Today, many networks require the transmission of large traffic and hence, developing high-throughput protocols and communication platforms is of great importance. In this work, we focus on developing high-throughput routing strategies for flying ad-hoc networks (FANETs). The resource constraints and dynamic nature of the FANETs introduce critical challenges to the design of the routing protocols. We focus on a new routing framework, called routing pipe, which improves the throughput compared to routing path.Multi-beam directional antennas (MBDAs) is a promising technology which can provide fast and high-throughput data communications through concurrent multi-directional transmissions. In our first work, we have developed a novel routing scheme, called volcano routing, which benefits from MBDAs to construct multiple routing pipes that can detour around the blocked areas. The routing process is composed of three phases: Main path search phase: Several main paths with the high potential of adding side nodes, are found. After selecting top-quality main paths, the volcano pipes are established by constructing side paths around each main path. A MBDA-oriented traffic scheduling and dispatching policy is also introduced to improve the performance. In the second work, a low-cost, distributed and adaptive routing protocol is proposed, which utilizes the structure of the formation-based FANETs. It consists of three phases: An addressing system which assigns geometric coordinates to each node based on the FANET structure; A pipe routing framework that benefits from the geometric addressing to select multiple forwarding candidates; An intelligent low-complexity learning strategy which determines how the data packets should be distributed inside the pipe in order to achieve load-balanced transmissions. The results confirm that the proposed scheme can significantly improve the throughput (up to 100%) compared to the single path routing protocols. In the third work, considering the availability of a powerful centralized entity which can execute more complex algorithms, we have developed an intelligent pipe (iPipe) routing using Deep Reinforcement Learning (DRL) to optimally decide on the direction of the pipe, based on the current network state and the prediction of the immediate future. It can achieve a better performance compared to the distributed solution.