Human-centric integrated safety and quality assurance in collaborative robotic manufacturing systems

Abstract

Safety concerns severely impede industrial adoption of emerging human-robot collaborative manufacturing systems. A human-centric anomaly detection framework rooted in decision theory is proposed for integrated safety and quality assurance—which is a marked departure from earlier, quality- or safety-exclusive process control approaches. The framework adapts deep learning models to track fast robot motions from surveillance cameras and provides real-time, risk-metered alerts of anomalous trajectory deviations with theoretical guarantees. Application to a shared human-robot assembly line suggests that the framework can outperform conventional statistical process control methods in reducing safety risks and allows for straightforward extensions to more involved manufacturing settings.