Abstract: The transportation system in open-pit mines is characterized by diverse equipment, distributed operation sites, spatiotemporal coupling of operational processes, and dynamically changing environments, which make efficient management and control challenging. In addition, difficulties in system testing under extreme working conditions and insufficient perception data in specific scenarios hinder comprehensive performance evaluation, resulting in poor adaptability of decision-making and control strategies to different scenarios, weak generalization capability, and inadequate assurance of operational stability and safety, thereby affecting overall system reliability. To address these issues, a parallel autonomous haulage system in open-pit mines based on parallel theory and the vehicle–road–cloud integrated control concept was proposed. The system consisted of three components, namely the physical system, the industrial internet, and the artificial system. The artificial system adopted a three-level cloud control architecture composed of edge cloud, mine cloud, and central cloud, and achieved multi-level coordinated control through the cloud control infrastructure platform and the cloud control application platform. The cloud control infrastructure platform served as the cloud computing center and met the diverse requirements for real-time performance and service coverage of cloud control applications at different levels. The cloud control application platform included a resource repository, as well as parallel learning and parallel collaboration for scenario generation. The resource repository consisted of a model library and an algorithm library, providing methodological support for selection of modeling methods and scenario definition, as well as algorithm training and collaborative control. Parallel learning established a closed-loop data flow between the artificial system and the actual transportation system through parallel training and parallel testing that integrated virtual and real environments, thereby addressing the problems of insufficient training data and difficulty in testing under extreme scenarios and promoting system self-evolution. Parallel collaboration, based on the three-level cloud control architecture and multi-agent technology, realized collaborative perception, situation prediction, and hierarchical distributed coordinated control, effectively addressing the complexity of multi-equipment collaborative management and real-time control. The results of the case study show that the parallel training method not only effectively reduces the workload of data collection and annotation, but also significantly improves the accuracy and robustness of object detection model.