Rapid urbanization has placed unprecedented demands on cities to manage infrastructure efficiently while ensuring sustainability and safety. Traditional methods for urban inspections—reliant on manual labour, fixed sensors, or manual drone deployments—often struggle to address the dynamic challenges of modern cities, from aging infrastructure to environmental hazards. These approaches are not only time-consuming but also limited in scalability, leaving gaps in real-time monitoring and decision-making.

Our research pioneers an integrated framework that bridges autonomous unmanned aerial systems (UAS), advanced artificial intelligence, and digital twin platforms to transform urban management. At its core lies a software architecture enabling single or multiple unmanned systems to corporate and operate autonomously across diverse environments, from dense cityscapes to complex indoor facilities. This system eliminates dependence on human pilots, reduces safety risks in hazardous inspections, and delivers actionable insights through seamless integration with digitized management platforms such as Building Information Modeling (BIM) and Geographic Information Systems (GIS).

Fig. 1 Overall system framework

To begin with, we embed AI into our unmanned aerial systems to achieve fully autonomous operations, allowing drones to adapt to real-world complexities. Equipped with lightweight edge computing modules, our UAS perform real-time environment perception, motion planning, and collision avoidance without manual interference. For example, in building façade inspections in high-density urban contexts, drones autonomously navigate around structural obstacles while identifying surface and subsurface defects through onboard AI models. Multi-drone coordination algorithms enable collaborative tasks, such as mapping large construction and industrial sites, where each unit dynamically adjusts its path based on shared data. This capability ensures uninterrupted operations even in GPS-denied environments, such as underground utilities or interior high-rise spaces, where traditional systems fail.

Furthermore, we developed high-performance 3D reconstruction and information modelling technologies as the backbone of our digital twin platform. Using AI-driven photogrammetry, the system generates millimetre-accurate 3D models of built assets. These models are then automatically enriched with semantic data, to form detailed BIM models. In a recent project inspecting all warehouses of China Resources Logistics, we have successfully applied the technology to efficiently generate high-resolution 3D models for our web-based defect management platform. The reconstruction process has been proved to be 10 times faster than commonly used commercial software tools with improved point cloud completeness up to 6 times. The technology has also been applied to heritage building conservation and real-time project progress monitoring, maintaining a unified digital repository for city-wide assets.

Finally, we integrate LLM-RAG (Large Language Model-Retrieval Augmented Generation) technology to enhance contextual understanding and decision-making in smart urban environments. By dynamically retrieving and synthesizing data from heterogeneous sources—including infrastructure maintenance logs, real-time IoT sensor streams, as well as inspection and management documents, the system generates actionable insights for anomaly detection and resource allocation. LLM-RAG enables semantic reasoning over complex urban datasets, identifying correlations between environmental factors and our management objectives. For predictive maintenance, this framework augments physics-informed models with domain-specific knowledge extracted from historical maintenance records and international standards, reducing false alarms in fault detection. In human-AI collaboration, the RAG component contextualizes operational decisions by cross-referencing live drone imagery with zoning regulations and emergency protocols, allowing operators to query natural-language scenarios and receive auditable recommendations. Adaptive learning mechanisms continuously update the retrieval database with new knowledge and drone-captured data, ensuring alignment with evolving urban dynamics.

Our research underscores the potential of autonomous unmanned systems and AI to achieve high efficiency and high accuracy urban environment inspection, modelling and management, aligning global urbanization with the principles of efficiency, sustainability, and resilience. The developed systems will redefine smart city infrastructure and establish the backbone for low-altitude economy ecosystems. Our team has presented above research outcome and achieved remarkable success at the 15th International Invention Fair in the Middle East (IIFME), winning the Gold Medal with the Congratulations of the Jury.

Fig. 2 Professor Benmei Chen and Professor Xi Chen won Gold Medal with
the Congratulations of the Jury at the 15^th International Invention Fair in the Middle East