Human–robot collaboration (HRC) systems are intelligent systems that guide robots to collaborate with humans based on a cognitive understanding of human intention, ensuring safe, flexible, and efficient collaboration between humans and robots in shared workspaces. In industrial settings, the current methods for constructing a human digital twin model rely on motion capture devices that require personnel to wear cumbersome equipment, which goes against the principle of flexible interaction advocated for HRC. Furthermore, the current methods do not model humans and robots in a unified space, which is both unintuitive and inconvenient for perceiving and understanding the overall environment. To address these limitations, this paper proposes a digital twin system for HRC. This system facilitates the construction of a digital twin scene, the mapping from the real space to the virtual space, and the planning and execution of collaborative strategies from the virtual to the real space. Designed explicitly for common workstation settings, a robust human mesh recovery algorithm is introduced to address the challenge of reconstructing occluded human bodies. Additionally, uncertainty estimation is employed to enhance the action recognition algorithm, ensuring a controllable level of risk in the recognition process. Experimental results demonstrate the superiority of the proposed methods over baseline methods. Finally, the feasibility and effectiveness of the HRC system are validated through a case study involving component assembly.

人机协作（HRC）系统是基于对人类意图的认知理解来引导机器人与人类协作的智能系统，确保人与机器人在共享工作空间中安全、灵活、高效的协作。在工业环境中，目前构建人体数字孪生模型的方法依赖于动作捕捉设备，需要人员佩戴笨重的设备，这违背了HRC倡导的灵活交互原则。此外，当前的方法没有在统一的空间中对人类和机器人进行建模，这既不直观也不方便感知和理解整体环境。为了解决这些限制，本文提出了一种 HRC 数字孪生系统。该系统有助于数字孪生场景的构建、现实空间到虚拟空间的映射以及从虚拟到现实空间的协作策略的规划和执行。专为常见工作站设置而设计，引入了强大的人体网格恢复算法来解决重建被遮挡人体的挑战。此外，还采用不确定性估计来增强动作识别算法，确保识别过程中的风险水平可控。实验结果证明了所提出的方法相对于基线方法的优越性。最后，通过涉及零部件装配的案例研究验证了HRC系统的可行性和有效性。