Date of Award
6-8-2026
Document Type
Thesis
Publisher
Santa Clara : Santa Clara University, 2026
Department
General Engineering
First Advisor
Fatemeh Davoudi Kakhki
Abstract
Work-related musculoskeletal disorders (WMSDs) remain a major challenge in manufacturing environments, particularly during repetitive assembly and materialhandling tasks. Although collaborative robots (cobots) are increasingly deployed to assist workers, most existing systems operate without awareness of the worker’s ergonomic condition and therefore cannot adapt their behavior in response to elevated physical risk. This thesis presents the design, integration, and evaluation of a real-time ergonomic risk monitoring and intervention framework for human–robot collaborative manufacturing environments. The proposed system combines multimodal wearable sensing and computer vision to continuously assess worker ergonomic risk. Surface electromyography (sEMG) and inertial measurement unit (IMU) data were collected using a Delsys Trigno wireless sensing platform, while markerless posture estimation was obtained using dual Luxonis OAK-D Pro depth cameras and a YOLOv8-based pose estimation framework. Sensor-derived physiological and kinematic information were fused within an augmented Rapid Entire Body Assessment (REBA) framework to generate continuous ergonomic risk estimates. When elevated risk levels were detected, a Universal Robots UR3e collaborative robot autonomously executed assistive interventions, including task sharing and material transport support. The framework was evaluated in a controlled assembly study involving fifteen participants performing a four-wheel robotic chassis assembly task under human-only (HO) and human–robot collaboration (HRC) conditions. The vision-based posture assessment system achieved 89% agreement with manual REBA scoring across ten assembly task phases. A Random Forest classifier achieved 71.6% overall accuracy (Macro-F1 = 0.69) for three-class ergonomic risk classification using leave-one-subject-out cross-validation. Incorporating vision-derived task context features improved ergonomic risk prediction performance, increasing the coefficient of determination (R2) from 0.61 to 0.84. Furthermore, robot-assisted interventions produced a statistically significant reduction of 4.81 REBA points during high-risk reach and transport activities, demonstrating the effectiveness of adaptive collaborative assistance for reducing ergonomic exposure. User evaluations indicated strong system acceptance, with a System Usability Scale score of 77.9/100 and average robot trust ratings of 6.0 out of 7. The results demonstrate the feasibility of integrating wearable sensing, computer vision, machine learning, and collaborative robotics into a unified human-centered intervention framework capable of continuously monitoring ergonomic risk and dynamically adapting robot assistance. This work contributes toward the development of intelligent, ergonomics-aware collaborative manufacturing systems that enhance worker safety, reduce physical workload, and support human-centered Industry 5.0 production environments.
Recommended Citation
Koch, Claire; Damico, Daniel; Whistler, Samuel; and Wanebo, Andreas, "Real-Time Multimodal Ergonomic Risk Assessment System Using Human-Robot Collaboration" (2026). General Engineering. 3.
https://scholarcommons.scu.edu/gen_eng/3
