In this paper we examine mobile ad-hoc networks (MANET) composed by unmanned aerial vehicles (UAVs). Due to the high-mobility of the nodes, these networks are very dynamic and the existing routing protocols partly fail to provide a reliable communication.We present Predictive-OLSR an extension to the Optimized Link-State Routing (OLSR) protocol: it enables efficient routing in very dynamic conditions. The key idea is to exploit GPS information to aid the routing protocol. Predictive-OLSR weights the expected transmission count (ETX) metric, taking into account the relative speed between the nodes. We provide numerical results obtained by a MAC-layer emulator that integrates a flight simulator to reproduce realistic flight conditions. These numerical results show that Predictive-OLSR significantly outperforms OLSR and BABEL, providing a reliable communication even in very dynamic conditions.