Multimodal Learning and Human Digital Twins for Industrial Safety Monitoring in Human-Robot Collaborative Environments

Abstract

The transition from Industry 4.0 to Industry 5.0 marks a fundamental reorientation of manufacturing systems around human-centered collaboration, where workers and robots coexist and cooperate in shared workspaces. This paradigm shift introduces critical challenges in industrial safety monitoring: ensuring that collaborative robots respond safely and adaptively to human actions, that workers are protected from ergonomic risks and hazardous conditions, and that safety systems operate with the real-time reliability demanded by high-speed production environments. Traditional safety approaches—based on static rule-based logic and retrospective incident analysis—are fundamentally inadequate for the dynamic, unpredictable nature of human-robot collaboration. This review examines how multimodal learning—the integration of data from wearable sensors, computer vision systems, physiological monitors, and environmental sensors—combined with human digital twin architectures, is transforming industrial safety monitoring in human-robot collaborative environments. Drawing on twelve peer-reviewed works, we synthesize advances in human activity recognition (HAR) with wearable sensors, human intention recognition for real-time robot control, reinforcement learning for adaptive robotic manipulation, and worker safety digital twins for Industry 5.0. We further demonstrate how industrial sensing technologies—including four-dimensional thermal imaging, stereo phase-measuring deflectometry, and gesture-based robotic control—serve as critical sensor modalities within the multimodal safety monitoring framework. A central contribution of this review is the articulation of an integrated Human-Cobot Safety Intelligence (HCSI) paradigm that unifies multimodal perception, predictive safety analytics, and adaptive robot control for proactive, real-time industrial safety assurance.