Human–robot collaboration (HRC) in industrial environments requires safety mechanisms that protect workers without sacrificing productivity. Conventional industrial robots typically lack built-in collaborative safety functions and contextual awareness, limiting the speed and separation monitoring (SSM) effectiveness. This paper proposes adaptive distance and affective perception technology (ADAPT), a perception-driven framework designed to enhance anticipatory and reactive safety without hardware modification.
ADAPT integrates facial emotion recognition (FER) and human pose estimation (HPE) within a dual-layer perception architecture. FER and HPE provides long-range affective monitoring and short-range spatial tracking, respectively. The system was implemented on an industrial robot using YOLO11-based models and Intel RealSense depth cameras. Experimental validation was conducted under real-time collaborative conditions to evaluate system performance, responsiveness and safety compliance.
The selected medium-scale configuration achieved stable real-time operation with total system latency below 140 ms. The FER module reached 99.26% temporal emotion-segment accuracy, while HPE module achieved a minimum mean-per-joint-position-error (MPJPE) of 23.45 mm, supporting reliable human–robot distance estimation. The integrated framework maintained compliant SSM and enabled adaptive and progressive velocity regulation during collaborative operations across varying robot speeds.
This study extends conventional industrial robot distance-based safety by integrating affective and spatial perception within a dual-layer perception framework. By combining FER with precise HPE separation monitoring, ADAPT upgrades conventional industrial robots into modular perception-aware collaborative systems without hardware modification, offering a practical and scalable enhancement for industrial HRC safety.
