Computer vision is rapidly transforming the construction industry through digitalization and robotic automation, yet its full potential remains limited by longstanding challenges in achieving robust, deployable performance across diverse environments. At the HCI, we lead research that addresses a fundamental question: How can high-performing computer vision models be developed to adapt seamlessly to diverse and dynamic construction settings with minimal effort? To this end, our team pioneers advancements in visual foundation model development, model architecture design, training dataset optimization, and pre-deployment performance evaluation. By bridging the gap between algorithmic innovation and field applicability, our research drives the next generation of reliable and scalable vision-based systems for construction.

Language models have emerged as powerful tools for assisting construction professionals in retrieving, analyzing, and interpreting complex, long-context regulatory documents. However, most existing general-purpose language models lack construction-specific terminology and domain knowledge, often leading to hallucinations or inaccurate reasoning. Besides, their large size and heavy computational demands limit their practicality for real-world construction applications. To tackle these challenges, our research explores how to build a construction-specialized language model that integrates industry-specific terminology and knowledge while remaining lightweight, efficient, and adaptable to diverse practical conditions. We also investigate methods to transform and structure complex construction documents so that language models can better comprehend and utilize their content. Through this research, we aim to advance AI adoption in the construction industry, enabling more accurate information retrieval, regulatory compliance, and decision support.

Generative AI (GenAI) holds tremendous potential for exploring and producing creative, high-performing design alternatives. However, current GenAI systems face inherent limitations in comprehending the complex interplay among designers’ intentions, performance requirements, constraints, and regulatory conditions. At HCI, we strive to bridge this gap by developing human-centered generative design systems that seamlessly integrate intent understanding, regulatory reasoning, and performance-driven optimization. Our goal is to transform GenAI from a mere creative assistant into a collaborative design partner—one that co-evolves ideas with humans while ensuring compliance, constructability, and meaningful alignment with real-world needs.