Abstract

The transformative advancement of critical infrastructure is being driven by the integration of artificial intelligence (AI) and machine learning (ML) with cyber-physical systems (CPS). In this review, two seemingly distinct domains-Intelligent Energy Systems, including smart grids and energy communities, and Intelligent Localization and Navigation Systems for GPS-denied environments-are unified through a common methodological lens. A shared framework of AI/ML paradigms, encompassing supervised and reinforcement learning, probabilistic models, sensor fusion, and physics-informed techniques, is established. Furthermore, the adaptation of these methods is examined under domain-specific constraints, such as stochastic energy generation, power flow limitations, and real-time robotic navigation. Through the analysis of high-impact studies from 2018-2026, cross-cutting trends are identified, including the use of graph neural networks for spatial modeling and multi-agent systems for decentralized decision-making. Persistent challenges regarding data scarcity, interpretability, and edge deployment are found to remain central across CPS applications. In response to these challenges, the CPS-Bench platform-a unified, cross-domain benchmark designed to standardize evaluation across energy and localization tasks-is proposed. Four primary contributions are presented: (1) a cohesive taxonomy by which two previously isolated CPS domains are connected, (2) a comparative analysis through which transferable techniques are uncovered, (3) a thorough assessment of domain-specific adaptations, and (4) the introduction of the CPS-Bench framework alongside a practical roadmap for interdisciplinary collaboration. It is concluded that the next generation of CPS will be defined by adaptive, embodied intelligence, where AI is co-designed with physical components. By such a vision, the energy transition is expected to be accelerated and resilient, human-centric infrastructure is expected to be realized.