Search for: All records

As robotics integrates into high-risk construction environments, cognitive load and emotional dynamics emerge as critical safety factors influencing Human–Robot Collaboration (HRC). While technical coordination is well-studied, comprehensive reviews of these psychological safety factors remain limited. This study addresses this gap through a mixed-method systematic review that combines scientometric mapping with an in-depth qualitative synthesis of 55 peer-reviewed journal articles. Five dominant research themes were identified: cognitive load and human factors, intelligent systems and cognitive architectures, social-emotional interaction, AI-driven decision-making, and emerging enabling technologies. More importantly, the findings demonstrate that cognitive and emotional processes are not independent but interact dynamically to shape safety outcomes. Building on these insights, the study develops an integrated cognitive–emotional safety framework that models how contextual factors, cognitive responses, and emotional states jointly influence worker safety during collaboration. The review further reveals a geographically concentrated evidence base, highlighting the need for broader, more inclusive research to support globally applicable safety strategies. The study advances safety science by repositioning cognitive and emotional dynamics as central, manageable risk factors in HRC. It calls for future research to develop task-specific cognitive load models, incorporate emotional safety into system design, and implement adaptive, human-aware robotic systems capable of proactively mitigating safety risks. This work provides a foundational framework for advancing safety science in technology-enhanced work environments. 

Overexertion, usually caused by manual material handling (MMH), often leads to work-related musculoskeletal disorders (WMSDs). These can be prevented during MMH tasks by redesigning them to include sufficient rest periods. Task characteristics play an important role in the redesign of the job. This study evaluated the effect of five task characteristics—weight, duration, frequency, distance, and angle of symmetry—at two levels on the maximum voluntary contraction recovery rate (Muscle Voluntary Contraction Recovery Time (MVCRT), defined as the time required for the muscle to recover to 80% of the initial Maximum Voluntary Contraction (MVC) value) of the muscles and developed a work-rest interval model based on these task characteristics. The results show that the weight and duration had a significant effect on MVCRT, increasing it by 32% and 21%, respectively. The interaction between frequency and distance also significantly impacted MVCRT. A simplified mathematical model was developed and validated by comparing a set of the model-generated data with experimental data. The simplified regression model demonstrated moderate predictive capability with an R2 value of 0.52. The ANOVA shows no significant difference between the data. The study’s results and mathematical models can be used during task design to reduce WMSDs. Because approximately 48% of the variability in MVCRT remained unexplained, direct application of the model to establish worker-specific work-rest intervals may result in underestimation or overestimation of recovery requirements. Therefore, the model should be interpreted as a preliminary estimation framework. 

Real-time prediction of human locomotion is vital for adaptive human-machine systems such as assistive exoskeletons. We present a multimodal pipeline that combines a lightweight local vision-language model for continuous first-person scene captions with a prompt-based language model that fuses three signals—user voice command, recent visual context, and short-term history of locomotion states—to infer the next mode. The model outputs a confidence score; when this score falls below a threshold, the system requests brief user clarification, improving safety and reducing silent errors. We tested the approach in a simulated live setting. The system reached 94.3% classification accuracy with a median end-to-end latency of 0.5 s, showing that low latency is compatible with high accuracy. These results indicate that the method offers a practical path to robust, context-aware, and interactive locomotion assistance and support its use as a building block for exoskeleton control and other time-critical applications. 

The construction industry struggles with persistent productivity challenges and a high incidence of work-related musculoskeletal disorders. While occupational exoskeletons present a promising technological solution, their widespread adoption is hindered by a lack of clear evidence regarding their economic viability. This study addresses this critical gap by developing and deploying a novel agent-based simulation (ABS) framework to quantify the cost-benefit trade-offs of integrating exoskeletons into construction projects. The model simulates complex interactions between heterogeneous worker agents, diverse task demands, and exoskeleton dynamics to evaluate impacts on project cost, duration, and worker health. The results demonstrate that strategic Exo adoption leads to significant reductions in total project costs by up to 6.7% and project duration by up to 7.8%. More critically, the technology drastically cuts health-related costs and injury recovery times by approximately 45%, yielding a return on investment (ROI) as high as 90% in large-scale projects. These findings provide the first quantitative, simulation-based evidence to support investment decisions, offering actionable insights for construction managers and safety planners considering the adoption of wearable assistive technologies. 

Excavator simulators provide safe training but present certain barriers to widespread adoption, including high cost, limited access, and the cognitive overload they impose on novices. In particular, our preliminary study indicates that novice users experience notable confusion and stress when first interacting with simulators' multi-purpose joystick controls, often failing initial tasks. To address these needs, this study designs a novel low-fidelity VR-based excavator simulator training program. The proposed work provides a simplified version of the simulator’s virtual environment and control design, focusing on the effortless practice of control interactions. By reducing non-essential details and complexity, the design aims to accelerate familiarization with the primary control interface. A between-group study with novice users was conducted to assess the effectiveness of the low-fidelity design on usability and cognitive load. Our results suggest that the low-fidelity environment effectively improves usability and user experience of targeted joystick control, compared to a high-fidelity system. This work provides a scalable, low-cost pre-training design that addresses workforce development challenges by building skills before using complex simulation training for construction equipment. This is a principle directly aligned with lean and operational efficiency in industrialized construction. 

Construction drawing sets are legally binding documents requiring complex multi-sheet, multi-discipline interpretation skills from practitioners. Despite the potential of multimodal large language models (MLLMs) to support plan review workflows, no systematic benchmark has evaluated their comprehension of complete construction document sets. This paper presents ConstructionQA, applied to a real 100% Construction Documents (CD) drawing set for an industrial warehouse facility, comprising approximately over 150 sheets across eight engineering disciplines. ConstructionQA structures evaluation around three progressively complex tasks: Information Extraction (IE), which tests value retrieval from single sheets using F1-BoW scoring; Sheet Reference Retrieval (SR), which tests navigation of the callout-and-reference system using F1-Sheets scoring; and Cross-Drawing Compliance (CDC), which tests multi-sheet, multi-discipline coordination reasoning using a composite score of YN Accuracy and ROUGE-L. Five MLLMs are evaluated under three input conditions: a distractor condition, in which plausible but irrelevant sheets augment the ground-truth context, a guaranteed-context condition, in which only relevant sheets are provided and a native-PDF condition. Models evaluated include GPT-5.4, Gemini 2.5 Flash, GPT-4.1 Nano, Gemma3:12b, and Qwen3-VL:8B. Results show that frontier models substantially outperform smaller models across all tasks, but performance remains insufficient for deployment in all categories. Sheet Reference Retrieval emerges as a critical bottleneck, with open-source models achieving near-zero SR scores across both conditions. These findings highlight the need for hybrid architectures that separate sheet retrieval from compliance reasoning in practical AEC deployment context. 

This paper introduces a novel multimodal action recognition framework designed to address the complexities of human activity recognition in manufacturing assembly tasks through spatio-temporal knowledge fusion. Traditional unimodal and naive multimodal fusion methods often fail to capture the intricate dependencies across space, time, and modality, especially in real-world industrial environments where actions are subtle, repetitive, and context-dependent. To overcome these challenges, we propose a unified architecture that incorporates: (i) a Multi-Stage Hierarchical Reconnection module for robust spatial and temporal feature disentanglement and reintegration; (ii) a spatio-temporal regularization technique, Optimized-MixUp (OMU), that jointly augments data along spatial and temporal axes to improve generalization; and (iii) a Cross-Modal Auxiliary Feature Learning component to enhance late fusion by exploiting modality-specific complementary information. Extensive experiments conducted on four benchmark datasets, NTU RGB+D, NTU RGB+D120, HA4M, and Northwestern-UCLA, demonstrate that our method outperforms recent state-of-the-art approaches, achieving top-1 accuracy of 98.4%, 93.5%, 92.0%, and 97.3%, respectively. These results confirm the framework’s robustness, scalability, and suitability for high-precision human activity understanding in manufacturing environments. The proposed method advances the field of information fusion by offering a principled approach to integrating heterogeneous spatio-temporal data for real-world action recognition tasks. 

In this study, we explore the performance, physiological responses, and emotional states of novice users of Simulator Excavators in a safe environment. The simulator excavator is a training system that employs haptic force feedback to provide a more realistic experience for users without the potential hazards of construction operations. The defined tasks are considered to be able to demonstrate various uses of the excavator to provide a meticulous overview of the possible capabilities of the machine. We used three different levels of excavator tasks to teach how to operate various uses of the machine, while monitoring participants’ pupil dilation (PD), gaze behavior and task performance. During the experiment, we identified potential problems contributing to participants’ poor performance. Problems that can be rooted in internal and external factors that increase stress, fatigue, and confusion. This study provides straightforward guidance for trainers to consider the factors that directly and indirectly affect the user’s performance. 

Work-related musculoskeletal disorders (WMSDs) significantly jeopardize construction worker safety, necessitating continuous ergonomic monitoring. Existing manual assessments are intermittent, while current automated solutions often struggle with the computational demands of real-time 3D analysis. This research proposes a novel framework for automated worker posture recognition utilizing skeleton-based transfer learning. By employing a co-learning strategy that fuses 2D and 3D datasets, the model effectively bridges the domain gap, enabling accurate 3D posture inference from 2D video. We introduce two variants: a high precision Pro-Pose model and a lightweight Lite-Pose model designed for edge deployment. Experimental results demonstrate state-of-the-art performance with a Mean Per Joint Position Error (MPJPE) of 44.36 mm on standard benchmarks and real-time processing speeds of 54 FPS. Integrated with the Rapid Entire Body Assessment (REBA) tool, this framework provides immediate, quantitative risk assessments to safeguard workforce health. 

Climbing A-frame ladders places high physical demand on the legs. We measured how lower-limb muscles work through each phase of climbing to identify when and where effort is greatest. Using surface electromyography, we found that the quadriceps are most active while supporting the body, the calf muscles assist during push-off, and the tibialis anterior engages mainly for foot clearance. During descent, the quadriceps appeared to playa substantialrole in controlling lowering through eccentric action. These findings provide insight into when leg support may be most needed and suggest potential timing windows for future exoskeleton assistance during ladder climbing.