Abstract: This paper proposes and implements an industrial digital twin platform for multi-robot collaborative control based on multi-agent reinforcement learning. Digital twin technology significantly enhances the efficiency of intelligent manufacturing through real-time interaction between digital and physical systems. This paper focuses on the construction process of the digital twin platform framework. The platform first trains a policy model in a simulation environment and then deploys the trained model in a physical system, enabling efficient and precise collaborative control between robotic arms. The results demonstrate that the real-time motion information of the twin robotic arms in the digital space can be fed back to the physical space, tracking and reflecting the status of real robotic arms during the assembly process. This interaction between the virtual and physical spaces enables synchronized control across both domains.