfaces, software and firmware integrity, autonomous cyber-physical actuation, and AI/ML-specific threats for a representative AWS are derived; enabling structured vulnerability analysis using the STRIDE model. These vulnerabilities are then contextualized within real-world adversarial behaviors through reference to the MITRE ATT&CK framework. To support evaluation of these risks, a proof-of-concept digital twin of an RCV-L platform is constructed to replicate the system’s decision-making logic and functional behavior, enabling controlled simulation of attack surfaces and observation of their impact on system behavior. Using this framework, the thesis examines key cyber-relevant vectors- including RF and wireless signal manipulation, adversarial machine learning considerations, and cyber-physical interfaces- within the context of autonomous system 1 operation. The results demonstrate a method for translating unclassified system knowledge to support the use of digital twin environments as viable testbeds for structured cybersecurity evaluation of AWS. Finally, this work discusses how the proposed repeatable framework may extend to more advanced AWS architectures, including Fully AWSs (FAWS), and outlines considerations for threat-informed design, instrumentation, and future red-team testing research as autonomy increases and human oversight is reduced.</p>

Name thesis:degree_name

Master of Science in Computer Science

Level thesis:degree_level

Thesis - Open Access

Discipline thesis:degree_discipline

Electrical, Computer, Software, and Systems Engineering

Year

2026

Author dc:creator

• Neubert, Thomas

• digital twin ; autonomous systems ; cybersecurity ; runtime assurance ; adversarial machine learning
• Artificial Intelligence and Robotics
• Other Electrical and Computer Engineering
• Other Operations Research, Systems Engineering and Industrial Engineering

Repository record dc:identifier

https://commons.erau.edu/edt/965

OAI identifier oai:identifier

oai:commons.erau.edu:edt-2008