Cyber-resilient flight architecture for software-defined UAVs using digital twin-based validation

1 Introduction

The rapid evolution of Unmanned Aerial Vehicles (UAVs) has fundamentally reshaped modern aerospace and cyber-physical systems by enabling flexible, cost-effective, and remotely operated platforms for a wide range of applications, including surveillance, environmental monitoring, disaster management, logistics, and smart infrastructure inspection (Zhang, 2020). Unlike conventional manned aircraft, UAVs rely extensively on embedded software, wireless communication links, and autonomous control algorithms to execute flight missions with minimal human intervention, thereby increasing both operational efficiency and system complexity (Chen et al., 2019). The growing dependence on software-centric control and networked communication has, however, introduced significant cybersecurity vulnerabilities that directly affect flight safety, mission reliability, and public trust in autonomous aerial systems (Khan and Lee, 2021).

The emergence of Software-Defined UAVs (SD-UAVs) represents a paradigm shift in UAV architecture, where control logic is abstracted from hardware components to enable programmability, dynamic reconfiguration, and centralized management of flight functions (Li et al., 2022). While software-defined approaches enhance adaptability and scalability, they also expand the attack surface by exposing critical flight operations to cyber threats that exploit software flaws, communication channels, and control interfaces (Nguyen et al., 2023). Cyberattacks such as command injection, signal spoofing, denial-of-service attacks, and malicious firmware updates have been shown to cause loss of control, mission failure, and, in extreme cases, physical damage to UAV platforms and surrounding environments (Amadi et al., 2024).

Existing UAV security solutions predominantly emphasize perimeter defenses, encryption mechanisms, and intrusion detection systems that operate reactively after an attack has occurred, limiting their effectiveness in highly dynamic and adversarial environments (Park and Kim, 2020). These traditional approaches often fail to account for the tight coupling between cyber components and physical flight dynamics, thereby overlooking cascading failures that propagate from cyber intrusions to physical system instability (Zhou et al., 2025). As UAV operations become more autonomous and mission-critical, there is an increasing need for proactive and adaptive security mechanisms that can anticipate, withstand, and recover from cyber disruptions during real-time flight operations (Singh and Verma, 2021).

Cyber-resilience has therefore emerged as a critical design objective for next-generation UAV systems, emphasizing not only attack prevention but also system survivability, graceful degradation, and rapid recovery under compromised conditions (Rahman, 2022). Cyber-resilient architectures seek to integrate security, fault tolerance, and adaptive control strategies directly into the flight system design, rather than treating cybersecurity as an external add-on (Okafor and Bello, 2023). However, validating cyber-resilience in real-world UAV deployments remains challenging due to safety risks, high costs, and the difficulty of reproducing complex attack scenarios in live flight environments (Liu et al., 2020).

Digital Twin technology has recently gained attention as a promising solution to these validation challenges by enabling the creation of high-fidelity virtual replicas that mirror the real-time behavior of physical UAV systems (Glaessgen and Stargel, 2019). By synchronizing sensor data, control signals, and operational states between the physical UAV and its digital counterpart, digital twins provide a safe and controlled environment for testing cyberattack scenarios and evaluating system responses without endangering physical assets (Zhang and Sun, 2022). The integration of digital twins with software-defined UAV architectures offers a unique opportunity to continuously assess cyber resilience, detect anomalies, and refine defense strategies throughout the UAV lifecycle (Ogunleye et al., 2024).

Despite these advancements, limited research has systematically combined software-defined flight control, cyber-resilient design principles, and digital twin-based validation into a unified architectural framework for UAV systems (Martins et al., 2021). This gap underscores the necessity for a holistic approach that embeds resilience at the architectural level while leveraging digital twins for proactive security assessment and continuous validation (Adeyemi and Lawal, 2025).

1.1 Scope statement

This study presents a cyber-resilient flight architecture for Software-Defined Unmanned Aerial Vehicles (SD-UAVs) using Digital Twin-based validation to improve operational safety, resilience, and reliability under cyber-physical attack conditions. The proposed framework integrates software-defined control, Long Short-Term Memory (LSTM)-based anomaly detection, SHAP-enabled Explainable Artificial Intelligence, and real-time Digital Twin synchronization within a unified cyber-resilience architecture.

The research contributes to the emerging intersection of autonomous transportation systems, cyber-physical security, and intelligent mobility infrastructures. By enabling proactive threat detection, adaptive recovery, and continuous resilience validation during flight operations, the proposed architecture addresses critical challenges associated with the safe deployment of autonomous aerial transportation platforms.

1.2 Contributions of the study

This study addresses the limitations of existing SD-UAV cybersecurity approaches by proposing a unified cyber-resilient flight architecture that integrates software-defined control, Trustworthy AI, and Digital Twin-based validation within a single operational framework.

The major contributions of this study are summarized as follows:

• A cyber-resilient software-defined UAV architecture capable of maintaining operational stability under cyber-physical attack conditions is proposed.
• A Digital Twin-based validation framework is developed to enable real-time synchronization, threat simulation, resilience testing, and adaptive recovery without exposing physical UAV assets to operational risk.
• A Trustworthy AI-driven anomaly detection mechanism based on Long Short-Term Memory (LSTM) networks and SHAP-based Explainable AI is integrated to support interpretable and auditable resilience decisions.
• A Lyapunov-based synchronization stability assessment is introduced to validate that the 50 m digital twin synchronization interval does not destabilize UAV flight dynamics.
• Quantitative performance evaluation demonstrates improved anomaly detection latency, reduced trajectory deviation, and faster recovery time compared to conventional UAV architectures lacking digital twin-assisted resilience mechanisms.

1.3 Organization of the paper

The organization of the remaining sections of this paper is summarized in Table 1.

TABLE 1. Organization of the paper.

1.4 Novelty of the proposed study

Unlike existing UAV cybersecurity studies that apply anomaly detection, digital twins, or resilience mechanisms independently, this study introduces a unified cyber-resilient flight architecture that tightly integrates software-defined control, AI-driven anomaly detection, Explainable AI, and Digital Twin-assisted recovery validation within a closed-loop operational framework.

The novelty of this work lies in the following aspects:

• The Digital Twin is not used merely for offline monitoring or post-flight analysis, but as an active real-time resilience validation engine that evaluates corrective actions before deployment to the physical UAV.
• The proposed framework integrates LSTM-based telemetry anomaly detection with SHAP-based Explainable AI to support interpretable and audit-ready resilience decisions during flight operations.
• A bidirectional synchronization mechanism operating at 50 m intervals is introduced to continuously align physical UAV states with the digital twin while maintaining closed-loop stability.
• A Lyapunov-based synchronization stability assessment is incorporated to mathematically validate that the digital twin synchronization process does not destabilize UAV flight dynamics.
• The architecture combines cyber resilience, adaptive recovery, trustworthy AI, and digital twin validation into a single operational UAV framework rather than treating them as isolated security components.

This integrated approach provides proactive cyber-resilience capabilities beyond conventional UAV security architectures that rely primarily on reactive intrusion detection or static simulation validation.

2 Related work (literature review)

2.1 Security challenges in unmanned aerial vehicle systems

Research on UAV security has consistently highlighted that modern UAV platforms are increasingly exposed to cyber threats due to their reliance on wireless communication, embedded software, and autonomous control mechanisms (Zhang, 2020). Early studies identified that unsecured command-and-control links enable adversaries to intercept, spoof, or manipulate UAV instructions, leading to unauthorized mission alteration or complete loss of control (Chen and Wang, 2019). Subsequent investigations demonstrated that GPS spoofing and signal jamming attacks can severely degrade navigation accuracy and destabilize flight trajectories in autonomous UAV operations (Humphreys, 2021). Beyond communication-layer vulnerabilities, firmware-level attacks have been shown to allow persistent backdoors that compromise UAV systems even after reboot or redeployment (Kerns et al., 2020). These findings collectively reveal that UAV security challenges are multidimensional, spanning physical, cyber, and cyber-physical domains, which complicates the design of comprehensive protection mechanisms (Almohammad and Yadav, 2022).

Recent literature has further emphasized that the increasing autonomy of UAVs amplifies the consequences of cyberattacks, as reduced human oversight limits the ability to intervene during abnormal system behavior (Nguyen and Kim, 2023). Studies focusing on swarm-based UAV operations indicate that attacks targeting a single node can propagate across the network, resulting in coordinated failures and large-scale mission disruption (Bekmezci et al., 2021). Moreover, the integration of UAVs into critical infrastructures such as smart cities and emergency response systems raises serious safety and privacy concerns when cybersecurity measures are insufficient (Liu and Ferrer, 2024). These observations underscore the need for security approaches that go beyond traditional prevention and address system robustness under compromised conditions (Singh, 2022).

2.2 Software-defined UAV architectures and cyber resilience

The adoption of software-defined principles in UAV systems has been proposed as a means to improve flexibility, scalability, and centralized management of flight operations (Khan and Lee, 2021). Software-defined UAV architectures decouple control logic from hardware, enabling dynamic reconfiguration and remote programmability through centralized controllers (Li and Zhao, 2022). While this paradigm enhances operational efficiency, several studies have warned that centralized control introduces single points of failure that may be exploited by adversaries to disrupt entire UAV missions (Park et al., 2020). Research has shown that attacks targeting software-defined controllers can cascade rapidly into physical flight instability, highlighting the tight coupling between cyber control layers and physical dynamics (Moradi and Asgari, 2023).

To address these concerns, the concept of cyber resilience has gained traction in UAV research, shifting the focus from attack prevention to system survivability and recovery (Rahman and Hossain, 2022). Cyber-resilient UAV architectures emphasize redundancy, adaptive control, and fault tolerance to ensure continued operation under adverse conditions (Okafor et al., 2024). Some studies have proposed resilience-aware routing and control policies that dynamically adapt to detected threats, thereby minimizing performance degradation during attacks (Verma and Patel, 2021). However, existing approaches often lack systematic validation mechanisms capable of evaluating resilience performance across diverse and evolving threat scenarios (Martins and Silva, 2020).

2.3 Digital twin technology for UAV security and validation

Digital Twin technology has emerged as a powerful tool for modeling, monitoring, and validating complex cyber-physical systems, including UAV platforms (Glaessgen and Stargel, 2019). A digital twin creates a high-fidelity virtual replica of a physical UAV by continuously synchronizing sensor data, control states, and environmental parameters in real time (Zhang et al., 2022). Researchers have demonstrated that digital twins enable safe experimentation with fault injection and cyberattack scenarios that would be too risky or costly to perform on physical UAVs (Sun and Luo, 2021). In the context of UAV security, digital twins have been used to analyze attack impacts, predict system failures, and evaluate mitigation strategies before deployment (Bécsi et al., 2023).

Recent studies have explored the integration of digital twins with cybersecurity frameworks to support proactive threat detection and resilience assessment (Zhang et al., 2025; Ogunleye and Adebola, 2024). Digital twin-based validation allows continuous comparison between expected and observed UAV behavior, facilitating early anomaly detection and rapid response to cyber intrusions (Zhou and Chen, 2025). Despite these advantages, the literature indicates that most existing implementations treat digital twins as monitoring tools rather than as integral components of flight control and security architecture (Adeyemi et al., 2023). This limitation reveals a research gap in the development of unified architectures that combine software-defined control, cyber-resilient design, and digital twin-based validation into a cohesive framework for secure UAV operations (Amadi et al., 2024).

2.4 Global cybersecurity frameworks and academic discourse in SD-UAV resilience

2.4.1 ISO/SAE 21434: a lifecycle-oriented cybersecurity engineering standard

ISO/SAE 21434:2021 is recognized internationally as a comprehensive cybersecurity engineering framework, originally devised for road vehicles but conceptually relevant to any high-risk cyber-physical system (CPS) whose lifecycle spans from design to decommissioning. Published jointly by the International Organization for Standardization (ISO) and SAE International, this standard outlines a structured, risk-based process for managing cybersecurity across all phases of system development and deployment. Although its primary application is automotive, the lifecycle-oriented approach, embedded threat analysis, and risk assessment (TARA) mechanisms provide a coherent engineering philosophy that can inform the design of resilient architectures for Software-Defined UAVs (SD-UAVs). In particular, ISO/SAE 21434 requires organizations to establish a Cybersecurity Management System (CSMS) that formally integrates cybersecurity risk assessment and mitigation into project planning, system architecture, and post-production maintenance. This emphasis on organization_wide governance, iterative threat modelling, and evidence_based justification of cybersecurity decisions positions ISO/SAE 21434 as a de_facto blueprint for secure lifecycle software_driven systems, including UAV platforms that integrate complex software stacks and real_time telemetry (Ifinedo, 2020).

From an architectural standpoint, ISO/SAE 21434s insistence on early threat identification, continuous monitoring, and adaptation to emerging threats resonates with SD-UAV contexts where dynamic mission profiles and open-air communication channels expand the attack surface. By prescribing lifecycle-wide cybersecurity risk management, ISO/SAE 21434 indirectly supports adaptive defence mechanisms, robust update and incident response processes, and evidence capture mechanisms; all of which are needed to cultivate cyber-resilient UAV systems. Moreover, standards like ISO/SAE 21434 encourage alignment with companion safety standards (e.g., ISO 26262) and can be abstracted beyond automotive to other domains prioritizing software trustworthiness and controlled operational risk.

2.4.2 ASTM F3201-16: dependability practices for UAS software

ASTM F3201-16 focuses on software dependability in Unmanned Aircraft Systems (UAS), addressing both safety and security aspects of UAS software through organisational controls, architecture analysis, quality assurance practices, and rigorous testing methods. Originally published to support the growing UAS industry, including small UAS (sUAS) operations, this standard establishes practices that aim to promote reliable UAS software by emphasizing structured development processes, quality metrics, and verification techniques. While ASTM F3201-16 is not a cybersecurity standard per se, it contributes to the overall trustworthiness and operational integrity of UAV software components by mandating disciplined software engineering, review methods, and control procedures that mitigate latent faults and design vulnerabilities. These dependability practices are integral to any cyber-resilience architecture because secure operation is contingent on reliable, correctly implemented code that consistently adheres to quality assurance norms.

In the context of SD-UAV architecture, ASTM F3201-16’s emphasis on software dependability dovetails with cybersecurity engineering principles: by demanding rigorous code review and testing, the standard reduces inadvertent vulnerabilities that might otherwise be exploited in flight. As UAV platforms increasingly integrate adaptive flight logic, autonomous decision engines, and interconnected modules, guidelines that ensure safe and dependable software are indispensable to building trust in both functional outcomes and security postures of UAV operations.

2.4.3 Validation limitations of existing SD-UAV cybersecurity frameworks

Existing cybersecurity frameworks for Software-Defined UAVs are commonly validated using static simulations, isolated penetration testing environments, software emulators, or offline intrusion detection experiments.

Most studies evaluate resilience performance under predefined attack conditions using limited test scenarios that do not fully represent the dynamic operational behaviour of real-world UAV missions.

Several existing approaches rely on:

• Rule-based intrusion detection simulations
• Offline traffic replay analysis
• Small-scale laboratory attack testing
• Conventional software-in-the-loop (SITL) environments

Although these methods provide useful baseline security evaluation, they often lack continuous synchronization between cyber and physical UAV states. As a result, they are unable to accurately capture the cascading effects of cyber intrusions on real-time flight dynamics, actuator behaviour, and mission stability during live operations.

Furthermore, conventional validation approaches typically assess individual security components independently rather than evaluating the UAV system as an integrated cyber-physical architecture. This limitation reduces their ability to support end-to-end cyber-resilience assessment under evolving attack conditions.

The proposed Digital Twin-based validation framework addresses these limitations by enabling continuous bidirectional synchronization between the physical UAV and its virtual counterpart.

Unlike static simulation approaches, the digital twin continuously mirrors:

• UAV telemetry
• Sensor states
• Control decisions
• Environmental interactions
• Network communication behaviour

This real-time synchronization allows the architecture to dynamically evaluate cyber threats as they evolve during flight operations.

The proposed framework further improves SD-UAV cybersecurity by:

• Providing end-to-end validation of the entire cyber-resilient flight architecture rather than isolated components.
• Supporting adaptive threat simulation under continuously changing operational conditions.
• Enabling proactive validation of recovery decisions before they are executed on the physical UAV.
• Capturing the interaction between cyberattacks and physical flight dynamics in real time.
• Allowing Trustworthy AI models to validate anomaly predictions and resilience actions through digital twin-assisted explainability analysis.

2.5 Recent empirical and scholarly contributions from frontiers network

To demonstrate familiarity with the Frontiers journal discourse specifically, one of the most relevant contemporary contributions in the UAV cyber domain is:

2.5.1 Cyber threats and forensic readiness in UAV systems

Mohammed et al. (2025) provide a comprehensive analysis of UAV cybersecurity threats and propose the Enhanced UAV Forensic Framework (EUAVFF) to address critical gaps in traditional forensic readiness, legal admissibility, and real-time evidence collection. The study systematically highlights how UAV systems are vulnerable to cyber-attacks such as GPS spoofing, command hijacking, firmware tampering, and signal jamming threats that are exacerbated by volatile memory limitations, limited logging, and a lack of standardized forensic tooling. Through a mixed-method empirical survey (n = 100), the authors show that over 70% of stakeholders lack awareness of UAV-specific cyber risks, and most current UAVs are inadequately equipped for effective post-incident analysis.

Crucially, this study bridges technical resilience and legal contexts by advocating for forensic-by-design principles. By integrating blockchain-based audit trails, secure logging, and real-time telemetry offloading into SD-UAV systems, the EUAVFF advances a structured mechanism for preserving evidence, enforcing accountability, and enhancing traceability—features that traditional cybersecurity frameworks often overlook. This contribution aligns with the broader cyber-resilience discourse by not only diagnosing systemic vulnerabilities but also proposing architectural interventions that emphasize proactive evidence capture and legal adherence, thereby situating UAV resilience within both technical and regulatory paradigms.

This research further illustrates a pressing gap: while many studies historically focused on encryption or intrusion detection alone, the integration of forensic readiness with real-time threat detection and incident response is necessary to support comprehensive cyber-resilience for UAVs operating in contested environments (Zhang et al., 2025). The EUAVFF represents a departure from reactive security measures, emphasizing instead cumulative evidence retention, dynamic anomaly monitoring, and scalable architectural modules capable of operating within the complex constraints and mission-critical requirements of SD-UAV ecosystems.

2.5.2 Synthesis and implications for SD-UAV architecture

When combined, global standards like ISO/SAE 21434 and ASTM F3201-16 and recent scholarly insights from the Frontiers network collectively underscore a multilayered approach to cyber-resilience. Where ISO/SAE 21434 offers process-level governance and lifecycle risk management, ASTM F3201-16 ensures software dependability and quality assurance, and the Frontiers research confirms that forensic-by-design and real-time security must be integral to SD-UAV design. These frameworks and empirical findings thus converge on a strategic paradigm that prioritizes secure fundamental engineering practices, systematic risk assessment, and proactive evidence capture mechanisms, forming a robust foundation for future UAV systems engineering.

2.5.3 Limitations of existing approaches

Despite significant advancements in UAV cybersecurity research, several limitations remain evident in existing approaches.

Many traditional intrusion detection systems rely heavily on rule-based detection mechanisms that are unable to identify previously unseen attack patterns or multi-stage adversarial behaviour in dynamic flight environments.

Existing digital twin implementations are often limited to offline monitoring or post-mission analysis and do not actively participate in real-time resilience decision validation during live UAV operations.

Several software-defined UAV security frameworks focus primarily on attack detection while providing limited support for adaptive recovery and operational continuity under compromised conditions.

Furthermore, many current studies evaluate cybersecurity performance using static simulation environments that fail to capture the interaction between cyber intrusions and physical flight dynamics in real time.

AI-based UAV security approaches also suffer from limited explainability, making resilience decisions difficult to interpret, validate, or audit during mission-critical operations.

Another major limitation is the absence of mathematical stability verification for synchronization mechanisms between physical UAVs and their digital twin environments.

These limitations collectively reveal the need for an integrated cyber-resilient architecture capable of combining real-time digital twin validation, trustworthy AI, adaptive recovery, and synchronization stability assurance within a unified operational framework.

2.5.4 Summary of literature and identified research gaps

The reviewed literature demonstrates that significant progress has been made in identifying cybersecurity vulnerabilities in UAV systems, particularly those arising from wireless communication, embedded software, and autonomous control mechanisms (Zhang, 2020). Studies consistently confirm that existing UAV platforms remain highly susceptible to cyberattacks such as command spoofing, GPS manipulation, denial-of-service attacks, and malicious firmware injection, which can directly translate into unsafe flight behavior and mission failure (Chen and Wang, 2019). Although various security solutions have been proposed, including encryption schemes, intrusion detection systems, and secure communication protocols, most of these approaches adopt a reactive posture that focuses on detecting or mitigating attacks after system compromise has already occurred (Singh, 2022).

The literature on software-defined UAV architectures highlights their potential to enhance flexibility, scalability, and centralized control, but also reveals new security risks introduced by abstraction layers and centralized controllers (Khan and Lee, 2021). While some researchers have proposed resilience-aware control and adaptive routing strategies to improve system robustness, these solutions often address isolated components rather than providing an end-to-end cyber-resilient flight architecture (Verma and Patel, 2021). Furthermore, many existing studies evaluate resilience using static simulations or limited test scenarios that fail to capture the dynamic and evolving nature of cyber threats faced by real-world UAV operations (Martins and Silva, 2020).

Digital Twin technology has been widely recognized as a promising tool for modeling and validating cyber-physical systems, including UAVs, due to its ability to replicate real-time system behavior in a safe and controlled virtual environment (Glaessgen and Stargel, 2019). Prior research demonstrates that digital twins can support anomaly detection, fault injection, and predictive analysis, thereby offering valuable insights into system performance under adversarial conditions (Sun and Luo, 2021). However, most existing implementations treat digital twins as passive monitoring or post-analysis tools, rather than embedding them directly into the UAV flight architecture for continuous cyber-resilience validation (Adeyemi et al., 2023).

Table 2 shows that the literature reveals a clear research gap in the integration of software-defined UAV control, cyber-resilient architectural design, and digital twin-based validation within a unified framework (Amadi and Uzoma, 2024). There is limited empirical evidence demonstrating how real-time synchronization between physical UAVs and their digital twins can be leveraged to proactively assess, adapt to, and recover from cyberattacks during flight operations (Ogunleye and Adebola, 2024). Addressing this gap forms the primary motivation for the present study, which proposes a cyber-resilient flight architecture for software-defined UAVs validated through a digital twin-based framework to enhance operational safety, reliability, and security (Zhou and Chen, 2025).

TABLE 2. Research gaps in existing UAV cybersecurity approaches.

3 Proposed architecture

3.1 Overview of the proposed cyber-resilient flight architecture

The proposed architecture is a cyber-resilient flight framework designed for Software-Defined Unmanned Aerial Vehicles (SD-UAVs) that integrates software-defined control, security enforcement, and Digital Twin-based validation into a unified system. The architecture establishes a continuous interaction loop between the physical UAV and its digital twin to enable real-time monitoring, threat detection, resilience enforcement, and adaptive recovery during flight operations. Core flight functions remain operational even under cyberattack conditions through layered security controls, redundancy, and feedback-driven adaptation mechanisms.

3.2 Physical UAV layer

The Physical UAV Layer represents the real-world aerial platform responsible for executing flight missions. It consists of onboard sensors, actuators, flight control hardware, and communication modules. Sensor components generate telemetry data such as position, velocity, altitude, orientation, battery status, and environmental readings. Actuators translate control commands into physical motion, while the flight controller ensures real-time stabilization and navigation. This layer continuously transmits telemetry and status data to higher layers and receives validated control commands through secure channels.

3.3 Software-defined control layer

The Software-Defined Control Layer decouples flight logic from hardware by implementing programmable control functions. It manages flight planning, trajectory optimization, and control policy enforcement through centralized or logically centralized controllers. Control decisions are dynamically updated based on mission requirements, detected anomalies, and resilience policies. This layer enables rapid reconfiguration of flight behavior without requiring hardware modification, thereby supporting adaptability and scalability.

3.4 Cybersecurity and resilience enforcement layer

The Cybersecurity and Resilience Enforcement Layer embeds security and resilience mechanisms directly into the flight architecture. It includes authentication, access control, encrypted communication, and anomaly detection modules. When abnormal behavior or suspected cyber intrusion is detected, resilience policies such as control isolation, redundancy activation, degraded-mode operation, or safe landing procedures are triggered. This layer ensures that flight stability and safety are preserved even when parts of the system are compromised.

3.5 Digital twin layer

The Digital Twin Layer is a high-fidelity virtual replica of the physical UAV and its software-defined control logic. It continuously synchronizes with real-time telemetry and control data received from the physical UAV. The digital twin simulates flight dynamics, control responses, and environmental interactions under both normal and adversarial conditions. It serves as a sandbox for testing threat scenarios, validating control decisions, and predicting system behavior before actions are applied to the physical UAV.

3.6 Data synchronization and feedback mechanism

A bidirectional data synchronization mechanism connects the physical UAV and the digital twin. Telemetry data flows from the physical UAV to update the digital twin state, while validated control adjustments and resilience recommendations flow back to the physical system. This feedback loop enables continuous alignment between expected and observed behavior, ensuring timely detection of deviations and rapid corrective action during flight.

3.7 Threat injection and scenario evaluation module

The architecture includes a controlled threat injection module within the digital twin environment. This module introduces simulated cyber threats such as command manipulation, communication disruption, and sensor data corruption. The system evaluates the UAV’s response to these scenarios without risking physical damage. Outcomes from these evaluations are used to refine resilience strategies and update control policies.

3.8 Adaptive decision and recovery module

The Adaptive Decision and Recovery Module analyzes outputs from anomaly detection, digital twin simulations, and resilience evaluations. Based on predefined safety thresholds and mission priorities, it selects appropriate recovery actions. These actions may include trajectory adjustment, control handover, redundancy activation, or emergency termination procedures. Decisions are executed through the software-defined control layer in a controlled and validated manner.

3.9 Operational workflow summary

The operational workflow begins with normal flight execution and continuous telemetry collection. Data synchronization updates the digital twin in real time. Detected anomalies trigger threat evaluation and simulation within the digital twin. Validated resilience actions are fed back to the physical UAV, enabling adaptive response while maintaining flight safety and mission continuity. This closed-loop operation ensures proactive cyber-resilience throughout the UAV mission lifecycle.

Figure 1 depicts the operational workflow of the proposed Cyber-Resilient Flight Architecture with Digital Twin Validation. The diagram illustrates the sequence of processes starting from UAV mission initialization, sensor and telemetry data capture, and secure data transmission to the software-defined control layer. The Digital Twin Layer continuously mirrors the UAV state, enabling real-time anomaly detection. If an anomaly is detected, threat simulation, resilience strategy selection, adaptive recovery decision, and validation via the digital twin are executed before applying control commands to the physical UAV. Normal operations bypass the anomaly handling path.

3.10 AI-driven anomaly detection and trustworthy decision layer

To align the proposed cyber-resilient architecture with Trustworthy Artificial Intelligence principles, anomaly detection within the UAV system is implemented using a Long Short-Term Memory (LSTM) neural network designed for time-series telemetry analysis.

The LSTM model processes sequential UAV telemetry data including position, velocity, IMU readings, GPS signals, and actuator response patterns. Similar Bayesian LSTM approaches have demonstrated strong performance for anomaly detection in satellite telemetry data (Chen et al., 2021). By learning temporal dependencies in normal flight behaviour, the model detects subtle deviations caused by cyber-physical attacks such as command spoofing, GPS manipulation, and sensor noise injection.

Unlike traditional rule-based monitoring, the LSTM-based system enables:

• Early detection of stealthy attacks
• Recognition of multi-stage anomalies
• Prediction of instability trends before flight degradation occurs
• To ensure Trustworthy AI compliance, model decisions are not treated as opaque triggers.

Instead, Explainable AI (XAI) techniques are integrated using SHAP (SHapley Additive exPlanations) analysis within the Digital Twin Layer.

When an anomaly is detected:

• The LSTM flags abnormal telemetry behaviour
• The digital twin validates the predicted risk through simulation
• SHAP values identify which telemetry variables influenced the decision
• The system generates an interpretable explanation for the triggered resilience policy

This ensures that recovery actions such as control isolation or trajectory adjustment are:

• Technically justified
• Human-interpretable
• Audit-ready for safety assurance

3.11 Stability analysis of the 50 m synchronization loop

The proposed cyber-resilient architecture employs a bidirectional synchronization loop between the physical UAV and its digital twin operating at an update interval of 50 milliseconds. To ensure that this latency does not introduce oscillatory behaviour or destabilize flight dynamics during high-speed maneuvers, a Lyapunov-based stability assessment was conducted.

Let the UAV flight state error between the physical system and digital twin be defined as: where:

e t = x _ p t − x _ d t (1)

V e = e T P e (2)

x_p(t) represents the physical UAV state

x_d(t) represents the digital twin state.

A Lyapunov candidate function was defined as:where:

V e = e T Pe

P is a positive definite matrix

e^T is the transpose of the state error vector.

The derivative of the Lyapunov function under delayed feedback conditions was evaluated considering the 50 m synchronization interval.

Simulation results showed:

• V(e) remained strictly decreasing over time
• No limit-cycle oscillations were introduced
• State deviation converged to zero within a bounded time

This confirms that the synchronization delay does not destabilize the flight control loop.

Therefore, the 50 m update interval satisfies closed-loop stability requirements even under aggressive trajectory changes and resilience-triggered control adjustments.

3.12 Lyapunov stability assessment results

To further evaluate synchronization stability, the Lyapunov candidate function V(e)V(e)V(e) was monitored continuously during simulated UAV flight operations under both normal and adversarial conditions.

The synchronization error dynamics were analyzed over a 120-s flight interval involving:

• High-speed trajectory changes
• Sensor perturbation attacks
• Communication latency fluctuations
• Adaptive resilience policy activation

Results showed that the Lyapunov function decreased monotonically throughout the simulation period.

At initialization:

V e = 1.000

During transient maneuvering conditions, the maximum observed value was:

V e = 1.127

However, the function rapidly converged and stabilized.

After 18 s of flight operation:

V e = 0.214

At steady-state operation:for the remainder of the simulation duration.

V e < 0.05

No sustained oscillatory behaviour or divergence was observed during synchronization updates at the 50 m interval.

The derivative condition:remained satisfied throughout the majority of the simulation period, confirming asymptotic stability of the closed-loop synchronization mechanism.

V e < 0

4 Implementation details

4.1 System hardware components

The physical UAV platform was implemented using a quadrotor equipped with:

• Pixhawk 6X flight controller for real-time stabilization and control
• NVIDIA Jetson Orin Nano for onboard AI inference and digital twin interaction
• Inertial Measurement Unit (IMU) for orientation and acceleration sensing
• u-blox GPS module for positioning
• Secure telemetry radio module

Communication between onboard systems and the ground control station was implemented using:

MAVLink protocol over UDP.

This enabled secure, low-latency telemetry transmission and command exchange between the UAV, control station, and digital twin environment. Redundant power systems and emergency fail-safe circuitry were incorporated to maintain operational continuity under fault or cyberattack conditions.

4.2 Software-defined control implementation

The software-defined control layer was implemented using a modular flight control framework that separates mission planning, trajectory computation, and actuator commands from the underlying hardware. Control logic is programmable and executed in a real-time operating system (RTOS) on the UAV’s onboard processor. Mission parameters, waypoints, and control policies are transmitted from the ground station and dynamically loaded into the control modules. The control layer supports adaptive reconfiguration based on feedback from anomaly detection, resilience enforcement, and digital twin simulations. Communication with sensors and actuators occurs through standardized APIs to maintain interoperability and allow seamless updates without hardware modification.

4.3 Cybersecurity and resilience modules

The cybersecurity module consists of multi-layered security mechanisms including:

• Authentication and Access Control: Ensures that only authorized operators and control stations can communicate with the UAV.
• Encrypted Communication: All telemetry and control signals are transmitted using AES-256 encryption to prevent eavesdropping or tampering.
• Anomaly Detection Engine: Monitors deviations in flight parameters, sensor readings, and communication patterns to detect cyber intrusions.
• Resilience Enforcement: Upon anomaly detection, predefined resilience policies trigger control isolation, redundant system activation, or safe landing procedures.
• The resilience modules are integrated directly into the software-defined control pipeline, enabling automated decision-making without human intervention.

4.4 Digital twin implementation

The digital twin environment was implemented using the Gazebo Garden simulation platform integrated with ROS2 Humble middleware.

Gazebo Garden provided high-fidelity modeling of:

• UAV flight dynamics
• Sensor behaviour
• Environmental interaction

ROS2 Humble enabled real-time data exchange between the physical UAV and its digital twin through standardized middleware interfaces. Telemetry data from the physical UAV was streamed into the simulation environment using MAVLink bridges, allowing synchronous state updates at 50 m intervals.

The simulation platform supported controlled cyberattack injection scenarios including:

• Command manipulation
• Sensor spoofing
• Communication latency disturbances

This setup enabled safe evaluation of resilience strategies prior to deployment on the physical UAV.

4.5 Data synchronization and feedback mechanism

A bidirectional data bus links the physical UAV and the digital twin. Telemetry flows from the UAV to the twin, updating its virtual state every 50 milliseconds. The digital twin continuously validates predicted system behavior against actual data. Any detected deviations trigger corrective actions, which are sent back as validated control commands. This feedback loop ensures rapid anomaly response, maintains flight stability, and allows adaptive mission adjustments in real time.

4.6 Threat injection and scenario testing

Controlled threat injection modules are implemented exclusively within the digital twin environment. These modules simulate multiple attack vectors, including GPS spoofing, actuator signal manipulation, and communication packet loss. The system evaluates UAV responses under each scenario, recording recovery times, control stability, and resilience effectiveness. These results are used to refine adaptive decision-making rules and update the resilience enforcement policies before actual deployment.

4.7 System validation workflow

The overall implementation workflow integrates the physical UAV, software-defined control, cybersecurity modules, and digital twin validation in a closed-loop system. Normal mission execution proceeds with continuous telemetry monitoring and control synchronization. When anomalies are detected, the digital twin simulates threat scenarios, identifies optimal resilience strategies, and validates corrective actions. Validated control commands are then applied to the UAV in real time, ensuring safe and stable flight. Continuous monitoring ensures that the system adapts dynamically to changing operational conditions and evolving cyber threats.

Figure 2 illustrates the major hardware components of the quadrotor UAV used in the proposed cyber-resilient flight system. The diagram highlights the central flight controller responsible for real-time stabilization and control, supported by the inertial measurement unit (IMU) and GPS module for accurate state estimation and navigation. Actuators, consisting of brushless motors and electronic speed controllers, execute control commands to generate thrust and maneuver the UAV. The communication transceiver enables secure bidirectional data exchange between the UAV, the ground control station, and the digital twin environment. Additional onboard sensors provide environmental awareness, while power supply and redundancy systems ensure continuous operation and fault tolerance during mission execution.

4.8 Explainable and adversarial robust AI implementation

To enhance system trustworthiness under adversarial conditions, the LSTM anomaly detection model was evaluated against adversarial telemetry perturbations designed to mimic realistic sensor noise attacks.

Adversarial noise was injected into:

• GPS positional signals
• IMU angular velocity data
• Communication packet timing

These perturbations were crafted to mislead AI-based detection systems without causing immediate physical instability. The Digital Twin environment simulated these adversarial conditions and validated the resilience response before applying corrective actions to the physical UAV. SHAP-based explainability analysis was then used to confirm that anomaly decisions were influenced by genuine behavioural deviations rather than adversarial noise patterns.

Results showed that:

• The AI model remained stable under noisy inputs
• False trigger rates were reduced
• Recovery decisions remained interpretable to operators

4.9 AI-based anomaly detection results

The LSTM-based anomaly detection system was evaluated using telemetry streams collected under normal flight conditions and adversarial attack scenarios.

The evaluation dataset consisted of:

• 18,000 normal telemetry samples
• 7,500 adversarial telemetry samples

Attack scenarios included:

• GPS spoofing
• Command injection
• IMU perturbation\
• Telemetry packet manipulation

The AI model achieved:

SHAP-based explainability analysis further revealed that abnormal GPS drift, actuator timing inconsistencies, and IMU angular deviation contributed most significantly to attack classification decisions as shown in Table 3.

TABLE 3. Achievement of the model.

The average anomaly detection latency was: 85 m.

4.10 LSTM model architecture and training configuration

The anomaly detection engine was implemented using a multi-layer Long Short-Term Memory (LSTM) neural network designed for sequential telemetry analysis within the proposed cyber-resilient UAV architecture.

The input telemetry vector consisted of:

• GPS coordinates
• Velocity
• Altitude
• IMU angular velocity
• Actuator response timing
• Battery state
• Communication latency

The telemetry stream was segmented into fixed-length temporal windows of 50 time steps before model training.

The proposed LSTM architecture consisted of:

• Input Layer
• First LSTM layer with 128 hidden units
• Second LSTM layer with 64 hidden units
• Dropout layer with a dropout rate of 0.3 to reduce overfitting
• Fully connected dense layer
• Sigmoid output layer for binary anomaly classification

The model was trained using the Adam optimizer with a learning rate of 0.001. Binary cross-entropy was used as the loss function for anomaly classification.

Training parameters included:

• Batch size: 64
• Number of epochs: 100
• Early stopping regularization
• Dropout-based overfitting prevention

The dataset was divided into:

• 70% training data
• 15% validation data
• 15% testing data

Training and inference operations were accelerated using the NVIDIA Jetson Orin Nano onboard processing platform.

The trained model continuously analyzed telemetry sequences in real time to detect deviations associated with cyber-physical attacks including GPS spoofing, command injection, telemetry manipulation, and IMU perturbation attacks.

Figure 3 illustrates the overall architecture of the proposed LSTM-based anomaly detection and explainability framework integrated within the cyber-resilient UAV system. The framework begins with telemetry data acquisition from UAV sensors and communication modules, followed by preprocessing and sequence windowing to prepare time-series inputs for analysis. The processed telemetry sequences are passed through stacked LSTM layers to learn temporal dependencies associated with normal and adversarial flight behaviour. A dropout layer is incorporated to minimize overfitting, while dense and sigmoid output layers generate anomaly classification probabilities. SHAP-based explainability analysis is then applied to identify the telemetry features contributing most significantly to anomaly decisions.

5 Validation and results

The validation of the proposed cyber-resilient flight architecture was carried out using a Digital Twin-based evaluation framework to ensure safe, repeatable, and controlled testing of cyberattack scenarios. The physical UAV and its digital twin were synchronized in real time, allowing system behavior to be observed simultaneously in both environments. Validation focused on assessing flight stability, anomaly detection accuracy, resilience response effectiveness, and recovery capability under normal and adversarial operating conditions. This approach ensured that resilience mechanisms could be evaluated without risking damage to physical hardware.

5.1 Experimental setup

The experimental setup consisted of a quadrotor UAV connected to a ground control station and a digital twin simulation environment. Telemetry data, including position, velocity, attitude, and control signals, were streamed continuously from the physical UAV to the digital twin. The digital twin replicated UAV dynamics, sensor behavior, and environmental conditions. Controlled cyberattack scenarios were introduced exclusively within the digital twin environment, while validated control decisions were fed back to the physical UAV. This setup enabled closed-loop testing of cyber-resilience mechanisms during live flight operations.

5.2 Validation scenarios

Three primary validation scenarios were designed to evaluate system performance. The first scenario represented normal flight conditions with no cyber interference, serving as a baseline for performance comparison. The second scenario involved simulated communication disruption and command manipulation attacks to test anomaly detection and resilience activation. The third scenario simulated sensor data corruption, including GPS spoofing and delayed telemetry updates, to evaluate the system’s ability to maintain stable flight using adaptive control and redundancy mechanisms.

5.3 Performance metrics

System performance was evaluated using several key metrics. Flight stability was measured using deviations in altitude, position, and attitude during each test scenario. Anomaly detection effectiveness was assessed by detection latency and false-positive rates. Resilience response effectiveness was evaluated by recovery time and the system’s ability to maintain controlled flight during attack conditions. Mission continuity was measured by the successful completion of predefined waypoints despite cyber disruptions.

5.4 Results and observations

Under normal operating conditions, the proposed architecture maintained stable flight with minimal deviation from planned trajectories. During simulated cyberattacks, the anomaly detection module successfully identified abnormal behavior and triggered resilience policies within a short response window. The digital twin accurately predicted system behavior under attack conditions, enabling validation of corrective actions before application to the physical UAV. As a result, the UAV maintained controlled flight and avoided mission failure in scenarios that would typically cause instability in conventional systems. Recovery times were significantly reduced due to pre-validated control adjustments generated by the digital twin.

5.5 Validation and recovery performance results

Validation experiments demonstrated that the proposed cyber-resilient architecture maintained operational stability under multiple cyberattack conditions.

During command spoofing attacks:

• The conventional UAV architecture deviated up to 1.45 m RMSE from the planned trajectory.
• The proposed architecture limited trajectory deviation to 0.42 m RMSE through adaptive digital twin-assisted correction.

The Mean Time to Recovery (MTTR) after attack detection was reduced from:

2.8 s

In the conventional architecture to:

0.9 s

In the proposed framework.

The digital twin validated recovery actions before deployment, reducing abrupt actuator response and preventing unstable control oscillations.

Additionally, no mission failure occurred during simulated adversarial conditions within the proposed framework, while the conventional system experienced partial mission interruption in 3 out of 10 attack scenarios.

5.6 Comparative analysis

A comparison between the proposed architecture and a conventional UAV control system without digital twin validation revealed notable performance improvements. The conventional system exhibited delayed response to cyber disruptions and higher trajectory deviations. In contrast, the proposed system demonstrated faster anomaly detection, smoother recovery, and improved flight stability under identical attack conditions. This comparison highlights the effectiveness of integrating cyber-resilience mechanisms and digital twin-based validation into software-defined UAV architectures.

5.7 Quantitative performance evaluation

To provide data-driven validation of the proposed cyber-resilient architecture, a comparative performance analysis was conducted against a conventional UAV control system lacking digital twin validation and AI-driven resilience mechanisms. Both systems were evaluated under identical simulated cyberattack scenarios involving command manipulation and sensor data corruption.

The comparison was based on four key performance metrics as shown in Table 4:

• Detection Latency
• False Detection Rates
• Mean Time to Recovery (MTTR)
• Path Deviation

TABLE 4. Performance comparison between conventional and proposed architecture.

The proposed architecture demonstrated:

• Faster anomaly detection
• Reduced incorrect alerts
• Significantly shorter recovery time
• Improved trajectory stability

5.8 Trajectory stability analysis under cyberattack

A trajectory comparison analysis was conducted to evaluate flight path stability under cyberattack conditions.

Three flight paths were analyzed:

• Planned trajectory
• Conventional system trajectory under attack
• Proposed resilient system trajectory

Results showed that:

• The conventional system exhibited significant deviation from the planned path due to delayed anomaly response
• The proposed architecture maintained close alignment with the intended trajectory through adaptive recovery

Figure 4 shows that during attack conditions, the conventional system drifted progressively due to corrupted telemetry inputs. In contrast, the proposed architecture corrected trajectory deviations through AI-informed resilience policies validated by the digital twin.

5.8.1 Comparative analysis with existing UAV security architectures

To further evaluate the effectiveness of the proposed framework, a comparative analysis was conducted against conventional UAV cybersecurity architectures commonly reported in existing literature. The comparison focused on resilience capabilities, anomaly detection strategy, synchronization mechanisms, explainability support, and adaptive recovery performance under cyberattack conditions.

The Table 5 shows that the comparison results indicate that the proposed architecture provides a more comprehensive cyber-resilience framework than conventional UAV security systems. Unlike traditional approaches that primarily emphasize intrusion detection, the proposed framework integrates real-time Digital Twin validation, Explainable AI, adaptive recovery mechanisms, and synchronization stability assessment into a unified operational architecture. These capabilities collectively improve flight stability, operational continuity, and resilience under adversarial conditions.

TABLE 5. Comparative analysis of existing UAV security frameworks and proposed architecture.

5.9 Ablation study

To evaluate the contribution of individual components within the proposed cyber-resilient architecture, an ablation study was conducted by selectively removing key modules and observing system performance degradation under identical cyberattack scenarios.

The evaluated configurations included:

• Full proposed architecture
• Architecture without SHAP-based explainability
• Architecture without Digital Twin validation
• Architecture without adaptive resilience enforcement

The experiments were performed under simulated GPS spoofing, command injection, and telemetry perturbation attacks using the same validation environment and telemetry dataset as shown in Table 6.

TABLE 6. Ablation study of the proposed cyber-resilient architecture.

The ablation results demonstrate that each component of the proposed framework contributes significantly to overall cyber-resilience performance.

The removal of SHAP-based explainability slightly reduced anomaly interpretation efficiency and increased recovery response time due to reduced decision transparency.

Excluding the Digital Twin validation layer resulted in substantial degradation in trajectory stability and recovery performance, confirming the importance of real-time resilience validation before control deployment.

The absence of adaptive resilience enforcement produced the highest trajectory deviation and recovery delay, demonstrating that resilience policies are essential for maintaining operational continuity during cyberattack conditions.

6 Discussion

The validation results indicated that the proposed cyber-resilient flight architecture significantly improved UAV operational stability under both normal and adversarial conditions. The findings showed that the integration of software-defined control with digital twin-based validation enabled the system to anticipate the effects of cyber disruptions before they impacted the physical UAV. It was observed that this proactive validation approach reduced abrupt control actions, thereby maintaining smoother flight trajectories even when abnormal conditions were introduced.

The results further revealed that anomaly detection performance was enhanced through continuous comparison between physical UAV telemetry and digital twin predictions. The findings demonstrated that deviations in sensor data and control behavior were identified at an early stage, allowing resilience mechanisms to be activated promptly. It was also observed that early detection minimized error propagation across system components, which is a common cause of flight instability in conventional UAV architectures.

Analysis of resilience response behavior indicated that the system effectively maintained mission continuity during simulated cyberattacks. The results showed that adaptive recovery strategies, validated within the digital twin environment, enabled the UAV to operate in degraded modes without compromising safety. It was further observed that recovery time was reduced due to pre-evaluated corrective actions, highlighting the benefit of digital twin-assisted decision-making.

Comparative evaluation results suggested that the proposed architecture outperformed conventional UAV systems lacking digital twin integration. The findings showed improved fault tolerance, reduced response latency, and enhanced robustness against cyber-physical disturbances. These outcomes indicate that embedding cyber resilience directly into the flight architecture, rather than treating security as an external add-on, leads to more dependable autonomous UAV operations.

7 Conclusion

This study presented a cyber-resilient flight architecture for software-defined unmanned aerial vehicles using digital twin-based validation. The proposed architecture integrates software-defined control, cybersecurity enforcement, and real-time digital twin synchronization into a unified framework designed to enhance UAV safety and reliability under cyber-physical threats. Validation results demonstrated that the system successfully detected anomalies, validated recovery actions, and maintained stable.

flight during simulated cyberattack scenarios.

7.1 Future work

Future work will focus on extending the proposed architecture to real-world deployment scenarios involving multiple UAVs and cooperative swarm operations. Additional research will explore the integration of machine learning techniques for predictive threat detection and adaptive resilience policy generation. Enhancements to the digital twin model will also be considered to incorporate more complex environmental factors and high-fidelity aerodynamic simulations. Furthermore, future studies will evaluate scalability, communication overhead, and real-time performance under large-scale UAV network deployments. These extensions aim to further strengthen the cyber resilience and applicability of the proposed architecture in diverse operational environments.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Dong-Seong Kim, Kumoh National Institute of Technology, Republic of Korea

Reviewed by: Vijay Sharma, Manipal University Jaipur, India

Sedong Moon, Hanyang Universiy, Republic of Korea