Self-reorganizing Dynamic Formations of Mobile Autonomous Robots

for Communication Network Optimization

Philip Necsulescu and Klaus Schilling

Chair of Computer Science VII, Universit¨at W¨urzburg, W¨urzburg, Germany

1 STAGE OF THE RESEARCH

This doctoral research intends to study a method to

autonomously self reorganize a formation of mobile

robots to optimize network performance.

Currently, a program has already been developed

that allows the robots to retrieve the received signal

strength of their neighbours along with positional in-

formation of themselves and their neighbours. A rout-

ing protocol has also been developed and tested that

uses the signal strength. It is intended to improve

this protocol with the addition of positional data of

each robot. Further studies will be conducted in un-

common environments, such as underground mines,

to further its applications.

Control algorithms have been developed and sim-

ulated to autonomously reorganize a small forma-

tion of car like robots to optimize communication

links. These algorithms will also be used to auto-

matically set up communication networks using drop-

pable, non-moving routing nodes. Along with being

simulated, experiments will be conducted in physical

environments.

2 OUTLINE OF OBJECTIVES

Although a lot of routing protocols for MANETs al-

ready exist and have been thoroughly simulated, there

has been little actual experimentation in realistic en-

vironments and scenarios. Using the signal strength

of each connection to establish routing is also a rarity,

especially in actual implementations on robots. Rel-

ative distance of each robot will also be used to im-

prove route selection.

Expending on studying the use of signal strength

for routing in MANETs, the PhD work will also

design and implement control algorithms for au-

tonomous topology control of swarms of robots to op-

timize communication network performance.

Both the routing algorithms and the autonomous

control of swarms will be studied in various environ-

ments, including indoors, outside, and in underground

environments.

The main objectives of this work are:

• Develop a routing protocol based on signal

strength.

• Incorporate relative distance information in rout-

ing protocol.

• MANET routing protocol experimental studies

and comparisons with various route cost calcula-

tion equations.

• Develop algorithm for placement of robots and

autonomous reorganization of their topology in a

swarm towards the goal of optimizing the commu-

nication network.

• Simulations of the robot control algorithm au-

tomation.

• Experimental studies of automatic placement of

robots and their autonomous self reorganization

towards the goal of optimizing the MANET‘s per-

formance..

3 RESEARCH PROBLEM

The ﬁeld of swarm robotics is quickly expending.

One on the key issues in controlling a swarm of robots

is the communication between the operator and the

robots, and between the robots themselves. The com-

munication network used by these swarms is typically

called a Mobile Ad-Hoc Network. (MANET).

MANETs generally use the IEEE 802.11 stan-

dard for wireless computing. Their dynamic nature

required routing protocols to be used as the routes

cannot be manually reprogrammed as with ﬁxed and

static networks. The use of signal strength of each

individual link between the agents in the MANET

requires the extraction of this information, the link

cost calculation, its propagation throughout the net-

work, and the total route cost calculation. This raises

some issues such as the amount of processing power

Necsulescu P. and Schilling K..

Self-reorganizing Dynamic Formations of Mobile Autonomous Robots for Communication Network Optimization.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

required and the amount of network overhead, espe-

cially when scaling to very large MANETs.

It is hoped that by using signal strength, along

with incorporating relative distance information for

cost calculations and studying various cost calculation

equations for different environments, the new proto-

col can improve the MANET performance under vari-

ous scenarios where other common protocols perform

poorly or fail. The main metric that is targeted is the

time needed for routing change, therefore reducing

the amount of time to establish a new route when a

link is broken. Other targeted metrics are reducing

latency and increasing bandwidth.

When controlling a swarm of robots, some of the

robots are normally autonomous. With only one op-

erator normally, only one robot can be controlled at

a time, the others must act autonomously, or remain

idle. The issue of controlling the autonomous robots,

with the scope of optimizing the MANET, is also a

main goal of this research.

4 STATE OF THE ART

Routing based on signal strength has been discussed

since the ﬁrst uses of MANETs, however, most of the

common wireless routing protocols do not make use

of this information. The proposed routing protocols

are rarely implemented and are usually only studied

in simulations.

The Signal Strength based Adaptive Routing

(SSA) Protocol is described in (Dube et al., 1997).

The protocol is a reactive protocol that selects routes

based on their continuous link connectivity duration

and the average signal strength of the connection.

SSA is composed of two separate protocols. The

Static Routing Protocol (SRP) is used to retrieve and

record the signal strength information in the signal

stability table. The second part, the Dynamic Rout-

ing Protocol (DRP), maintains the routing table.

Using the beacons sent by the wireless adapters

ﬁrmware, the signal strength is simply classiﬁed as

either strongly or weakly connected. The length of

time that the beacons have been continuously received

over that link is also stored. When a node attempts

to forward a packet, it ﬁrst searches for a route in the

routing table maintained by DRP. If the route does not

exist, a route search is initiated. When a route is bro-

ken, the node that is attempting to forward the packet

will notify the source node, which will then search for

another route. If the notiﬁcation of the broken route

is unable to arrive to the source, the source will send

a request for a new route after a timeout period.

The route selected will use only strongly con-

nected links when possible. If a route consisting

solely of strongly connected links is unavailable, it

will search for a route that includes weakly connected

links.

The SSA routing protocol is simulated and com-

pared with a simple routing algorithm that chooses

the smallest number of hops. It is determined that the

SSA routing protocol signiﬁcantly reduced the num-

ber of route reconstructions vs. the simple hop min-

imizing algorithm. The usefulness of the continuous

connectivity lifetime is found to vary greatly depend-

ing on the scenario and must be conﬁgured properly

for each case. It does not seem that this protocol has

been implemented outside of a simulation environ-

ment.

A mechanism to improve TCP performance in ad-

hoc networks using signal strength is proposed by

(Klemm et al., 2005). It is found that the causes

of 802.11 link performance degradation can be at-

tributed to congestion of the shared channel medium

or to the mobility of the nodes. The paper‘s objectives

are to reduce the packet losses due to the mobility of

nodes in MANETs.

Suggestions to reduce link failures are made to re-

act to the loss of a link. When a link failure is de-

termined to be due to a node moving out of range,

the transmission power of the network adaptor is tem-

porarily increased and a search for a new route is ini-

tiated. Care is taken not to include routes of links that

have temporarily increased transmission power. By

using signal strength information, it is possible to de-

termine when a route is approaching failure and then

pro-activelysearch for a new route. If the route degra-

dation is due to congestion and it is determined that

the neighbour is most probably in range from prior

signal strength data, the node will persist overa longer

period of time to communicate with its neighbour in-

stead of searching for better routes.

A problem is found that signal strength degrades

as the networks channel becomes congested. To de-

termine if the links degradation is due to congestion

or due to the node moving out of range, the number

of packets overheard by the node is counted. If the

count is large, then the degradation can be attributed

to channel congestion. Care must be taken not to in-

crease transmission power in this case as it can lead

to further deterioration of the communication link.

It is determined through simulations using net-

work simulator 2 that under low load, the TCP per-

formance is increased by up to 75%. On a congested

network under heavy load, TCP performance is in-

creased by 14-30%. TCP performance is determined

from the number of packets lost.

In the above papers, by using simulations, the use

Self-reorganizingDynamicFormationsofMobileAutonomousRobotsforCommunicationNetworkOptimization

of signal strength information in route ﬁnding is dis-

covered to be very beneﬁcial. It is however very rare

to ﬁnd any of these proposals implemented and tested

on real hardware. Because loss in signal strength and

network performance can be due to multiple causes,

further improvement in routing can be accomplished

by determining its cause.

(Necsulescu and K.Schilling, 2013) investigated

the cost calculation and performance evaluation for a

signal strength based MANET routing protocol. In

this paper, a signal strength based routing protocol

is evaluated. The evaluation consists of an experi-

ment where a mobile, car-like robot, is remotely op-

erated down a hallway with multiple network access

nodes located in the ofﬁces. This results in a dynamic,

multi-hop WiFi network. A camera is mounted to the

front of the robot providing live video streaming for

the operator. The performance of the protocol is then

evaluated with respect to video frames received per

section for various parameters. Results using BAT-

MAN, OLSR, and BABEL protocols are also pro-

vided for comparison.

(Qian et al., 2012) and (Nagatani et al., 2011) pre-

sented interesting results regarding the use of signal

intensity in controlling robot motion for the case of

nuclear radiation source ﬁnding. These results were

found relevant for the development of the robot con-

trol using signal intensity in this paper.

5 METHODOLOGY

5.1 Signal Strength Routing Protocol

The calculation of the cost of a link between two

nodes is derived from the free space path loss of radio

waves. The path loss can also be denoted as the lo-

cal average received signal power at the receiver node

relative to the transmission power of the transmission

node. The equation for path loss ( 1) is retrieved from

(Andersen et al., 1995):

L = L(d

) + 10∗ n∗ log

+ X

(1)

Where: L = path loss in (dB), d is distance in m,

L(d

) is the path loss at a known, close distance d0, n

is the power law relationship between distance and re-

ceived power, and X

is a zero mean Gaussian random

variable

By letting X

= 0, d

as 4π/λ, L(d

) = 0(L →

0 as d → 0), and n = 2 for free space as speciﬁed by

(Andersen et al., 1995), the path loss becomes ( 2). It

is then solved for d in ( 3), which is shownin Figure 1.

L = 20log



4πd



(2)

d =

4π

∗ 10

(3)

Through experimentation, it was determined that

the connection between two nodes becomes unusable

around 91-92dB. A cost of 1000 is therefore chosen

to represent 91dB, which corresponds to a distance

of 314.6m from equation ( 3), or d = 314.6 for L =

91dB. To assign a cost of 1000 for the value of 91dB,

( 3) is multiplied by the constant a.

cost

strength

= a ∗ d (4)

where:

a =

1000

L(91)

1000

314.5576

= 2.8334 (5)

Figure 1: Distance with respect to path loss.

Figure 2: Cost Calculated from Path Loss.

An experiment was conducted to test this protocol

and compare its performance against other commonly

used wireless routing protocols.

The experiment consists of a mobile node and

three ﬁxed nodes. The ﬁxed nodes are identical

Lenovo X61t laptops with Intel based WiFi adap-

tors. The mobile node is a 4 wheeled car like robot

which has an on-board x86 PC with an Atheros5000

based WiFi network adaptor. The laptops are run-

ning Xubuntu 12.04 and the mobile robot is running

ICINCO2014-DoctoralConsortium

Ubuntu 12.04 Server. An AXIS IP camera is mounted

to the front of the mobile robot.

An operator, using a laptop, sits in a corner ofﬁce.

The two other laptops are placed such that the opera-

tor and the farthest laptop have no direct connection.

The transmission values of the laptop and robot WiFi

adaptors were modiﬁed as well and automatic power

management turned off. The robot is then placed at

one end of the corridor and is driven remotely by the

operator to the other end. The route changes accord-

ing to the signal strengths. The initial setup of the

experiment is shown in Figure 3.

The signal strength routing protocol performed

better than OLSR and BABEL routing protocol and

was comparable to the performance of BATMAN in

this experiment. A historgram of best performance of

the FPS received for the signal strength routing proto-

col for this experiment is shown in Figure 4.

Figure 3: Signal Strenght Routing Protocol Experiment

Setup.

Figure 4: Best Result from Various Parameters of the Signal

Strength Routing Protocol.

Future experimentation using different cost func-

tions as well as more complex scenarios are planned.

5.2 Addition of Distance to Routing

Protocol

The absolute distance of each robot is determined by

using odometry. This is achieved by using the on-

board gyroscope and encoders. The absolute distance

is then kept track of by each robot through dead reck-

oning, and then broadcast throughout the MANET.

After calculating the relative distance the route

score is calculated using path loss equation stated in

(Andersen et al., 1995). Similar to the determination

of the cost function due to signal strength, the cost

function uses the same derivation for cost, however,

the right hand side of the Equation is inverted. This is

shown in ( 6).

cost

dist

= 270log

4πd

relative

(6)

Only preliminary testing of this protocol has been

ﬁnished. Further work to combine this information

along with signal strength into the routing protocol

is still needed. Future experiments are planned to test

the newcost calculation method along with trying var-

ious cost calculation equations.

5.3 Automation of Swarm Agents with a

Dynamic Topology to Optimize

Communication Network

Swarms of robots often have un-predictable forma-

tions due to their dynamic nature. When a swarm

moves or expends to a new area, the topology

changes. Due to these topology changes, the distances

between the various swarm agents also change, their

neighbours change, and therefore, the routes will also

change. In this section of the doctoral work, we look

at optimizing the topology of the swarm to improve

network communication performance. This is accom-

plished by developing control algorithms for the au-

tonomous robots such that they place themselves in

such a way as to optimize the network.

As a ﬁrst step to accomplishing this, we develop a

control algorithm and a Matlab simulation of a simple

case with two mobile robots and one operator. The

operator will operate one of the robots down a hall-

way, while the other robot is autonomous and places

itself in such a way as to maintain a good connection

with the operated robot.

5.3.1 Control Algorithm Formulation

An intermediate robot is assumed to be autonomous

with a controller that searches for locations that most

efﬁciently relay the signal from a ﬁxed station to the

tele-operated robot to increase operational range and

network performance. In contrast to the case of tra-

ditional robot control where a desired position is de-

ﬁned, in this case such a position has to be found by

the robot while moving (Qian et al., 2012) and (Na-

gatani et al., 2011).

Self-reorganizingDynamicFormationsofMobileAutonomousRobotsforCommunicationNetworkOptimization

In actual applications signal intensity at a relay

robot is obtained from measurement on board (An-

dersen et al., 1995) and (Klemm et al., 2005). For

simulations signal intensity has to be calculated as-

suming known relative distance d of the source with

regard to the relay robot.

The signal intensity I(d) in dB, at a distance d

from the signal source with intensity I(0), is subject

to a logarithmic loss 20log[

2π

d]. The radio engineer

formula that is based on the Friis transmission equa-

tion (Gaertner and Nuallain, 2012) and (Necsulescu

and K.Schilling, 2013), is

I(d) = I(0) − L(d)

= I(0)

= I(0) − 20log



2π



(cosθ − jsinθ)

(7)

It is assumed that the position vectors of the in-

termediate robot, located at (x

) with regard to the

tele-operated robot d1 , located at (X,Y), and with re-

gards to the ﬁxed base, d

are known at all times (See

Figure 5).

Figure 5: Schematic of Hallway Experiment.

= d

(cosθ

= jsinθ

) (8)

= d

(cosθ

= jsinθ

) (9)

Intermediate robot control should drive the robot

toward where the signal intensity vectors, I

and I2

are:

= I

(cosθ

= jsinθ

) (10)

= I

(cosθ

= jsinθ

) (11)

and I

are along the vectors d

and d

. While the

signal strength I

and I

are sensed on board, the an-

gles are determined from the relative position vectors

and d

The controller of the intermediate robot has to ﬁnd

the position of the robot where the difference I

− I

becomes insigniﬁcant, where:

− I

(0) − 20log



2π



(cosθ

− jsinθ

)

− (I

(0) − 20log



2π



(cosθ

− jsinθ

)

(12)

for

(X − x

)

+ (Y − y

)

(13)

+ y

(14)

Signal loss vectors are given by:

) = 20log



2π



(cosθ

− jsinθ

) (15)

) = 20log



2π



(cosθ

− jsinθ

) (16)

A second goal of the intermediate robot is to avoid

collisions during motion. Here, collision avoidance

will be presented for the case that the robots move

in a corridor of width 2D, assumed for convenience

along x-axis (See Figure 5).

The distance of the intermediate robot with regard

to the upper wall at y = D is D− y

and with regard to

the lower wall at y = −D is D+ y

(See Figure 5). Ve-

locity command to avoids collisions with these walls

is given by:

−1

D− y

+ ε

D+ y

− ε

(17)

where ε, of a very small value, is included to avoid

singularities. These functions give zero velocity com-

mand when y

= 0 and 1/ε, a very large value, when:

= D or y

= −D (18)

The velocity command vector for the intermediate

robot results as

− K{[I

(0) − L

)](cosθ

+ jsinθ

)

− [I

− L

)](cosθ

+ jsinθ

}

+ jK



−1

D− y

+ ε

D+ y

− ε



(19)

or, after separating real and imaginary part:

ICINCO2014-DoctoralConsortium

− K{[I

(0) − L

)]cosθ

− [I

− L

)](cosθ

}

+ j{[I

(0) − L

)]sin[I

(0) − L

)]sinθ

}

+ jK



−1

D− y

+ ε

D+ y

− ε



(20)

K and K

are gains to be determined at the con-

troller design stage.

In actual applications, I

and I

are obtained from

on-board signal intensity measurement, while the rel-

ative distances D− y

of the relay robot with regard

to the upper wall and with regard to the lower wall

D+ y

are obtained from on board proximity sensors.

Denoting:

= A + jB (21)

Results in:

+ B

(22)

θd = tan

−1





(23)

After each iteration the new position of the inter-

mediate robot is calculated as:

(t + ∆t) = x

(t) + v

∆tcosθ

(24)

(t + ∆t) = y

(t) + v

∆tcosθ

(25)

The results of the simulation showed that the

equations work and appear to be suitable for use in

controlling a robot.

In Figure 6, the x position of both the tele-operated

and relay robot are plotted with respect to time. It is

shown that the relay robot places itself at the point

where the signal intensities are equal.

Figure 6: X-position of tele-operated robot and relay robot

with respect to time.

In Figure 7, the magnitude of the velocity of the

relay robot is shown. The maximum speed it reaches

is 0.8m/s which is reasonable for robotus in an indoor

environment.

Figure 7: Absolute value of velocity with respect to time.

Future work is planed to do the simulation in an

experiment in the hallway used for the signal strength

routing experiment. Further work is also planned to

expend the method for more complicated scenarios

and to simulate those scenarios.

6 EXPECTED OUTCOME

At the end of the PhD work, it is hoped that the fol-

lowing is accomplished:

• A new routing protocol, customizable to use sig-

nal strength and/or distance, adaptable to the en-

vironment, and with different possible route cost

functions is completed. It is hoped that it can im-

prove performance and a variety of different envi-

ronments and scenarios.

• Methods to automate swarm agents for commu-

nication network optimization are developed and

simulated. It is hoped to also use these methods

in actual deployments and realistic scenarios with

dynamic topologies.

• To develop methods to setup wireless communi-

cation networks autonomous by using static rout-

ing nodes that are dropped by an autonomous

robot. The placement of these nodes will be op-

timized through various techniques, including the

use of 3D mapping. The above developed routing

protocols will also be tested on these networks.

• Further expansion of the different technologies to

be used in uncommon environments, such as un-

derground mines, with harsh conditions. This in-

cludes high humidity, poor lighting, and large lev-

els of airborne particles.

REFERENCES

Andersen, J., Rappaport, T., and Yoshida, S. (1995). Prop-

agation measurements and models for wireless com-

munications channels. Communications Magazine,

IEEE, 33(1):42–49.

Self-reorganizingDynamicFormationsofMobileAutonomousRobotsforCommunicationNetworkOptimization

Dube, R., Rais, C. D., yeh Wang, K., and Tripathi, S. K.

(1997). Signal stability based adaptive routing (ssa)

for ad-hoc mobile networks. IEEE Personal Commu-

nications, 4:36–45.

Gaertner, G. and Nuallain, E. O. (2012). Link Quality Pre-

diction in Mobile Ad-Hoc Networks, pages 61–94. In-

Tech.

Klemm, F., Ye, Z., Krishnamurthy, S. V., and Tripathi, S. K.

(2005). Improving tcp performance in ad hoc net-

works using signal strength based link management.

Ad Hoc Netw., 3(2):175–191.

Nagatani, K., S., K., Okada, Y., Tadokoro, S., Nishimura,

T., Yoshida, T., Koyanagi, E., and Hada, Y. (2011).

Redesign of rescue mobile robot quince. In Safety,

Security, and Rescue Robotics (SSRR), 2011 IEEE In-

ternational Symposium on, pages 13–18.

Necsulescu, P. and K.Schilling (2013). Signal strength

based manet routing protocol: Cost calculation and

performance evaluation. In Proceedings of the

3rd IFAC Symposium on Telematics Applications

(TA2013), Seoul, South Korea.

Qian, K., Song, A., Bao, J., and Zhang, H. (2012). Small

teleoperated robot for nuclear radiation and chemi-

cal leak detection. International Journal of Advanced

Robotic Systems, pages 1–9.

ICINCO2014-DoctoralConsortium