The real-time Railway Traffic Management Problem (rtRTMP) is a classical problem in the field of railway operations research. It consists in defining the passing orders and the arrival and departure times of trains in stations and selecting their route across the network. The aim is to minimize delay propagation, given a perturbed timetable. Several models and algorithms have been developed to solve this problem and provide optimized decision support systems. However, most of the systems have been tested in controlled environments without considering the train movement uncertainties that might be encountered in real implementation. For example, train driving behaviour is a source of noise for train operations that can significantly alter the traffic predictions used in the rtRTMP. In this paper, we study the robustness of the rtRTMP solution in terms of its ability to cope with train movement noise, which is inevitable in practical deployment. We consider a closed-loop framework that integrates OpenTrack, a commercial railway traffic simulator, with RECIFE-MILP, an rtRTMP solver. The solver periodically communicates with the simulator, providing optimized traffic management decisions for a given time horizon and receiving traffic feedback. The quality of traffic management decisions made in closed-loop is compared to the application of the First Come First Serve (FCFS) approach. Moreover, we compare the former with the use of RECIFE-MILP in open-loop, i.e., performing a single traffic management optimization at the beginning of the time horizon and sticking to the decisions made for the whole time. Computational experiments are performed on a portion of the French rail line between Paris and Le Havre. The analysis shows that the closed-loop rtRTMP framework significantly enhances solution robustness by reducing the impact of noise on total delay.