Jun Ma

As the architecture, engineering, and construction (AEC) industry become increasingly data-centric and AI-enabled, CAD drawings remain a major and important source of reusable engineering knowledge for design review, coordination, checking, and maintenance. Retrieving this knowledge from archives remains difficult because relevant evidence is distributed across symbols, annotations, title blocks, and context. Engineers must therefore identify relevant sheets, verify their document identity, and inspect the local regions that carry decision-relevant evidence. This paper proposes a novel context-conditioned evidence-retrieval framework for construction CAD archives, in which the drawing sheet serves as the context-bearing unit for archival identity and a local region serves as an evidence unit surfaced within that sheet-level context. SemCAD grounds sheet identity in title-block OCR evidence verified by vision-language-model analysis, ranks sheets by fused sparse and dense retrieval over the resulting semantic representation, and restricts visual patch matching to a retrieved sheet. On 150 sheets from 33 projects across five engineering domains, SemCAD reaches mean reciprocal rank (MRR) of 0.947 and an end-to-end evidence hit rate of 86.2%, compared with 69.0% for archive-wide patch ranking. Removing title-block grounding changes MRR by only 0.001 yet reduces complete drawing-register accuracy from 70.6% to 36.0%. These results show that retrieval effectiveness, record auditability, and evidence inspectability are coordinated but distinct capabilities required for reusable engineering drawing archives.

The siting of neighborhood‐scale service facilities requires balancing accessibility, facility capacity, and feasible land under environmental and regulatory constraints. Municipal solid waste collection points provide a representative case, especially across contrasting urban and rural settlement forms. Existing approaches often address only parts of this problem: GIS‐based suitability analysis identifies candidate areas but rarely incorporates explicit assignment and capacity control, while clustering and location‐allocation methods may overlook siting feasibility under exclusion constraints. This paper develops an integrated geospatial siting framework, the Spatial Capacitated Allocation Network (SCAN), which combines capacitated assignment, soft service radius, and exclusion‐aware location refinement within a single workflow. SCAN is solved through an Alternating Assignment Relocation procedure integrating mixed‐integer assignment with continuous relocation while preserving siting feasibility. Two case studies from Shenzhen and Sichuan show that SCAN generates feasible locations and balanced allocations while revealing how settlement structure shapes accessibility and feasibility trade‐offs.

Urban air pollution severely impacts pedestrians on sidewalks, yet traditional fixed-site monitoring lacks the spatial coverage needed for fine-scale exposure assessment due to high costs. This study proposes a novel machine-learning framework to predict short-term sidewalk PM2.5 and PM1 concentrations using multimodal audiovisual features extracted from self-collected street-view videos, alongside meteorological and background pollution data. Based on a mobile monitoring campaign in Shenzhen, China, we evaluated multiple models (linear regression, XGBoost, and LightGBM) across different temporal resolutions (10 s and 1 min) and validation strategies. LightGBM achieved the best performance, yielding R2 values of 0.64–0.65 for 10 s predictions and 0.80 for 1 min predictions under random cross-validation. Under rigorous spatial cross-validation, the model maintained moderate generalizability, with R2 reaching 0.41–0.48 at the 1 min resolution. Furthermore, developing a hybrid model that incorporated static geospatial context further improved the overall predictive accuracy. Variable interpretation revealed that while background PM and meteorology were dominant predictors, dynamic audio-derived features and visual indicators provided substantial additional predictive power. These findings demonstrate that integrating multimodal audiovisual sensing with ancillary data enables scalable, high-resolution estimation of street-level PM, effectively complementing conventional monitoring for urban air-quality management.

Despite advancements in Building Information Modelling (BIM), a significant gap remains in automated compliance checking (ACC) for Passive Fire Protection (PFP). Existing semi-automatic systems often lack the flexibility to interpret complex regulatory logic while maintaining the performance required for large-scale BIM environments. This paper proposes ChatRevit FireCode, an end-to-end system integrated within Revit that automates the transition from natural-language regulations to executable C# code. By leveraging Large Language Models (LLMs), the system dynamically determines query intent, structures complex fire safety requirements, performs multi-hierarchical spatial reasoning, and generates check code in Revit environment without intermediate human intervention. Experimental results across 104 diverse compliance queries demonstrate a 61.54% success rate with a mean processing time of approximately 40 seconds, proving the system’s cost-effectiveness and technical viability for real-world AEC applications.

As an important part of urbanization, the development monitoring of newly constructed parks is of great significance for evaluating the effect of urban planning and optimizing resource allocation. However, traditional change detection methods based on remote sensing imagery have obvious limitations in high-level and intelligent analysis, and thus are difficult to meet the requirements of current urban planning and management. In face of the growing demand for complex multi-modal data analysis in urban park development monitoring, these methods often fail to provide flexible analysis capabilities for diverse application scenarios. This study proposes a multi-modal LLM agent framework, which aims to make full use of the semantic understanding and reasoning capabilities of LLM to meet the challenges in urban park development monitoring. In this framework, a general horizontal and vertical data alignment mechanism is designed to ensure the consistency and effective tracking of multi-modal data. At the same time, a specific toolkit is constructed to alleviate the hallucination issues of LLM due to the lack of domain-specific knowledge. Compared to vanilla GPT-4o and other agents, our approach enables robust multi-modal information fusion and analysis, offering reliable and scalable solutions tailored to the diverse and evolving demands of urban park development monitoring.

Understanding urban environment change is essential for sustainable development. However, current approaches, particularly remote sensing change detection, often rely on rigid, single-modal analysis. To overcome these limitations, we propose MMUEChange, a multi-modal agent framework that flexibly integrates heterogeneous urban data via a modular toolkit and a core module, Modality Controller for cross- and intra-modal alignment, enabling robust analysis of complex urban change scenarios. Case studies include: a shift toward small, community-focused parks in New York, reflecting local green space efforts; the spread of concentrated water pollution across districts in Hong Kong, pointing to coordinated water management; and a notable decline in open dumpsites in Shenzhen, with contrasting links between nighttime economic activity and waste types, indicating differing urban pressures behind domestic and construction waste. Compared to the best-performing baseline, the MMUEChange agent achieves a 46.7% improvement in task success rate and effectively mitigates hallucination, demonstrating its capacity to support complex urban change analysis tasks with real-world policy implications.

Existing change detection methods often lack the versatility to handle diverse real-world queries and the intelligence for comprehensive analysis. This paper presents a general agent framework, integrating Large Language Models (LLM) with vision foundation models to form ChangeGPT. A hierarchical structure is employed to mitigate hallucination. The agent was evaluated on a curated dataset of 140 questions categorized by real-world scenarios, encompassing various question types (e.g., Size, Class, Number) and complexities. The evaluation assessed the agent's tool selection ability (Precision/Recall) and overall query accuracy (Match). ChangeGPT, especially with a GPT-4-turbo backend, demonstrated superior performance, achieving a 90.71 % Match rate. Its strength lies particularly in handling change-related queries requiring multi-step reasoning and robust tool selection. Practical effectiveness was further validated through a real-world urban change monitoring case study in Qianhai Bay, Shenzhen. By providing intelligence, adaptability, and multi-type change analysis, ChangeGPT offers a powerful solution for decision-making in remote sensing applications.

Urban air mobility (UAM) introduces electric vertical takeoff and landing (eVTOL) systems, creating new requirements for infrastructure planning. Vertiport siting is central, yet existing approaches such as multi-criteria decision analysis and optimization often rely on fixed criteria and seldom integrate ground transportation, which is critical for first- and last-mile access. Large language models (LLMs) show strong capabilities in reasoning and tool orchestration, but their role in siting tasks remains underexplored. This study proposes a strategy-aware LLM-based framework that connects heterogeneous spatial data with planning strategies expressed in natural language. A reflective loop connects the planner, executor, and validator for iterative refinement using two methods: multi-criteria decision analysis for interpretable mapping and a genetic algorithm for nonlinear optimization. Experiments in Los Angeles highlight both the potential and challenges of applying LLM agents to siting: outcome evaluation shows that strategies can be translated into distinct trade-offs, while process evaluation demonstrates the benefits of iterative refinement. The study suggests that LLM-based agents can formalize qualitative strategies into reproducible workflows, indicating their potential for UAM siting and promise for broader use in urban planning.

This paper presents a comprehensive review of emerging innovations and future research directions leveraging Large Language Models (LLMs) for urban data analytics, examining how cities generate, structure, and use information to support planning and operational decisions. While LLMs show promise in addressing critical challenges faced by urban stakeholders—including data integration, accessibility, and cross-domain analysis—their applications and effectiveness in urban contexts remain largely unexplored and fragmented across disciplines. Through our systematic analysis of 178 papers, we examine the impact of LLMs across the four key stages of urban data analytics: collection, preprocessing, modeling, and post-analysis. Our review encompasses various urban domains, including transportation, urban planning, disaster management, and environmental monitoring, identifying how LLMs can transform analytical approaches in these fields. We identify current trends, innovative applications, and challenges in integrating LLMs into urban analytics workflows. Based on our findings, we propose a 3E framework for future research directions: Expanding information dimensions, Enhancing model capabilities, and Executing advanced applications. This framework provides a structured approach to emphasize key opportunities in the field. Our study concludes by discussing critical challenges, including hallucination, scalability, fairness, and ethical concerns, emphasizing the need for interdisciplinary collaboration to fully realize the potential of LLMs in creating smarter, more sustainable urban environments for researchers and urban practitioners working to integrate LLMs into data-driven decision processes.