Abstract: Digital twin is playing an important role in manufacturing system. However, in the complex manufacturing scene for human-robot collaboration, human-robot-environment and its digital twin system show the characteristics of heterogeneous and complex tasks, dynamic environment and real-time interaction. At present, the research on intelligent methods in the digital twin collaboration process of human-robot-environment integration is poor, especially the transfer and reinforcement of digital twin model in collaboration, so as to meet the robustness and adaptive ability of manufacturing system. The paper puts forward the digital twin collaboration technology for human-robot-environment integration, and launches the scientific problem of human -robot integration in digital twin collaboration from the two cores of environment and task. Firstly, the digital twin model of collaborative assembly environment is given to provide understanding for human-robot-task interaction in the form of virtual assembly; Secondly, the corresponding spatial model and collaboration model are established to provide theoretical support for the twin collaboration of integration; Finally, taking the most typical human-robot integrated manufacturing scenario (assembly task) as an example, the transfer learning algorithm is used to provide assembly operation guidance for the robot at the decision-making level, and the reinforcement learning algorithm is used to optimize the specific execution actions of the robot. In different types of products, the corresponding human-robot collaborative assembly planning schemes can be generated, which proves the feasibility of the proposed method.

Key words: human-robot collaboration, environment understanding, digital twin, transfer learning, reinforcement learning