Aerial Multi-hop Sensor Networks

Unmanned Aerial Vehicles (UAVs) recently enabled a myriad of new applications spanning domains from personal entertainment and industrial inspection, to criminal surveillance and forest monitoring. A combination of sensor collection, wireless communication and path planning between multiple distributed agents is the natural way to support applications. Several small UAVs working collaboratively can rapidly provide extended reach, at low cost, and efficiently stream sensor information to operators on a ground station. A significant amount of previous work has addressed each of these topics independently, but in this dissertation we propose a holistic approach for joint coordination of networking and topology (placement of mobile nodes). Our thesis is that this approach improves user-interactive control of UAVs for live-streaming applications in terms of throughput, delay and reliability. In order to defend these claims, this dissertation begins by experimentally evaluating and modeling the wireless link between two UAVs, under different conditions. Due to limited link range, and the need for wide-area operation, the model is extended to encompass a multi-hop topology. We show that the performance of such networks using COTS devices is typically poor, and solutions must rely on coordination of network protocol and topology, simultaneously. At the network layer, we introduce a novel Time-division Multiple Access (TDMA) scheme called Distributed Variable Slot Protocol that relies on adaptive slot-length. We prove its convergence as well as its meliorated performance experimentally validated, namely 50% higher packet delivery. In terms of network topology, we show that without node placement control overall performance of the network is severely penalized, due to natural link asymmetries. We propose a novel protocol, named Dynamic Relay Placement, that is able to do both online link quality model-estimation and in a distributed fashion decide the best location for each network node, increasing throughput by 300%. Finally, we demonstrate the end-to-end system in a multi-vehicle monitoring mission. We show that coordination of multiple UAVs increases the sensor sampling rate up to 7 times in wide areas when compared to a naive approach. This work considers environmental constraints such as wind, as well as the intrinsic limitations of the vehicles such as maximum acceleration.