B.A.T.Mobile: Leveraging Mobility ControlKnowledge for Efficient Routing in Mobile Robotic

NetworksBenjamin Sliwa, Daniel Behnke, Christoph Ide and Christian Wietfeld

Communication Networks InstituteTU Dortmund University

44227 Dortmund, Germanye-mail: {Benjamin.Sliwa, Daniel.Behnke, Christoph.Ide, Christian.Wietfeld}@tu-dortmund.de

Abstract—Efficient routing is one of the key challenges ofwireless networking for unmanned autonomous vehicles (UAVs)due to dynamically changing channel and network topology char-acteristics. Various well known mobile-ad-hoc routing protocols,such as AODV, OLSR and B.A.T.M.A.N. have been proposed toallow for proactive and reactive routing decisions. In this paper,we present a novel approach which leverages application layerknowledge derived from mobility control algorithms guidingthe behavior of UAVs to fulfill a dedicated task. Thereby aprediction of future trajectories of the UAVs can be integratedwith the routing protocol to avoid unexpected route breaksand packet loss. The proposed extension of the B.A.T.M.A.N.routing protocol by a mobility prediction component – calledB.A.T.Mobile – has shown to be very effective to realize thisconcept. The results of in-depth simulation studies show that theproposed protocol reaches a distinct higher availability comparedto the established approaches and shows robust behavior even inchallenging channel conditions.

I. INTRODUCTION

Mobile robotic networks are an important subset of mobilead-hoc networks (MANETs) and form a class for a wide rangeof different network types. Applications range from dynamictraffic management in the field of vehicular ad-hoc networks(VANETs) to maintaining robust swarm communication forUnmanned Aerial Vehicles (UAVs) exploring hazardous areas.The provision of reliable end-to-end communication in thiskind of networks is a challenging topic due to the highrelative mobility. Established routing protocols can barelycope with the frequently changing network and fail to adoptto the new channel and topology conditions. This issue iswidely known and has been onesidedly addressed from twodifferent perspectives: a mobility-centric view and a routing-centric view. In this paper, we combine these approachesto enhance the overall routing performance and the stabilityof communication paths in UAV networks. Application layermobility control data is used to predict future node positions.This information is then used to enable a forward-lookingrouting approach to optimize the packet forwarding process.We analyze the system behaviour with multiple mobilityalgorithms (see Fig. 1), which are described in Section IV-A.The remainder of this paper is structured as follows: afterdiscussing the related work, we present the system modelof our solution approch, which contains the novel prediction

method and our proposed routing protocol as subchapters. Inthe next section we describe the used mobility algorithms, thetraffic model and our simulation environment. Finally, detailedresults of multiple simulation evaluations are presented, whichcompare our proposal to existing approaches. The results showthe high efficiency of our proposed methods and prove theirsuitability for highly dynamic mobile robotic networks.

II. RELATED WORK

Challenges of wireless communications with UAVs arediscussed in [1]. The high mobility is one important challengefor these networks. Recent approaches try to solve the issueby optimizing the mobile behavior but neglect the influenceon the routing. A classification about different methods ofusing mobility information to enhance the routing process isgiven in [2]. A common approach is the estimation of linkexpiration times in order to forward packets via the neighborwith the highest link-availability [3]. Most approaches assume

Swarm exploration Plume exploration

Swarm centroid

Controlled WaypointRandom Walk

Fig. 1. Typical trajectories for different mobility algorithms

©2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, including reprinting/republishing this material for advertising or promotional purposes, collecting new collected works for resale or redistribution to servers or lists, or reuse of any copyrighted

component of this work in other works.

arX

iv:1

607.

0122

3v2

[cs

.NI]

3 J

an 2

017

constant movement vectors, which cannot be fulfilled for real-world UAV exploration tasks. The usage of movement tracesin order to predict future trajectories is evaluated in [4] in thefield of VANETs. The authors of [5] use mobility predictionwith a geographic routing protocol to avoid routing voids.Knowledge about planned trajectories is used by the TrajectoryAware Geographical (TAG) routing protocol in [4] in orderto avoid link-breaks in Cognitive Radio Ad Hoc Networks(CRAHNs). The authors are able to decrease the end-to-end delay significantly. The authors of [6] present Mobilityand Load aware OLSR (ML-OLSR) as an extension to thewell-known Optimized Link State Routing (OLSR) protocoland uses information about the nodes movement direction tooptimize the multipoint relay selection. The results show animproved packet delivery ratio (PDR). An overview about bio-inspired routing protocols is given in [7] and the ’Mobilityaware-Termite’ algorithm is presented, which uses the node’sposition history to improve the path availability in combina-tion with pheromone-based routing. It is able to outperformestablished reactive protocols in terms of throughput and delay.However the node-distances are estimated from the receivedsignal strength, which will cause frequent positioning errorsin real-world channel conditions. The discussed approachesindicate the scientific interest and the relevance of the topic.Existing methods are often based on position histories andmovement directions but do not interact with the mobilitycontrol layer in order to optimze the prediction accuracy.

III. CROSS-LAYER SOLUTION APPROACH

Our proposed system model is illustrated in Fig. 2. TheUAV control software implements the mobility algorithms andacts as a database for the agent’s current mobility information,which is used by our proposed multifactoral prediction methodto determine the future agent trajectory. For the routing process

Mobility algorithms

Forward-looking routing protocol

Prediction method

Application

Improved packet delivery ratio

Controlled MobilityCW, CAE, DDD

B.A.T.Mobile(Section III-B)

Leveraging Mobility Control Knowledge

(Section III-A)

Exploration

predicted trajectory

reply: current mobility data

optimized neighbor selection

request

task definition

pro

po

sal

CW: Controlled WaypointCAE: Cluster Area Exploration [9]DDD: Distributed Dispersion Detection [11]

Goal

Fig. 2. Cross-layer solution approach: utilizing predicted mobility informa-tion for the routing decisions

the basic Better Approach To Mobile Adhoc Networking

(B.A.T.M.A.N.) [8] routing protocol is used to handle theexchange of routing messages between the agents because ofits stigmergic approach, which provides a sophisticated wayof distributing path quality information through the network.The forwarding decision is performed by our novel forward-looking metric and utilizes the results of the mobility pre-diction process in order to optimize the neighbor selection,leading to an improved packet delivery ratio. In the followingsubchapters we will give a detailled description of the novelmobility prediction algorithm and present B.A.T.Mobile as anextension to the B.A.T.M.A.N. protocol.

A. Leveraging mobility control knowledge for node trajectoryprediction

The stepwise prediction process uses different kinds ofmobility information and is illustrated for an example agent inFig. 3. A new position estimate is calculated for each next stepuntil the final iteration Np is reached. Within each iteration i,the prediction method with the assumed highest precision isselected from all available methods to perform the calculationfor the next position estimation step. Since the prediction

Prediction width

real positionfuture positionpredicted position

+

y

x

z

Obstacle

Assumed movement in waypoint direction with velocity

Extrapolation using the last positions

Waypoint rangeCurrent steering vector

Fig. 3. Application of the predicion method for an example agent

process is iterative, previously predicted position values canalso be used by the extrapolation process. Fig. 4 summarizesthe prediction algorithm as a flow chart. The steering vector ~σirepresents the vector to the desired vehicle position ~Pi+1 in thenext update step i+ 1 and is calculated by the vehicle controlsoftware. Since it is a weighted superposition of individualsteerings taking into account exploration, collision avoidanceand swarm coherence, it cannot be predicted itself and is onlyavailable in the first step of the prediction process. For theprediction of the next step, ~σi needs to be scaled from theupdate interval ∆tu of the mobility algorithm to the actualtime difference ti+1 − ti (see Eq. 1) .

~P′i+1 = ~Pi + (ti+1 − ti)

~σi∆tu

(1)

Individual vehicle waypoints act as orientation points and in-dicate the current movement direction of the agent. Dependingon the mobility algorithm the frequency of waypoint changesis variable. Eq. 2 is used to determine the next position on theline-of-sight to the current waypoint ~W with a defined velocityv.

Predict with steering vector

Use further mobility data

Target iteration reached?

Continue prediction

Steering vector available?

Current waypoint available?

Predict with waypoint direction

Predict with position extrapolation

[false]

[true]

[true]

[false]

[false]

[true]End

Start

Fig. 4. Flow Chart for the iterative prediction process for B.A.T.Mobile:in each iteration the estimation of the next position is determined by theprediction method with the highest available precision

~P′′i+1 = ~Pi +

~W − ~Pi

|| ~W − ~Pi||· (ti+1 − ti) · v (2)

The distance of the predicted position to the current waypointis contineously determined to check if the agent is inside thewaypoint range rw and can be considered ”reached”. A changeof the current waypoint is performed after a waypoint has beenreached and further waypoints are available. If no waypointsare remaining, this prediction method can no longer be used.As a fallback solution, the extrapolation prediction methodis used, if no other information is available. It calculates theaverage movement vector from the last Ne positions as shownin Eq. 3.

~P′′′i+1 = ~Pi +

ti+1 − tiNe − 1

Ne−2∑

j=0

~Pi−j − ~Pi−j−1ti−j − ti−j−1

(3)

B. Predictive routing with B.A.T.MobileThe main task of the routing process is the choice of

a forwarder node from the available neighbors for a givendestination. Most established protocols only maintain a singlepath for each destination in their routing tables, limiting the

capabilities for proactive avoidance of path losses in highlydynamic networks. Moreover the usage of simple decisionmetrics (e.g. hop count) leads to a high frequency of routeswitches. To overcome these issues, every node N uses aNeighbor Ranking (Fig. 5) for each destination D, whichcontains all neighbor nodes of N and a score as an indicatorabout their suitability to be the next hop on the path D.The routing decision is simplified to selecting the neighborwith highest score from the ranking. We use B.A.T.M.A.N.

A

B

D

C

E

0.95

0.8

0.7

0.8

0.7

0.5

Neighbor Ranking C

max

Routing Table A

Fig. 5. Score-based maintenance of multiple routing paths using neighborrankings in an example network. Node A calculates the neighbor scores forthe destination C from the received messages.

as the basic routing protocol to forward the required mobilityinformation with its Originator Messages (OGMs). All mes-sages are extended with the current forwarder position ~P , thepredicted forwarder position ~P ′ and the current path scoreS. If a node generates a broadcast message, it initializes Swith the maximum value 1 and sets both position values toits own positions. On reception of a routing packet, the newpath score S′ is calculated by multiplicating the received pathscore S with the link score SL to the forwarder node as shownin Eq. 4. This method implicitely punishes paths with highhopcounts. S′ and the position values of the receiver nodeare then used to update the routing message and forward itaccording to the rules of the basic protocol.

S′ = S · SL (4)

For the link score SL the calculation is based on the distanced to the forwarder node, which is set in relation to a maximumdistance dmax. The latter is obtained with a defined channelmodel for a desired minimal received signal strength Pe,min.Depending on the potential dynamics of the network topologyin the considered scenario, the parameter α is used to controlthe tradeoff between the influence of absolute distances andrelative mobility on the total score.

SL = min

[1−

(d

dmax

)α, 1−

(d′

dmax

)α]+ ptrend (5)

The relative agent mobility is taken into account by Eq. 6,which correlates the predicted development of the distanceto the forwarder with the maximal possible distance using thevehicle’s step width dstep. The resulting value range is definedby the parameter ptrend,max.

ptrend =

{0 : NP = 0

d′−d2·dstep·NP

· ptrend,max : else (6)

Since routing packets are usually sent with best-effort delivery,packet losses occur and may cause false routing decisions.

B.A.T.Mobile addresses this issue with the introduction of theNeighbor Score Buffer B (see Fig. 6). New path scores are

Timer

Fig. 6. Compensation of lost routing packets through score buffering

only used to update the current score candidate SC . The timertu controls the update phase, which sets the value of SC tothe score of the best received path to the destination. Afterthe timeout occurence SC is shifted to the neighbor scorebuffer and is then reset to zero. The resulting neighbor scoreis calculated as the mean of the buffer and assigned to theforwarder node in the neighbor ranking.

IV. SIMULATION-BASED SYSTEM MODEL

In this section, the simulation-based system model that isused for the performance evaluation is presented. It consists ofthe description of the mobility algorithms, the actual routingsimulation with scenario definition and the data traffic model.

A. Mobility algorithms

We use the Random walk mobility model in order to deter-mine a lower bound for the benefit through using the predictionmethod. In each iteration the movement vector is determinedrandomly. For the general performance evaluation of ourproposed methods, we use a Controlled Waypoint algorithm,which uses a trajectory of multiple random positions. Therandom positioning causes frequent changes of the networktopology, leading to challenging situations for the routingprotocols. In contrast to the well-known Random Waypointalgorithm, future waypoints are known from the beginningand can therefore be utilized by the prediction method. Im-plications for real-life UAV applications are derived from an-alyzing the routing behaviour using two different explorationalgorithms. Cluster Area Exploration (CAE) [9] performs aswarm-based exploration by selecting random waypoints forthe swarm centroid. The algorithm is used in combination withCommunication Aware Potential Fields (CAPF) [10], whichmaintains the swarm coherence and builds up a chain structureto the base station at runtime. The Distributed DispersionDetection (DDD) [11] algorithm is used for plume detectionand is able to maintain the swarm coherence on its own. Forthe exploration task, the swarm uses a mesh structure withhigh relative agent mobility.

B. Traffic model

Our reference scenario is defined by a swarm of autonomousagents exploring a mission area. One of the agents is randomlyselected to continously stream User Datagram Protocol (UDP)video data to the base station, which is centered inside theterritory. Telemetry information is periodically broadcasted byall agents in order to keep the swarm coherence and collisionavoidance steerings updated with recent data.

C. Routing simulation with OMNeT++/INETMANET

We use the discrete event-based simulation environmentOMNeT++ [12] and its INETMANET framework for theevaluation of the routing protocols. For the integration ofapplication layer mobility data, INETMANET has been en-hanced with a dedicated location service, which makes thoseinformation available for network layer routing protocols.Additionally a new base module ”GeoAssistedRoutingBase”has been added in order to provide score-based routing tablesand our novel path score metric to further protocols. Thesimulation parameters for the reference scenario are defined inTab. I. Deviations from the default assignment are explicitelymarked where they are required.

TABLE ISIMULATION PARAMETERS FOR THE REFERENCE SCENARIO IN

OMNET++/INETMANET

Simulation parameter ValueMission area 500m x 500m x 250mNumber of agents 10Mobility model Controlled WaypointVelocity v 50 km/hChannel model Friis, Nakagami (γ = 2.75)Videostream bitrate 2 Mbit/sTelemetry broadcast interval 250 msTelemetry packet size 1000 ByteOGM broadcast interval (B.A.T.M.A.N.) 0.5 sHELLO interval (OLSR) 0.5 sTopology Control (TC) interval (OLSR) 1 sMedium Access Control (MAC) layer IEEE802.11gTransport layer protocol UDPUDP Maximum Transmission Unit (MTU) 1460 ByteTransmission power 100 mWCarrier frequency 2.4 GHzReceiver sensitivity -83 dBmSimulation time per run 300 sNumber of simulation runs 50GNSS positioning error 0, [0,120] mNeighbor Score Buffer size 8Mobility update interval ∆tu 250 msExtrapolation data size Ne 5Prediction width Np 15, [0, 30]Grade of relative mobility α 7Maximum path trend ptrend,max 0.1

V. RESULTS

In this section we present the results achieved with ourproposed protocol extension. We consider the PDR of thevideo stream to the base station as our main key performanceindicator.

A. Comparison with established routing protocols

We compare the results of B.A.T.Mobile with B.A.T.M.A.N.and the established routing protocols OLSR [13] and Ad-hoc On-demand Distance Vector (AODV) [14] for differentmobility algorithms. General characteristics of the routingbehaviour can be identified analyzing the time behaviour,which is exemplarily shown in Fig. 7. Using the proposedrouting metric, B.A.T.Mobile avoids packet losses and and isable to maintain high PDR-values for longer time intervalscompared to its competitors. On rare occasions the forward-looking approach leads to situations, where the current PDRis temporarily below the value of the established protocols.Each routing decision is a trade-off between the current andthe predicted network state, therefore it is sometimes required

to spare current PDR peaks in order to avoid drops in the nearfuture.

150 160 170 180 190 200 210 220 230 240 250

Time [s]

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Cu

rren

t P

acke

t D

eliv

ery

Ratio

AODV

OLSR

B.A.T.M.A.N.

B.A.T.Mobilerouting failure of established protocols

long plateau phases

forward-looking decision with lower current PDR

Fig. 7. Example temporal behaviour comparison of B.A.T.Mobile withestablished routing protocols in urban environments

Fig. 8 shows the statistical results for a scenario withControlled Waypoint mobility in two application scenarios.We use a Friis channel model for rural scenarios and aNakagami channel model (m = 2) for urban applications.B.A.T.Mobile outperforms the established protocols signifi-cantly, while B.A.T.M.A.N. shows the lowest PDR in bothscenarios. AODV achieves the highest PDR value for the

AODV OLSR B.A.T.M.A.N. B.A.T.Mobile

Routing Protocol

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Pa

cke

t D

eliv

ery

Ra

tio

low impact of packet losses

Rural Urban

PDR decrease due to lostRREQ / RREP packets

AODV OLSR B.A.T.M.A.N. B.A.T.Mobile

Fig. 8. Comparison of the routing protocols for Controlled Waypoint mobilityin two channel models

established protocols and is fitting for mobile applications withlow packet loss probabilities due to its reactive nature. Packetlosses are more probable in urban scenarios, causing lowerPDR values for all protocols. AODV suffers heavily fromthe loss of route request (RREQ) and route reply (RREP)packets, which results in a high PDR decrease. The impactof the channel conditions on the overall routing performanceis much lower for B.A.T.Mobile than for its competitors. Thepath score is contained in all routing messages and acts anindicator for the quality of the whole path to the messageoriginator. The effect of packet losses is furthermore reduced

by the neighbor score buffer. Fig. 9 shows the resultingPDR values of B.A.T.Mobile and B.A.T.M.A.N. for differentmobility algorithms. As expected, the Random Walk routingperformance cannot be significantly improved using predictivemethods, although the score buffering slightly increases thePDR. For the Swarm exploration algorithm the general PDR isrelatively low due to the chain structure, which contains manysingle point of failure links. The best routing performance isachieved with the plume exploration algorithm. B.A.T.Mobilehighly benefits from the inherent path choice possibilities ofthe dynamical mesh structure and is able to enhance the meanPDR value above 90%.

Random Walk Swarm exploration Plume exploration

highest PDR due to high amount of possible routing paths

chain structure limits path choice possibilities

slight benefits from message buffering

B.A.T.M.A.N. B.A.T.Mobile B.A.T.M.A.N. B.A.T.Mobile B.A.T.M.A.N. B.A.T.Mobile

Fig. 9. Performance of B.A.T.Mobile in relation to B.A.T.M.A.N. in differentcontrolled mobility scenarios (Rural)

B. ParameterizationIn order to evaluate the influence of individual mobility

information on the total routing performance, we comparemultiple parameter configurations. Fig. 10 shows the resultingPDR values for the different configurations depending on theGNSS positioning accuracy. Generally we can identify threeeffects:

1) If the positioning error range is very low, all configura-tions achieve nearly equally high PDR values.

2) Individual waypoints act as static orientation points andreduce the influence of positioning errors on the overallrouting performance. Further mobility information onlybring minor benefits.

3) The extrapolation prediction method fails, if the posi-tioning error is higher than a threshold. Enhancementscan be achieved by integrating the steering vector.

The prediction width is one of the key parameters forthe performance of the routing decisions. It is depending onthe intensity of the dynamic of the network topology andinfluenced by the mobility pattern of the agents. Fig. 11 showsthe selection of the optimal prediction width for multiplemovement speed parameterizations. The possible benefit ofusing predictive methods is proportional to the velocity. For

lower speeds the general PDR is high, thus limiting the

0 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120

Positioning Error [m]

0.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

Pack

et D

eliv

ery

Ratio

no prediction

individual waypoints

high PDR for all prediction methods

SE

E

0

SWNE

SWE

WNE

WE

Configurations

Mobility informationS Steering vectorW Current waypointN Next waypointE Extrapolation0 No prediction

Fig. 10. Sensitivity analysis: Impact of positioning errors on the routingperformance for different mobility control information configurations

possible space for further improvements. For higher speedvalues the PDR can be highly increased by the predictionbut the dependency to an optimal parameter choice is alsointensified.

Velocity [km/h]

Fig. 11. Choice of the prediction width depending on the vehicle’s velocity

VI. CONCLUSION

In this paper we presented B.A.T.Mobile as a novel cross-layer approach to avoid losses of communication paths inhighly dynamic robotic networks. The proposed protocol usesa novel prediction algorithm in order to optimize the selectionof forwarder nodes by taking the relative agent mobility intoaccount. We compared our proposal with multiple establishedrouting protocols and demonstrated its superiority in differentmobility scenarios. The results of the simulation proved theability of B.A.T.Mobile to provide reliable communicationeven under challenging environmental conditions. Furthermore

we showed that the utilization of application layer mobilityinformation can significantly enhance the overall routing per-formance. In the future, we will investigate the suitability ofB.A.T.Mobile to be used in the field of VANETs applications.Additionally, we want to evaluate the usage of our predictionmethod and the path score routing metric with more routingprotocols.

ACKNOWLEDGMENTPart of the work on this paper has been supported by Deutsche Forschungs-

gemeinschaft (DFG) within the Collaborative Research Center SFB 876 “Pro-viding Information by Resource-Constrained Analysis”, project B4 and hasreceived funding from the European Union Seventh Framework Programme(FP7/2007-2013) under grant agreement n◦607832 (project SecInCoRe). Thetext reflects the authors’ views. The European Commission is not liable forany use that may be made of the information contained therein. For furtherinformation see http://www.secincore.eu/.

REFERENCES

[1] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless communications withunmanned aerial vehicles: opportunities and challenges,” IEEE Com-munications Magazine, vol. 54, no. 5, pp. 36–42, May 2016.

[2] K. S. A. Lahouari Ghouti, Tarek R. Sheltami, “Mobility prediction inmobile ad hoc networks using extreme learning machines,” in The 4thInternational Concference on Ambient Systems, Networks and Technolo-gies (ANT 2013), 2013, pp. 305–312.

[3] D. V. Le, H. Oh, and S. Yoon, “A controllable mobility (cm)-aided rout-ing protocol using mobility prediction in manets,” in 2013 InternationalConference on ICT Convergence (ICTC), Oct 2013, pp. 427–428.

[4] M. Harounabadi, A. Puschmann, O. Artemenko, and A. Mitschele-Thiel, TAG: Trajectory Aware Geographical Routing in Cognitive RadioAd Hoc Networks with UAV Nodes. Cham: Springer InternationalPublishing, 2015, pp. 111–122.

[5] L. Lin, Q. Sun, S. Wang, and F. Yang, “A geographic mobility predictionrouting protocol for ad hoc uav network,” in 2012 IEEE GlobecomWorkshops, Dec 2012, pp. 1597–1602.

[6] Y. Zheng, Y. Wang, Z. Li, L. Dong, Y. Jiang, and H. Zhang, “A mobilityand load aware olsr routing protocol for uav mobile ad-hoc networks,”in Information and Communications Technologies (ICT 2014), 2014International Conference on, May 2014, pp. 1–7.

[7] K. Manjappa and R. M. R. Guddeti, “Mobility aware-termite: a novel bioinspired routing protocol for mobile ad-hoc networks,” IET Networks,vol. 2, no. 4, pp. 188–195, December 2013.

[8] A. Neumann, C. Aichele, M. Lindner, and S. Wunderlich,“Better approach to mobile ad-hoc networking (b.a.t.m.a.n.),”Working Draft, IETF Secretariat, Internet-Draft, April 2008.[Online]. Available: http://www.ietf.org/internet-drafts/draft-wunderlich-openmesh-manet-routing-00.txt

[9] N. Goddemeier, K. Daniel, and C. Wietfeld, “Role-Based ConnectivityManagement with Realistic Air-to-Ground Channels for CooperativeUAVs,” IEEE Journal on Selected Areas in Communications (JSAC),vol. 30, no. 5, pp. 951–963, Jun. 2012.

[10] N. Goddemeier, S. Rohde, J. Pojda, and C. Wietfeld, “Evaluation ofPotential Fields Mobility Strategies for Aerial Network Provisioning,”in Proc. of IEEE Globecom Workshop on Wireless Networking forUnmanned Autonomous Vehicles. Houston, TX, USA: IEEE, Dec. 2011.

[11] D. Behnke, P.-B. Bok, and C. Wietfeld, “UAV-based Connectivity Main-tenance for Borderline Detection,” in IEEE 77th Vehicular TechnologyConference (VTC-Spring), Dresden, Germany, June 2013.

[12] A. Varga and R. Hornig, “An overview of the omnet++ simulationenvironment,” in Proceedings of the 1st International Conference onSimulation Tools and Techniques for Communications, Networks andSystems & Workshops, ser. Simutools ’08. ICST, Brussels, Belgium,Belgium: ICST (Institute for Computer Sciences, Social-Informaticsand Telecommunications Engineering), 2008, pp. 60:1–60:10. [Online].Available: http://dl.acm.org/citation.cfm?id=1416222.1416290

[13] T. Clausen and P. Jacquet, “Optimized link state routing protocol (olsr),”Internet Requests for Comments, RFC Editor, RFC 3626, October2003. [Online]. Available: http://www.rfc-editor.org/rfc/rfc3626.txt

[14] C. E. Perkins and E. M. Royer, “Ad-hoc on-demand distance vector rout-ing,” in Proceedings of the 2nd IEEE Workshop on Mobile ComputingSystems and Applications, 1997, pp. 90–100.