AV/OT Adversarial ML Framework

Doctoral Dissertation Research · Walsh College (Troy, MI)
DBA — AIML Leadership · PhD in Technology — Cybersecurity
Lamonte Smith · Expected Completion: December 2027

---

Dissertation Title

Designing a Resilience-Oriented Cyber-Secure Machine Learning Framework for Operational Technology in Autonomous Vehicle Systems Enhanced by Advanced Telecommunications Infrastructure

Status: Proposal formally approved · Research in progress

---

Research Overview

This research addresses a critical gap at the intersection of adversarial machine learning, operational technology (OT) security, and autonomous vehicle (AV) systems. As AV ecosystems increasingly rely on ML-driven decision-making embedded in OT environments — and as 5G/6G/V2X wireless infrastructure expands the attack surface — the need for resilience-oriented, cyber-secure ML frameworks becomes urgent.

The dissertation applies adversarial attack simulation (FGSM and PGD) against ML models embedded in AV/OT control systems, evaluating degradation under attack and designing defensive frameworks that maintain operational resilience.

---

Research Questions

1. How do adversarial ML attacks (FGSM/PGD) affect the operational integrity of ML models embedded in AV/OT systems?
2. What design principles constitute a resilience-oriented, cyber-secure ML framework for AV/OT environments?
3. How does advanced telecommunications infrastructure (5G/6G/V2X) affect the threat surface and defensive posture of AV/OT ML systems?

---

Theoretical Framework

This research is grounded in a hybrid theoretical framework spanning three domains:

Domain	Framework	Application
Cyber-Physical Systems Security	CPS Security Theory	Modeling threat surfaces across physical and cyber layers in AV/OT systems
Adversarial Machine Learning	Adversarial ML Theory	FGSM/PGD attack modeling, robustness evaluation, and adversarial training
Resilience Engineering	Resilience Engineering Framework	Designing systems that absorb, adapt, and recover from adversarial conditions

---

Methodology

Design: Convergent mixed-methods

Quantitative strand:

• Adversarial attack simulation using FGSM (Fast Gradient Sign Method) and PGD (Projected Gradient Descent)
• ML model degradation measurement under attack across multiple OT scenarios
• Simulation environments: CARLA · ROS · SUMO · OMNeT++
• Metrics: model accuracy degradation, decision latency, system recovery time

Qualitative strand:

• Expert interviews with AV cybersecurity and OT security practitioners
• Thematic analysis using established qualitative coding methods

Integration point: Quantitative simulation results and qualitative expert insights converge to inform framework design principles

---

Simulation Environment Suite

Tool	Role
CARLA	Autonomous vehicle simulation — sensor modeling, scenario generation
ROS (Robot Operating System)	AV control system integration and middleware
SUMO	Traffic simulation — V2X communication modeling
OMNeT++	Network simulation — 5G/6G/V2X infrastructure modeling
PyTorch	Adversarial attack implementation (FGSM/PGD)

---

Dissertation Structure

Chapter	Title	Status	Target Term
Chapter 1	Introduction & Problem Statement	Planned	27/WI
Chapter 2	Literature Review	Planned	27/WI
Chapter 3	Methodology	Planned	27/SP
Chapter 4	Results & Analysis	Planned	27/SU
Chapter 5	Discussion & Framework Design	Planned	27/SU

---

Repository Structure

av-ot-adversarial-ml-framework/
├── docs/
│   ├── proposal/          # Approved dissertation proposal
│   ├── literature/        # Annotated bibliography and literature notes
│   └── framework/         # Evolving framework design documents
├── simulations/
│   ├── carla/             # CARLA scenario configurations
│   ├── ros/               # ROS integration files
│   ├── sumo/              # SUMO traffic simulation configs
│   └── omnetpp/           # OMNeT++ network simulation configs
├── adversarial/
│   ├── fgsm/              # Fast Gradient Sign Method implementation
│   ├── pgd/               # Projected Gradient Descent implementation
│   └── evaluation/        # Attack evaluation metrics and results
├── models/
│   ├── baseline/          # Baseline ML models for AV/OT tasks
│   └── defended/          # Adversarially trained / defended models
├── data/
│   └── README.md          # Data sources and collection notes
├── results/
│   └── README.md          # Experimental results (added as research progresses)
├── SECURITY.md
└── README.md

---

Academic Context

Field	Detail
Degrees	DBA (AIML Leadership) · PhD in Technology (Cybersecurity)
Institution	Walsh College, Troy MI
Proposal Status	Formally approved
Expected Completion	December 2027
Advisor Institution	Walsh College Doctoral Program

---

Related Work & Domains

• Adversarial robustness in autonomous driving perception systems
• OT/ICS cybersecurity and resilience frameworks
• V2X communication security under 5G/6G infrastructure
• ML model hardening for safety-critical embedded systems
• IT/OT convergence threat modeling

---

Security & Ethics

This research involves adversarial attack simulation in controlled environments only. All simulations are conducted against synthetic/virtual systems. No real vehicle systems, infrastructure, or production environments are targeted. Research follows Walsh College IRB guidelines and ethical research standards.

See SECURITY.md for repository security policy.

---

License

MIT License — see LICENSE

---

_{Doctoral Dissertation · Lamonte Smith · Walsh College · DBA + PhD · Expected December 2027}