The safety of workers in modular construction remains a concern due to the dynamic hazardous work environments and unawareness of the potential proximity of equipment. To avoid potential contact collisions and to provide a safe workplace, workers’ trajectory prediction is required. With recent technology advancements, the study in the area of trajectory prediction has benefited from various data-driven approaches. However, existing data-driven approaches are mostly limited to considering only the movement information of workers in the workplace, resulting in poor estimation accuracy. In this study, we propose an environment-aware worker trajectory prediction framework based on long short-term memory (LSTM) network to not only take the individual movement into account but also the surrounding information to fully exploit the context in the modular construction facilities. By incorporating worker-to-worker interactions as well as environment-to-worker interactions into our prediction model, a sequence of the worker’s future positions can be predicted. Extensive numerical tests on synthetic as well as modular construction datasets show the improved prediction performance of the proposed approach in comparison to several state-of-the-art alternatives. This study offers a systematic and flexible framework to incorporate rich contextual information into the prediction model in modular construction. The observation of how to integrate construction data analytics into a single framework could be inspiring for further future research to support robust construction safety practices.

Although much attention has been paid to the safety risk of construction sites and ergonomic risk assessment of workers, the automation of ergonomic risk assessment has not been significantly developed. This article presents a non-intrusive, automated ergonomic risk assessment approach based on computer vision, machine learning, and Rapid Entire Body Assessment (REBA). The method is called Computer-Vison Based Rapid Entire Body Analysis Estimation (CVRE). This approach is expected to realize automated monitoring and early-stage warning of ergonomic risks by automating the procedure of calculating REBA scores for construction site workers. This method consists of machine learning-based key joints and joint angles estimation of human bodies and computer-vision-based automated risk estimation. With the extensive development of machine learning and computer vision, researchers have been paying attention to assessing ergonomic risks with machine learning techniques. The proposed method has been further validated using the experimental data obtained by a motion capture system.

Modular construction sites are often reported as one of the most hazardous workplaces where the complex environments can lead to near misses and life-threatening collisions. To avoid contact collisions and provide a safe workplace, forecasting workers' trajectories on dynamic construction sites is demanding yet remains challenging. Existing approaches for trajectory prediction are mostly limited to only considering the objects moving information. In this paper, an environment-aware distance worker trajectory prediction model is designed to fully exploit the contextual information on construction sites. Incorporating the interactions among workers and distances between workers and static elements into the prediction model, the proposed approach offers a reliable prediction of worker positions. To further exploit the contextual cues, an environment-aware direction scheme taking directional information of the static elements into account is put forth. Extensive numerical tests on synthetic as well as modular construction datasets showcase the improved prediction performance of the proposed approaches in comparison to several state-of-the-art alternatives.

Construction workers are frequently exposed to postural ergonomic risks, yet conventional AI-based ergonomic risk assessment (ERA) systems generally provide risk classifications or scores rather than interpretable and interactive explanations. This study investigates whether domain adaptation can improve a generic vision–language model’s ability to identify and explain construction ergonomic risks from images. A dataset with 1900 construction image–text pairs was curated. The domain-adapted model was compared with the baseline model. Performance was evaluated using visual question answering (VQA) accuracy, nine image-captioning metrics, and a blinded human evaluation involving 50 participants with varying levels of ergonomics knowledge. ErgoChat improved performance across the nine caption-evaluation metrics and VQA. In the human evaluation, ErgoChat-generated descriptions were selected as more accurate in 81.69% of comparisons. Limitations include dataset size and class imbalance, potential VLM hallucination, and visual-perception errors. Future research will expand and balance real-site data, evaluate prompt robustness, and strengthen integration with structured ergonomic assessment procedures. The primary contribution is therefore a construction-ergonomics-specific domain-adaptation and evaluation framework that enables interactive VQA and natural-language ergonomic risk reasoning.