The cooperation of heterogeneous unmanned systems , for instance, between aerial engines and terrestrial engines, relies on reliable communication. Data delivery is ensured by routing protocols, but traditional routing approaches, MANET and DTN, are not efficient in such networks. In this paper, we propose the S-ROGUE routing protocol combining the paradigms MANET and DTN and switching between them according to the network connectivity. On the one hand, the S-ROGUE MANET algorithm relies on a proactive approach and a novel metric to anticipate link disruptions and detect unidirectional links. On the other hand, the S-ROGUE DTN algorithm uses on a reinforcement learning technique to select the best routing action. It implements also a replication control and packet prioritization to improve routing performances. We lead a performance evaluation of S-ROGUE with similar routing protocols in realistic simulated environments and conclude that S-ROGUE has the best routing performance regardless the scenarios.