Urban rail transit plays a significant role in alleviating urban traffic congestion, and the coordinated control of multiple urban rail vehicles has been a research hotspot in recent years. The multi-vehicle coordinated computing task is limited by communication, leading to issues such as poor resource allocation balance, slow system response to environmental changes, and limited cooperative operation capabilities. The integration of 5G communication and Mobile Edge Computing (MEC) can effectively improve the real-time and accuracy of task processing, enhancing the overall system performance. This paper designs an autonomous coordinated computing architecture for urban rail vehicle operation control systems based on 5G and MEC. According to the characteristics of multi-vehicle coordinated control tasks, the problem of edge server selection in multi-vehicle coordinated computing offloading is modeled as a Multi-Armed Bandit (MAB) learning model, and a solution based on the Upper Confidence Bound (UCB) algorithm is proposed to minimize the overall energy consumption and latency of the urban rail vehicle multi-vehicle coordinated control system. Simulation results show that the proposed algorithm model has significant performance advantages in terms of average reward, best selection probability, average execution latency, and weighted total cost.