Intelligent Robotics and LLM-Based Decision Support for Sustainable Construction Operations

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

The construction industry is increasingly dependent on digital sensing, three-dimensional perception, automated decision-making, and intelligent operational control to improve productivity while reducing resource consumption and environmental impacts. Intelligent robotics can provide physical capabilities for inspection, material handling, positioning, and site monitoring, whereas large language models (LLMs) can support interpretation, reasoning, task coordination, and human–machine interaction. However, the integration of these capabilities remains constrained by the reliability of spatial perception, the complexity of construction environments, and the difficulty of translating high-level decisions into executable robotic actions. This research and review paper develops a conceptual framework for integrating intelligent robotics, three-dimensional computer vision, and LLM-based decision support for sustainable construction operations. The methodology synthesizes the provided literature on 3D deep learning, RGB-D semantic segmentation, point-cloud understanding, object detection, and instance segmentation. The analysis indicates that robust 3D perception should constitute the foundational layer of an LLM-enabled construction intelligence architecture. Point-cloud and RGB-D models can provide spatially grounded information, while LLM-based reasoning can transform such information into interpretable operational recommendations. The proposed framework positions the LLM as a decision-support and coordination layer rather than an autonomous source of physical truth. This distinction is important because sustainable construction requires decisions that simultaneously consider operational efficiency, material utilization, safety-related constraints, and environmental performance. The resulting architecture provides a theoretical basis for integrating perception, reasoning, planning, and robotic execution while recognizing limitations associated with data quality, domain transfer, computational requirements, and decision reliability.

Links and sources

Comments