Toward autonomous infrastructure management: Agentic AI for structural health monitoring, robotics, and digital twins

Modern sensing, robotics, artificial intelligence (AI), and digital twins (DTs) are reshaping the design, construction, operation, and maintenance of civil infrastructure. This review examines the evolution of structural health monitoring (SHM) from predictive models toward agentic AI for autonomous infrastructure management. Current research has achieved substantial progress in vibration-based monitoring, vision-based inspection, robotic sensing, and transformer-enabled analytics; however, most workflows remain diagnostic and lack autonomous reasoning and decision-making capabilities. This review defines the emerging concept of Agentic SHM, synthesizes recent advances in reinforcement learning, large language models, robotics, and DTs, and proposes layered architecture integrating perception, reasoning, planning, execution, and governance. Key challenges, including trustworthy autonomy, uncertainty-aware decision-making, reproducibility, standards compliance, and multi-agent coordination, are identified. The review concludes that progressively validated, human-supervised agentic AI provides a practical pathway toward safer, more resilient, and lifecycle-oriented infrastructure automation, while providing a roadmap for future research and practical implementation across infrastructure.