Levels of automation implemented in the railway domain are already high, especially with automated metro. Nevertheless, high levels of automation in the urban sector cannot be directly transposed to train control in mainline or freight sectors, since the train driving environment is more likely to face various unexpected events. The autonomous train is still the final objective; nevertheless, before reaching such a high level of automation, the first step will be to design intermediate levels where human operator is kept in the loop of train control. As for the scope of the TC-Rail project, presented in this paper, the design of remote driving is a necessary first stage that serves two objectives: the optimization of freight railway traffic and a way to recover from an autonomous train, as a degraded driving mode. Railway literature rarely addresses these topics as remote driving may lead to highly risky and complex situations. The goal of this paper is to identify how such risks can be addressed through a Human-machine cooperation approach. A first analysis has been conducted and some reflections are presented. The different steps proposed by this approach are illustrated by the use case stemming from the framework of the TC-Rail project.