<p>Human–robot collaboration, which integrates human dexterity with robotic precision, demonstrates significant potential for enhancing productivity. The concept of a digital twin, as an advanced technology facilitating virtual-real interactions, offers a novel framework for an intelligent human–robot collaboration paradigm. The utilization of digital twin-based approaches in human–robot collaboration has been extensively advocated to simulate collaborative processes, validate collaboration strategies, and monitor operational status. The human–robot collaboration digital twin represents a sophisticated system that encompass both physical and virtual collaboration scenarios. The capability to respond to risks within this system is essential for sustaining stable operations and ensuring the effective execution of tasks. The objective of this paper is to refine the theoretical foundations of the application of digital twins in human–robot collaboration. It introduces an innovative assessment framework for comprehensively measuring the risk response capabilities of the human–robot collaboration digital twin. This framework encompasses 4 risk response capacity levels, 5 assessment dimensions, and 18 evaluation factors. Furthermore, the paper proposes the Analyze-Evaluate-Calculate-Recommend methodology for assessing risk response capacity. The implementation of this method is elucidated through a detailed case study focused on human–robot collaborative assembly. This study presents a comprehensive approach to evaluating the risk response capabilities of digital twins in human–robot collaboration. The proposed methodology offers practitioners in the domain of human–robot collaboration a framework and criteria for determining the stability of the established digital twin systems.</p>