Combining mobile platforms such as manned or unmanned vehicles and peer-assisted wireless communication is an enabler for a vast number of applications. A key enabler for the applications is the routing protocol that directs the packets in the network. Routing packets in fully connected mobile ad hoc networks (MANETs) has been studied to a great extent, but the assumption on full connectivity is generally not valid in a real system. This case means that a practical routing protocol must handle intermittent connectivity and the absence of end-to-end connections. In this paper, we propose a geographical routing algorithm called location-aware routing for delay-tolerant networks (LAROD), enhanced with a location service, location dissemination service (LoDiS), which together are shown to suit an intermittently connected MANET (IC-MANET). Because location dissemination takes time in IC-MANETs, LAROD is designed to route packets with only partial knowledge of geographic position. To achieve low overhead, LAROD uses a beaconless strategy combined with a position-based resolution of bids when forwarding packets. LoDiS maintains a local database of node locations, which is updated using broadcast gossip combined with routing overhearing. The algorithms are evaluated under a realistic application, i.e., unmanned aerial vehicles deployed in a reconnaissance scenario, using the low-level packet simulator ns-2. The novelty of this paper is the illustration of sound design choices in a realistic application, with holistic choices in routing, location management, and the mobility model. This holistic approach justifies that the choice of maintaining a local database of node locations is both essential and feasible. The LAROD-LoDiS scheme is compared with a leading delay-tolerant routing algorithm (spray and wait) and is shown to have a competitive edge, both in terms of delivery ratio and overhead. For spray and wait, this case involved a new packet-level implementation in ns-2 as opposed to the original connection-level custom simulator.