Convergence of Generative AI, Digital Twins, and Edge Intelligence for Autonomous Cyber-Physical Systems: A Review
Abstract
ABSTRACT
The rapid expansion of Autonomous Cyber-Physical Systems (ACPS) in smart
manufacturing, vehicle swarms, and intelligent grids demands real-time
adaptability and cognitive control. Traditional Cyber-Physical Systems (CPS)
rely on deterministic control loops and centralized cloud computing, which
suffer from high network latency, single-point bandwidth bottlenecks, and an
inability to adapt under unobserved physical anomalies. To address these
issues, a transformative paradigm is emerging at the convergence of
Generative Artificial Intelligence (GenAI), Digital Twins (DT), and Edge
Intelligence (EI). This review presents a comprehensive analysis of the
convergence of GenAI, DT, and EI within next-generation ACPS
architectures. We investigate how generative architectures—specifically
Generative Adversarial Networks (GANs), Diffusion Models, and Physics
Informed Transformers—can be deployed across edge nodes to synthesize
telemetry, generate counterfactual scenarios, and enable rapid zero-shot
decision planning. Furthermore, we examine how Digital Twins provide high
fidelity, physics-bound state representations that ground GenAI outputs within
physical conservation laws. Key architectural components, low-latency
closed-loop synchronization protocols, model compression strategies, and
federated generative learning are detailed. Additionally, domain case studies
across vehicle swarms, factory robotics, and microgrids are evaluated.
Finally, open challenges regarding temporal drift, hallucination risks,
compute constraints, and security are discussed, establishing future research
pathways.
KEYWORDS: Generative AI, Digital Twins, Edge Intelligence, Autonomous
Cyber-Physical Systems, Federated Generative Learning, Physics-Informed
Neural Networks.
The rapid expansion of Autonomous Cyber-Physical Systems (ACPS) in smart
manufacturing, vehicle swarms, and intelligent grids demands real-time
adaptability and cognitive control. Traditional Cyber-Physical Systems (CPS)
rely on deterministic control loops and centralized cloud computing, which
suffer from high network latency, single-point bandwidth bottlenecks, and an
inability to adapt under unobserved physical anomalies. To address these
issues, a transformative paradigm is emerging at the convergence of
Generative Artificial Intelligence (GenAI), Digital Twins (DT), and Edge
Intelligence (EI). This review presents a comprehensive analysis of the
convergence of GenAI, DT, and EI within next-generation ACPS
architectures. We investigate how generative architectures—specifically
Generative Adversarial Networks (GANs), Diffusion Models, and Physics
Informed Transformers—can be deployed across edge nodes to synthesize
telemetry, generate counterfactual scenarios, and enable rapid zero-shot
decision planning. Furthermore, we examine how Digital Twins provide high
fidelity, physics-bound state representations that ground GenAI outputs within
physical conservation laws. Key architectural components, low-latency
closed-loop synchronization protocols, model compression strategies, and
federated generative learning are detailed. Additionally, domain case studies
across vehicle swarms, factory robotics, and microgrids are evaluated.
Finally, open challenges regarding temporal drift, hallucination risks,
compute constraints, and security are discussed, establishing future research
pathways.
KEYWORDS: Generative AI, Digital Twins, Edge Intelligence, Autonomous
Cyber-Physical Systems, Federated Generative Learning, Physics-Informed
Neural Networks.