Human-robot collaboration is a crucial research area in developing flexible and complex manufacturing systems. With industries moving to collaborative robots, the future workforce needs proper training in working with such systems. Learning factories, simulating realistic manufacturing environments, have emerged as promising solutions for such training. However, the high costs, heterogeneous hardware requirements, and safety concerns of collaborative systems pose significant challenges for implementation and training. This study takes a human-centric approach by applying digital twins and mixed reality to enhance collaborative robotics training. This is done in complex manufacturing assembly environments that involve multiple human and robot agents. The digital twin provides real-time system information to off-site collaborators and trainers, allowing for remote training and real-time decision-making through the mixed-reality interface. This enabled onsite trainees to intuitively communicate with and control robots, improving task efficiency and reducing cognitive load. The approach was applied to a gear mechanism assembly case study to evaluate the system’s effectiveness through task completion time measurement. The findings indicate that the mixed reality interface improves the ability of on-site trainees and remote trainers to coordinate with robots, leading to reduced risk and increased access to high-end robotic systems. In addition, the study highlights the potential of MR to democratize access to advanced manufacturing technologies, allowing for more inclusive and flexible training, especially through virtual learning factories. This research contributes to the growing field of human-robot interaction training by demonstrating the practical benefits of MR in multi-human-robot collaborative scenarios, paving the way for the next generation of learning factories.