CO-DESIGN FOR WIRELESS NETWORKED CONTROL OF AN

INTELLIGENT MOBILE ROBOT

Amine Mechraoui, Zeashan Hameed Khan, Jean-Marc Thiriet and Sylviane Gentil

Gipsa-Lab, Control departement INPG-UJF-CNRS, BP 46, 38402 Saint Martin d’H

eres, France

Keywords:

Autonomous mobile robots, Networked control systems, Wireless network, 802.15.4 protocol, Horizontal

handoff, TrueTime simulation.

Abstract:

This paper describes a wireless network based control of a Khepera mobile robot moving in a distributed

infrastructure. Due to critical dependence on wireless communication, a procedure for reconﬁguration of the

network is proposed as a possibility to maintain communication between control station and the mobile robot

in a successful manner. The network handoff is made under a criterion that takes into account key application

dependent performance parameters. The controlled system and the communication network are simulated

respectively with Matlab/Simulink and TrueTime.

1 INTRODUCTION

Networked control systems (NCS) in mobile robotics

are getting very popular today. With the rapid

progress in communication techniques, especially the

wireless networks, distributed control and decision

have become mandatory to reduce the onboard pro-

cessing overhead. It includes control, decision, obsta-

cle avoidance etc; as it effects the battery consump-

tion in miniature robots with space and weight as key

design constraints. However, introducing a wireless

network in the control loops presents some disadvan-

tages such as band limited channels, sampling delays

and packet dropouts (Hespanha et al., 2007). Further-

more, the mobility of the robot also adds some prob-

lems, e.g. increasing the distance between the con-

trol station and the robot increases the number of lost

packets due to decreased signal strength and increased

bit error rate (BER) (Zhu et al., 2005).

The communication architecture in mobile robotics

may be centralized, in which case there is a ﬁxed

or mobile node, which communicates with all the

other nodes, or decentralized, where individual mo-

bile nodes should ideally operate without any central

control (Schwager et al., 2007). In the decentralized

control scheme, each component solves a part of the

problem and shares memory without having a global

view of the mission. There is less emphasis on com-

putation than communication. In distributed control

systems, communication is an important parameter

and individual components don’t need to share mem-

ory (Martinez et al., 2007). In related research work,

many approaches have been used for distributed con-

trol of mobile robots. In (Fierro and Lewis, 1996) the

dynamic model of the mobile robot is controlled by

means of neural networks. In (Aicardi et al., 1995)

and (canudas de Wit and Sordalen, 1992), a nonlinear

control approach has been introduced. Another re-

search area, related to the hybrid architecture of con-

trol for autonomous navigation robots is studied in

(Benzerrouk et al., 2008).

The objective of this paper is to deﬁne a communi-

cation architecture for a mobile robot moving in a

2D space with several ﬁxed stations (wireless infras-

tructure communication). According to the position

of the robot, communication is possible in a speciﬁc

coverage area with one or several stations (see Fig.

1). When several stations are reachable, the robot

will choose one of the stations (Horizontal Handoff

(HHO) strategy (Wang et al., 2007)) that can allocate

sufﬁcient resources to ensure a good level of commu-

nication. This comprises of optimal choice of payload

and delay based on distance between robot and station

and thus maintaining the necessary Quality of Service

(QoS) in order to ensure that the Quality of Control

(QoC) is sufﬁcient (distributed control). When there

is no station in the reachable space, the robot will

have to be absolutely autonomous (embedded con-

trol), maintaining a sufﬁcient QoC despite a degrada-

tion of the communication QoS. The QoC in wireless

318

Mechraoui A., Hameed Khan Z., Thiriet J. and Gentil S. (2009).

CO-DESIGN FOR WIRELESS NETWORKED CONTROL OF AN INTELLIGENT MOBILE ROBOT.

In Proceedings of the 6th International Conference on Informatics in Control, Automation and Robotics - Robotics and Automation, pages 318-324

DOI: 10.5220/0002219303180324

 SciTePress

Figure 1: Problem description.

NCS is deﬁned as the performance delivered by each

closed-loop operation. Stability is the main property

that must be guaranteed but control error and response

time are also important to analyse.

In the literature, many researchers have proposed a

HHO strategy. They proposed making a decision

by taking into account the Received Signal Strength

(RSS), the power consumption and the cost of com-

munication (Chen et al., 2004). We propose to add

the packet dropouts caused by the propagation delay

or the distance and orientation between the robot and

the station, which have a consequence on the QoC.

The paper is organized as follows. The second section

presents a brief description of the model of the Khep-

era robot, notably the kinematic model and the dy-

namic one. In the third section, the controller design

is described and simulations of the tracking trajectory

are presented. Section 4 presents the control over Zig-

Bee wireless network and the inﬂuence of the inte-

gration of this network on control performance. After

that, the proposed HHO architecture is described with

one, two and three stations. Finally, the conclusion

and perspectives are presented.

2 MODELING AND CONTROL

This section presents the study of a unicycle Khepera

robot (Lambercy and Caprari, 2007). Consider a uni-

cycle robot (Khepera) as shown in Fig. 2. Let x, y and

θ be the state variables where x ∈ ℜ and y ∈ ℜ are the

Cartesian coordinates, θ ∈ [0,2π[ is the robot’s orien-

tation with respect to the X-axis. We consider ‘v’ and

‘ω’ respectively as the linear and the angular veloci-

ties of the robot. The kinematic of the robot can be

modeled as

˙x = v cosθ, ˙y = vsin θ,

θ = ω (1)

The kinematics model of the mobile robot has two

control inputs ω

le f t

and ω

right

i.e. the left and right

wheels velocities. These are related to the linear ve-

locity v and the angular velocity ω of the robot ac-

cording to the following equations

right

= v + Rω, V

le f t

= v − Rω (2)

where R is half the distance between the two robot’s

wheels. The dynamic model of the robot wheels is

characterized by the equations of the DC motors driv-

ing the wheels. They are represented by a ﬁrst order

model

∗

(le f t,right)

τs + 1

(3)

where U is the voltage applied to the motor and

∗

(le f t,right)

are angular velocities generated by each

motor. τ is the time constant (τ = 0.63s) and K is the

gain (K = 5.3).

2.1 Controller Design

The objective of this section is to present how to con-

trol the robot to track any trajectory. Two levels of

controllers are required. The ﬁrst one is needed to

control the angular velocities of the motors. PI con-

trollers are implemented. The second one controls

the linear and angular velocities of the robot. Let

us consider the controller presented in (Toibero et al.,

2007), where the robot can reach a desired target point

]. Errors are deﬁned as

x = x

− x,

y = y

− y (4)

and the tracking error (Eq. 5a) and the orientation

error (Eq. 5b) are calculated as

d =

(5a)

θ = θ

− θ = tan

−1

(

x) − θ (5b)

According to (Toibero et al., 2007), the following

control actions are deﬁned

v =

max

1 + |d|

d cos(

θ) (6a)

ω =

max

1 + |d|

cos(

θ)sin(

θ) + K

tanh(k

θ) (6b)

where v

max

is the maximum linear velocity that the

robot can reach (v

max

= 0.3m/s) and K

are con-

stants. Those controllers are stable according to

(Toibero et al., 2007) using the Lyapunov candidate

function V

/2 + d

/2 (7)

Figure 2: Robot Model.

CO-DESIGN FOR WIRELESS NETWORKED CONTROL OF AN INTELLIGENT MOBILE ROBOT

319

Figure 3: Real trajectory of the robot.

Fig. 3 shows the reference trajectory with initial con-

ditions [X

,θ

] = [0, 10,0] and the real trajectory

of the robot using those controllers. X, Y and θ are

measured each sampling time (T

= 0.03s). These re-

sults are obtained with Matlab/TrueTime. The PI con-

trollers are discretized and the voltage applied to the

motor is obtained with a zero order hold. The trajec-

tory is known in advance and the references x

, y

and

are updated every 0.28s.

3 CONTROL OVER NETWORK

In this section, the controller is digitized and Zigbee

wireless network is integrated (see Fig. 4). The ef-

fects of network on control system are analyzed.

3.1 WPAN 802.15.4

ZigBee is a speciﬁcation for small, low-power digital

radios based on the IEEE 802.15.4 standard for wire-

less personal area networks (WPANs). The low cost

allows the technology to be widely deployed in wire-

less control and monitoring applications. The low

power-usage allows longer life with smaller batter-

ies. The mesh networking provides high reliability

and larger range. The main reasons for adopting this

wireless network in our application is its low power

consumption and its ad-hoc networking capabilities.

Figure 4: Control over Zigbee network.

3.2 Control Over Zigbee

Control feedback loops are closed through a real-time

network as shown in Fig. 4.

To perform this study, the following conditions are

considered:

• A bit rate of 250 kb/s for Zigbee in the physical

layer.

• The sensor ﬂow and the controller ﬂow use 248

bits each. The sensor ﬂow uses 3.3% and the con-

troller one uses 3.3% of the network capacity with

a sampling period T

equal to 30 ms.

TrueTime simulator is used to simulate Zigbee Net-

work. Two tasks are programmed, the ﬁrst is the con-

troller task that generates the controller ﬂow and the

second one is a periodic sensor task that generates

sensors ﬂow. The controller task is event-triggered,

which means that the controller calculates and sends

the control signals V

and ω

when it has received all

measures X

and θ

. With the TrueTime Simula-

tion of ZigBee Network, there is only the CAP period

(Contention Access Period which uses CSMA/CA

protocol), therefore there is no priority mode like CFP

(Contention Free Period) which allows Guaranteed

Time Slots (GTS) (see (van den Bossche et al., 2007)

for more information about 802.15.4 protocol).

The sensor and control data are critical to keep the

stability of the system, hence packet losses are unde-

sirable. Practically, loss of packets cannot be entirely

eliminated in wireless networks. A lot of methods e.g.

Forward Error Correction (FEC) (Kurose and Ross,

2004) are proposed to reduce the probability of packet

losses. The retransmission of lost packets is proposed

too but for a real-time application, this is an inappro-

priate solution. When a packet has been discovered to

be lost and is retransmitted, the sensors state evolve to

a newer one and thus the retransmitted packet will be

based on the old information and the calculated con-

trol will be wrong. The solution to ensure the best

control is to do over sampling, with a sampling rate

higher than what is needed, which therefore makes

control more tolerant to packet losses. However,

this solution increases the Use Request Factor (URF)

and causes more packet delays and losses (Mechraoui

et al., 2008). The minimal sampling period to ensure

a good QoC is 0.3 s in our example. To increase the

QoC over wireless network and decrease the proba-

bility of losing packets of critical data, the sampling

period of 30 ms is required. Fig. 5 shows that the

robot can communicate with the station only within

the coverage range of the station. When it losses the

communication, it keeps the last value of the linear

and angular velocities and it deviates from the refer-

ence trajectory. The progress of WPAN and the util-

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

320

Figure 5: Trajectory of the robot with one Zigbee station.

isation of this kind of wireless networks allow the

robot to ﬁnd another station (PAN) to communicate

for control. Therefore, we assume that we have dif-

ferent stations that the robot can communicate with.

The next section explains how to change stations to

ensure sufﬁcient QoS.

4 STATION SELECTION

4.1 Handoff Decision

For several stations, mobility management is pro-

posed through redundancy in network communication

by searching for the best available protocol based on

a cost function. Horizontal and vertical handoff can

be proposed. A horizontal handoff (HHO) is the link

transfer between two network access points that use

the same network technology while vertical handoff

(VHO) is between two network access points using

different network technologies. (Chen et al., 2004)

describes a smart decision model to decide the best

network interface and best time for VHO. (Stevens-

Navarro and Wong, 2006) gives a comparison be-

tween four different VHO decision algorithms w.r.t

to bandwidth and delay for different trafﬁc classes.

(Wang et al., 1999) describes a policy-enabled hand-

off system that allows users to express policies, based

on the estimation of current network conditions and

determining the best network at any moment by char-

acteristics e.g. cost, performance and power con-

sumption. (Nkansa-Gyekye and Agbinya, 2007) pro-

poses a distributed additive weighting based VHO

mechanism to reduce the processing overhead in the

mobile terminal by delegating the calculation of hand-

off metrics for network selection to the visiting net-

works. (Angermann and Kammann, 2002) evaluates

different structures for cost functions and presents a

simulation for the inﬂuence of various parameters.

Figure 6: Network HHO strategy for a mobile robot control.

The challenge in handoff is to maintain the applica-

tion session alive while the physical connection inter-

face is changed. A generic weighted ‘Network Rating

Function’ evaluation for each network is based on

NRF

= W

RSSI

RSSI,i

NL,i

D,i

(8)

RSSI

= 1 (9)

where W

RSSI

, W

and W

are weighting functions

of RSSI, network load (NL) and delay (D) respec-

tively. N

RSSI,i

, N

NL,i

and N

D,i

indicate the scores of

interface. NRF

is between 0 and 1. W

is the weight

of the factor i, which emphasizes the importance of

each contributing factor and N

i, j

represents the nor-

malized score of the interface

for factor

The embedded logic in the mobile robot, assigns

different “weights” to the handoff decision parame-

ters in order to determine the level of importance of

each parameter for the information to be exchanged

between neighbors. In our case W

RSSI

> W

The best station connection interface at any given mo-

ment is then chosen as the one that achieves the high-

est score among all candidate interfaces.

In our scenario, factors within N

i, j

include QoS pa-

rameters i.e. the RSSI, the network load (NL) and

Delay or latency (D) of the candidate network. Now

we calculate NRF

and NRF

to choose accordingly

(which ever is greater) the station that maximizes the

Figure 7: HHO state diagram.

CO-DESIGN FOR WIRELESS NETWORKED CONTROL OF AN INTELLIGENT MOBILE ROBOT

321

decision criteria (Fig. 7).

Additionally, there is a corresponding function for

each term N

RSSI,i

, N

NL,i

, and N

D,i

, and the ranges of

the functions are bounded between 0 and 1. The func-

tions are given in Eq. (10).

RSSI,i

(10a)

NL,i

= 1/e

| 0 ≤ NL ≤ 1 (10b)

D,i

(

1 if S

is the current station & RSSI > −48dBm

0 if not

(10c)

In our case, the transmitted power is equal to −3dBm

and the receiver signal threshold is equal to −48dBm.

The RSSI is calculed as follow

(11a)

RSSI(dBm) = 10log(P

(mW )) (11b)

where P

and P

are the power in mW received and

sent respectively. d is the distance between the two

nodes in meters, and α is a parameter that can be cho-

sen to model different environments.

Fig. 8 shows the trajectory of the robot with changes

in station if necessary (QoS is the best to ensure the

best QoC). The stations positions are

• For the ﬁrst station (S

) (0, 5) (m)

• For the second station (S

) (-3, 30)

• For the last station(S

) (40, 35).

For this experiment, the weight values are W

RSSI

0.8, W

= 0.15 and W

= 0.05.

We assume that S

and S

have a common coverage

area, and S

is far. According to the initial position of

the robot (here (X

,θ

) = (0,0,0)) the robot com-

municates ﬁrst with S1. It continues its trajectory un-

til the robot detects the second station. In this case,

Khepera executes Algorithm 1.

To understand this algorithm, symbols are used. r is

the mobile robot, S

is the current station, S

{1,2,..k}

are

all the stations and S

{m,...,n}

are the stations that are

detected by the robot. We note the fact that the robot

detects a set of stations by the symbol .

When the robot is out of range of all stations, it

keeps the last values of the angular and the linear ve-

locities which is not a good solution because it can

result in an important error with respect to the refer-

ence trajectory and cause problems (see Fig.8). We

propose in this case to change the controller and use

an embedded one.

Figures 10 and 11 show the change of stations. In

Algorithm 1: Decision algorithm.

if r  {S

,..., S

} | m ∈ {1,..., k} and

n ∈ {1,..., k} | m 6= n then

calculate NRF for all S

,...S

for all i and l ∈ {m, ...,n} do

if NRF

< NRF

then

i = l

else if NRF

> NRF

then

i = i

Wait

else

i = i

end if

end for

end if

Figure 8: Trajectory of the robot with HHO with NL=0.

Figure 9: Evolution of the linear and angular velocities.

Fig.10, the simulation is made with NL=0 for all sta-

tions. The change from S

to S

depends thus on the

distance between the robot and the station. The robot

changes the station only if the RSSI of the current sta-

tion is poor. In Fig.11, the simulation is made under

the assumption that the NL

= 0.8 which means that

the network is 80% loaded and NL

= 0.2. In this

scenario the robot decides to change the station ac-

cording to the RSSI of both stations and the network

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

322

Figure 10: Evolution of communication with different sta-

tions with HHO with NL=0 (corresponds to Fig.8).

Figure 11: Evolution of communication with different sta-

tions with HHO with NL

= 0.8 and NL

= 0.2.

load of each station. Comparing Fig.10 and Fig.11 the

HHO of the station is quite sensitive to parameter NL.

4.2 Control Decision

As the robot moves outward towards another control

station, the signal strength decreases. The low value

of RSSI is sensed by the mobile robot, which then re-

quests the station for an estimation of next way-point.

The control station sends the trajectory information as

well as broadcasts the “standby” message to all con-

trol stations in the infrastructure to minimize the time

of connection of robot for the next cell. Thus, a PAN

slot is reserved in advance for the coming robot as

the trajectory is already estimated. This proposition

permits to maintain a sufﬁcient QoC despite a degra-

dation of the QoS.

The embedded control design is the same than the

distributed one. The only difference is that the sam-

pling period is changed. The controller computes

control signals each 0.3 s. This change is very im-

portant to reduce the energy consumption and also

the execution time. Algorithm 1 and 2 is executed

(“poor” in Algorithm 2 means that the RSSI of the

current station comes near the receiver signal thresh-

old). Fig.12 show the simulation when the control is

switched within the non covered area.

Figure 12: The trajectory of the robot with switching con-

troller.

Figure 13: Evolution of linear and angular velocities of the

robot with switching controller.

5 CONCLUSIONS

This work aims to study, in a co-design approach, the

inﬂuence of a wireless network QoS on the QoC of

a teleoperated robot. A wireless network based con-

trol of a unicycle mobile robot in a distributed infras-

tructure mode is described. The simulations are per-

formed using Matlab/TrueTime toolbox. The problem

faced was a loss of communication when the mobile

robot moves out of range or there is an excessive net-

work load that prohibits successful communication.

The ﬁrst strategy consists of adapting the network

QoS to the control requirements. An algorithm for

reconﬁguration by HHO is proposed as a solution

to maintain communication between control stations

and the mobile robot in multi stations scenarios. The

HHO is made with a criterion that takes into account

Algorithm 2: Decision algorithm.

if NRF = poor and r  S

then

Switch controller

end if

CO-DESIGN FOR WIRELESS NETWORKED CONTROL OF AN INTELLIGENT MOBILE ROBOT

323

the QoC for the robot as well as the QoS of wireless

network.

The second strategy deals with the reconﬁguration of

the control, for the robot to be autonomous, when the

communication link is out of order. Therefore, if the

robot is out of range of all stations, the control mode

is switched to embedded control, increasing the sam-

pling period to reduce computations and the robot is

completely autonomous. Integration of WLAN could

be a choice for extended zone coverage for mobile

robots.

In the future work, a quantiﬁcation of QoS and QoC

will be dealt. A combination of infrastructure and ad-

hoc architecture will also be investigated in order to

maintain sufﬁcient QoC in multi robots perspective.

ACKNOWLEDGEMENTS

This work is partially supported by the Safe NeCS

project funded by the French Agence Nationale de la

Recherche under grant ANR-05-SSIA-0015-03.

REFERENCES

Aicardi, M., Casalino, G., Bicchi, A., and Balestrino, A.

(1995). Closed loop steering of unicycle like vehicles

via Lyapunov techniques. In IEEE Robotics and Au-

tomation Magazine, volume 2(1), pages 27–35.

Angermann, M. and Kammann, J. (2002). Cost metrics

for decision problems in wireless ad-hoc networking.

In IEEE CAS Workshop on Wireless Communications

and Networking, Pasadena, California, USA.

Benzerrouk, A., Adouane, L., Martinet, P., and Andreff,

N. (2008). Toward an hybrid control architecture for

a mobile multi-robot systems. In Proceeding of the

National Conference on Control Architectures of

Robots, Bourges, France.

canudas de Wit, C. and Sordalen, O. (1992). Exponen-

tial stabilization of mobile robots with non-holonomic

constraints. In IEEE Trans. Automatic Control, vol-

ume 37, pages 1792–1797.

Chen, L.-J., Sun, T., Chen, B., Rajendran, V., and Gerla,

M. (2004). A smart decision model for vertical hand-

off. In Proceedings of the 4

International Workshop

on Wireless Internet and Reconﬁgurability, Athens,

Greece.

Fierro, R. and Lewis, F. (1996). Practical point stabiliza-

tion of a nonholonomic mobile robot using neural net-

works. In Proceeding Conference on Decision and

Control, Kobel, Japan.

Hespanha, J., Naghshtabrizi, P., and Xu, Y. (2007). A sur-

vey of recent results in networked control systems. In

Proceeding of IEEE special issue on Technology of

Networked Control Systems, volume 95, pages 138–

162.

Kurose, J. and Ross, K. W. (2004). Computer Networking:

A Top-Down Approach Featuring the internet. Addi-

son Wesley, 3

edition.

Lambercy, F. and Caprari, G. (2007). Khepera3 user manual

version 2.2. Technical report.

Martinez, S., Corts, J., and Bullo, F. (2007). Motion coordi-

nation with distributed information. In IEEE Control

Systems Magazine, volume 27(4), pages 75–88.

Mechraoui, A., Thiriet, J.-M., and Gentil, S. (2008). Con-

trol of an intelligent robot system over a wireless net-

work. In Proceeding of the 23

IAR Workshop on

Advanced Control and Diagnosis, Coventry, United

Kingdom.

Nkansa-Gyekye, Y. and Agbinya, J. (2007). Vertical hand-

off decision algorithm for umts-wlan. In Proceed-

ing of the 2

International Conference on Wire-

less Broadband and Ultra Wideband Communica-

tions, Sydney, Australia.

Schwager, M., Slotine, J., and Rus, D. (2007). Decen-

tralized, adaptive control for coverage with networked

robots. In Proceeding of IEEE International Confer-

ence on Robotics and Automation, Roma, Italy.

Stevens-Navarro, E. and Wong, V. (2006). Comparison be-

tween vertical handoff decision algorithms for hetero-

geneous wireless networks. In Proceeding of the 63

IEEE Vehicular Technology Conference, volume 2,

pages 947–951.

Toibero, J. M., Carelli, R., and Kuchen, B. (2007). Switch-

ing control of mobile robots for autonomous naviga-

tion in unknown environments. In Proceeding of IEEE

International Conference on Robotics and Automa-

tion, Roma, Italy.

van den Bossche, A., Val, T., and Campo, E. (2007). Proto-

typing and performance analysis of a qos mac layer

for industrial wireless network. In Proceeding of

the 7

IFAC International Conference on Fieldbuses

and Networks in Industrial and Embedded Systems,

Toulouse, France.

Wang, H., Katz, R., and Giese, J. (1999). Policy-enabled

handoffs across heterogeneous wireless networks. In

Proceeding of the 2

IEEE Workshop on Mobile

Computing Systems and Applications, New Orleans,

Louisiana, USA.

Wang, L., Chen, A., and Chen, H. (2007). Network se-

lection with joint vertical and horizontal handoff in

heterogeneous wlan and mobile wimax systems. In

Proceedings of the 63

IEEE Vehicular Technology

Conference, Melbourne, Australia.

Zhu, H., Yang, Q., and Kwak, K. (2005). Performance anal-

ysis of fast handoff with mobility prediction. In Pro-

ceeding of IEEE International Symposium on Commu-

nications and Information Technology, Beijing, China.

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

324