Indoor geometric quality inspection plays a crucial role in construction quality control. Light detection and ranging (LiDAR) can obtain point cloud data of the indoor environments in full range, which then can be utilized to efficiently extract geometric features of building elements for inspection. However, existing data collection of indoor environments using LiDAR is still manually planned and implemented, which is quite time-consuming as the scanning space is usually unknown and the movability of equipment is limited. Therefore, this paper proposes an automated approach for indoor geometric quality inspection data collection based on Building Information Model (BIM) and quadruped robot equipped with LiDAR. First, BIM with accurate geometric and semantic information of building is integrated with heuristic algorithm to automate scan planning, including scan stations and order. Second, Simultaneous Localization and Mapping (SLAM) integrates with BIM is equipped into a quadruped robot to achieve automated data collection capability. Finally, an intelligent execution module for scanning point cloud data in indoor environments is introduced. The proposed approach is validated in real indoor environments. Compared with traditional data collection methods, the experiment results show that the proposed approach can save 42% of scanning time, and the scanned point cloud data has better quality and enough density, which significantly improves the efficiency of inspection data collection for indoor geometric quality management. By unifying BIM-driven planning, SLAM-based localization, and quadruped robot mobility into a single framework, this study introduces a novel approach for large-scale automated indoor inspection, with attention to connected areas to ensure continuity across multiple spaces.

The growing complexity of construction projects demands decision processes that can integrate diverse professional perspectives across design, cost control, construction, and acceptance. Traditional sequential management approaches often leave conflicts unresolved until late stages, causing delays, cost overruns, and rework. To address this issue, this study develops a cross-stage workflow for construction projects that embeds artificial intelligence into collaborative decision-making. The workflow employs digital agents representing different roles, which engage in structured debates supported by evidence from building information modeling models, engineering drawings, cost records, and technical standards. Through iterative exchanges, alternative solutions are proposed, challenged, and refined, with the process producing decisions that are transparent, traceable, and adaptable to changing project conditions. Results from four representative scenarios confirm significant improvements over conventional practice: unresolved design clashes were reduced by more than 40%, cost forecast deviations narrowed from around 12% to below 5%, schedule variance under disruption decreased by about 18%, and first-pass acceptance rates increased from 74% to 88% with fewer reworks. A longer-term case study further demonstrated reductions in cost variance and project duration, together with higher stakeholder satisfaction. By coupling AI-enabled debate mechanisms with established digital construction environments, the workflow turns disciplinary conflicts into structured reasoning that enhances decision quality. The findings highlight how intelligent and collaborative approaches can deliver measurable gains in cost, time, and quality, while offering a practical pathway for advancing smart construction practices.

Effective scheduling in construction projects increasingly depends on allocating scarce multi-skilled labour under strict precedence and capacity constraints. Reinforcement learning (RL) techniques have presented outstanding performances in addressing Resource Constrained Project Scheduling Problem (RCPSP) instances. However, studies are lacking that utilise RL algorithms in solving extended RCPSP like the Multi-Skilled RCPSP (MSRCPSP). MSRCPSP is a nondeterministic polynomial time (NP)-hard problem that enables activity start times and allocation of multi-skilled resources to be determined simultaneously. Unlike the classical RCPSP, each activity specifies explicit skill requirements rather than anonymous capacity units. This study formulates scheduling as a Markov decision process and compares five optimisation agents, including the genetic algorithm, particle swarm optimisation, black-winged kite algorithm, deep Q-Network (DQN), and proximal policy optimisation (PPO), on benchmark instances from the MSLIB library. The results showed that solutions produced by PPO and DQN concentrate near the reference optima. PPO, in particular, attains the largest number of optimal or near-optimal schedules and the shortest makespans. In several instances, the deep reinforcement learning (DRL) agents also outperform the benchmark solutions, plausibly because CPLEX, constrained by a 60-second time limit, returns suboptimal solutions. Overall, the comparative results provide a practical benchmark of widely used heuristics and metaheuristics against DRL baselines, offering guidance for future algorithm selection and hybridisation for MSRCPSP.

Buildings account for a substantial proportion of urban energy demand, making it essential to understand the interrelationships between the built environment, urban heat island (UHI) effects, and energy demand. This study investigates the impacts of UHI on urban building energy demand in Mumbai, India, using a multi-scale framework. First, UHI intensity is assessed by generating and analyzing Land Surface Temperature maps and calculating the Urban Thermal Comfort and Vulnerability Index. This assessment identifies UHI hotspots and regions with potential thermal discomfort. Subsequently, energy demand modelling is conducted across two locations with distinct thermal comfort conditions, spanning from the individual building to the urban scale. Detailed building information and site-specific climate data from the two contrasting locations are collected to support urban building energy modelling (UBEM). Results reveal that UHI increases cooling energy demand by 7.31% at the individual building level; however, its impact on urban-scale energy demand is significantly mitigated by building geometry and surrounding structures. Specifically, variations in building height and inter-building shading outweigh the influence of UHI, leading to a substantial 15.9% reduction in the mean cooling energy demand intensity. These findings highlight the critical role of urban morphology and density in shaping cooling energy demand under UHI conditions. By extending beyond individual building-level assessments to UBEM, the study contributes new evidence regarding UHI-energy interactions in a year-round warm tropical context such as Mumbai. Furthermore, the results position urban-scale cooling energy demand assessments as a transferable framework to integrate UHI research with energy policy, ultimately supporting climate-responsive planning.

The construction industry is at a critical juncture, facing both unprecedented opportunities and challenges driven by emerging technologies like BIMS-GPT, which combines Building Information Models (BIMs) with Generative Pre-trained Transformer (GPT) language models. This study investigates the drivers and barriers to the adoption of BIMS-GPT in the construction industry through empirical research using questionnaires and expert interviews. The results indicate that the most significant drivers are BIMS-GPT’s ability to automate tasks, enhance decision-making, improve safety, and optimize processes across the entire building lifecycle. The main barriers include high development and training costs, lack of legal frameworks, data security concerns, and resistance to change among employees. Furthermore, the study analyzes differences in perceptions among respondents based on their years of experience and departmental roles. Those with 6-10 years of experience show the highest interest in adopting BIMS-GPT, while management and technical departments prioritize different aspects of the technology. The findings provide valuable insights for construction companies looking to implement BIMS-GPT and establish a solid foundation for its promotion and implementation. By understanding stakeholders’ attitudes and addressing the identified drivers and barriers, the industry can fully leverage the potential of this transformative technology, paving the way for a smarter, more efficient, and sustainable future in construction.

Sustainability in the construction industry entails measures to reduce the environmental impact of construction projects, achieve economic viability, and ensure a livable environment before, during, and after construction. However, construction activities generate substantial pollution, including dust and noise, and given their large scale, long duration, and involvement of multiple public service departments, considerable information is produced throughout each project. Traditional approaches to pollution control have proven largely ineffective, whereas platform-based models have demonstrated efficiency and cost-effectiveness in many sectors. This study therefore explores the digital transformation of public service governance for construction pollution through a platform-based approach. Timed Petri nets are employed to model and analyze both traditional and platform governance modes, providing a theoretical foundation for the digital upgrading of the construction industry using public data (e.g., construction pollution data). Results indicate that the platform governance mode reduces governance time by over 40% and substantially enhances information exchange efficiency compared with the traditional mode. By contrasting time delays between the two approaches, the superior efficiency of the platform model is highlighted. The paper further summarizes key findings, offers recommendations for institutional support and future research, and validates the proposed governance model through an empirical case study.