HIERARCHICAL MULTI-ROBOT COORDINATION

Aggregation Strategies Using Hybrid Communication

Yan Meng, Jeffrey V. Nickerson, Jing Gan

Stevens Institute of Technology, Castle Point on Hudson, Hoboken, NJ 07030, USA

Keywords: Multi-robot coordination; hybrid communications; aggregation strategy.

Abstract: Multi-robot coordination is important for searching tasks. Usually discussions of this coordination

presuppose a reliable explicit communication infrastructure. However, limited power, low radio range, and

an ever changing environment all hinder communication. Maintaining weakened connections will cause

robots to cluster during searching, which may be suboptimal with respect to the searching time. In this

paper, hierarchical-based aggregation strategies are proposed to coordinate a team of robots with limited

communication. To speed up the reconnection procedure for the proposed aggregate strategies, a hybrid

communication approach is proposed in this paper to establish a movement plan to recover the explicit

communication through vision sensors. Simulation results are presented and discussed. Experiments with 3

Pioneer robots have been conducted, and the experimental results show that our proposed strategies using a

hybrid communication mechanism are feasible and efficient in a searching task. The proposed strategies can

be extended to a large-scale searching environment as well as to a combination of humans and robots.

1 INTRODUCTION

As a community, we would like to be able to deploy

a team of robots to explore the environment in order

to assist in tasks such as searching. Most multi-robot

searching approaches assume that robots will

maintain radio (explicit) communication with each

other during the searching. However, since the on-

board wireless device of each robot has limited

power and a low radio range, producing a well

connected network with these small wireless devices

while maximizing the searching efficiency is a

challenging task, especially in a changing

environment. Mobile ad hoc networks must

continuously deal with the connectivity topology

changing. Robots may fail, robots or other elements

of the environment move around, and weather can

change which nodes are within radio range of each

other.

In an adversary environment, such as combat

environment, continuous radio communication is

easily to be attacked or hacked by the adversary. Or

in a hazardous environment, radio communication

may be very difficult, if not impossible, to perform

well due to the spectrum or signal constraints. Under

these situations, visual communication mode would

be a more appropriate and convenient way for multi

robots.

In the searching task, we eventually want the

robots to integrate information on the success of their

search. If we relax the requirement of constant

connection, the searching task can be conducted in

parallel and has the potential to cover more areas in a

given timeframe. Without planning, however, the

robots might have to search for each other after they

have completed their search and their reconnection

can not be guaranteed.

In human survival manuals, there is a simple

method recommended for coordinating after a

communication loss. Members of a team agree ahead

of time on a place to meet, called a rally point (DOD,

1992). This technique has been studied in relation to

robotic communication in emergencies (Nickerson,

2005). In the area of robotic search, the use of a

rendezvous between two searching robots at a pre-

arranged spot has been studied (Roy, 2001), drawing

from work in the theory of search games (Alpern &

Gal, 2003).

As we know, the longest searching time of a

mobile robot is totally depends on the on-board

battery. To extend the searching time of the overall

multi robot system, a power-efficient hierarchical

architecture is proposed in this paper. Based on this

architecture, several heuristic aggregation strategies

are proposed to manage the coordination between a

team of searching robots which had difficulty to

communicate. To speed up the integration procedure

289

Meng Y., V. Nickerson J. and Gan J. (2006).

HIERARCHICAL MULTI-ROBOT COORDINATION - Aggregation Strategies Using Hybrid Communication.

In Proceedings of the Third International Conference on Informatics in Control, Automation and Robotics, pages 289-295

DOI: 10.5220/0001212802890295

 SciTePress

when robots have lost radio communication, a hybrid

communication approach combining implicit

communication via vision with explicit

communication via radio is proposed. When radio

communication is broken, vision is applied to

establish a movement plan to get back into radio

connection.

2 RELATED WORK

Extensive research has been carried out on the topics

of multi-robot coordination, where communication is

critical for the success of coordination. In general,

the communication mechanism can be classified into

two categories: implicit communication and explicit

communication.

Implicit communication transmits information

through the environment or through the observation

of behaviors of other robots. Some research has been

conducted on the implicit communications (Arkin,

1992; Balch, 1994; Kuniyoshi, 1994) in multi-robot

system. Arkin (1992) indicated that explicit

communication is not always required to achieve an

increase in utility. In a follow-up study (Balch, 1994),

he concludes that "(Explicit) communication is not

essential in tasks which include implicit

communication" but that "(Explicit) communication

improves performance significantly in tasks with

little environmental communication."

(Roy and Dudek, 2001) addressed the rendezvous

problem of two heterogeneous robots with limited

communication range exploring unknown

environments. The basic idea of their approach is

that the robots have an agreed-on notion of what

constitutes a good rendezvous point. At a pre-

arranged time, the robots go to the best rendezvous

point, and wait for the other robots to arrive. They

can then fuse their map and suitably partition any

remaining exploration to be done.

Most previous work in multi-robot coordinate

relies on explicit communication to keep robots in

communication with each other (e.g. (Hu, 1998;

Pimentel, 2003)). However, in related empirical

work, it is known that the CRASAR teams at the

World Trade Center had a difficult time

communicating with their robot since at the World

Trade Center, 25% of communication between

wireless robot and control unit was extremely noisy

and therefore useless. Bandwidth problems, loss of

communication resulted in the loss of one robot

(Murphy, 2004).

One way to enhance the communication reliability

is to proactively adjust a robot’s behaviors to try to

avoid communication failure before it occurs (Arkin,

2002; Sweeney, 2002; Anderson, 2003). This method

relies on maintaining a clear line of sight between the

communicating robots. Another way is to design a

reactive approach to deal with the network failure

when it occurs so that the network can be recovered

(Ulam, 2004; Dias, 2004).

Some research has focused on architectures for

multi-robot cooperation. Grabowski et al.

(Grabowshi, 2000) consider teams of miniature

robots that overcome the limitations imposed by their

small scale by exchanging mapping and sensor

information gathered by the other robots. In this

architecture, a team leader integrates the information

gathered by the other robots. Furthermore, the leader

directs the other robots to move around obstacles or

sends them to unknown areas. Stroupe et al. (Stroupe,

2004) recently presented the MVERT-approach.

Their system uses a greedy approach that elects

robot-target pairs based on proximity. The goal of

the action selection is to maximize cooperative

progress toward mission goals.

3 AGGREGATION STRATEGIES

We assume a team of heterogeneous mobile robots

working cooperatively to explore an environment

with a preliminary map, seeking for randomly

scattered targets, where the number of the targets is

given in advance. Due to the large scale of robot

systems and large scale of the searching area, a team

of robots are divided into several sub-teams, where

each sub-team has one host and several searching

robots locally connected with short-range mobile ad

hoc network. The global communication between

the sub-team can be conducted via long-range mobile

ad hoc network between the hosts, which is shown in

Fig.1. The host robot integrates the information from

its local searching robots, and sends this collected

information to other hosts. This hierarchical

communication mechanism is power-efficient since

only low-power communication is needed for each

sub-team.

Host

Figure 1: A hierarchical structure of multi-robot system in

a searching task.

ICINCO 2006 - ROBOTICS AND AUTOMATION

290

The host robots make high-level decisions, such as

task assignments, global map building, global target

information, whereas the searching robot only holds

local perceptual data and the hosts’ status. The

robots will be dispersed to different searching areas

looking for the randomly scattered targets. The

objective is to minimize the searching time, which is

defined as the time from the starting point to the time

when the host robot receives all of the information of

the expected targets.

3.1 Static Rally Point (SRP)

Approach

Without any plan, disconnected robots might search

at random for targets, and then search at random to

find each other and compare results. Such a

technique is obviously inefficient, and so we look for

a simple organizing principle.

In the first strategy, for each sub-team, all

searching robots which have lost communication

move to a rally point when they have finished their

own searching area. At the rally point, all the

information will be exchanged and collected by the

host robot. Assuming an ad-hoc network, the robots

do not have to physically meet the host or each other,

but might stop moving at the point at which they

connect to the rally point. We call this strategy the

static rally point (SRP).

The location of the rally point for each sub-team

depends on the environment and the rally points of

other sub-teams. Usually these rally points should

be set up within the long-range communication area

between the hosts. The host assigns different

searching areas to each robot, and each robot uses its

path planner to cover their assigned area, and moves

to the rally point as soon as it finishes its searching

area or finds a target, whichever comes first. In this

approach, the host robot for each sub-team is located

on the rally point for information integration, and

does not move after stationing itself.

3.2 Mobile Rally Point (MRP)

Approach

The SRP strategy is simple to implement, but it lacks

flexibility for different target distribution

environments, especially for large scale searching

areas. Therefore, we consider a mobile rally point

(MRP) strategy. In this technique a mobile host robot

for each subteam fulfills the function of a rally point.

All of the other robots periodically reconvene at the

host robot at pre-assigned times in order to integrate

the searching information. Effectively, the robots

perform a series of synchronizations. The searching

task will be finished when the host robot has the

information of all the expected targets after a

reconvening session, which may happen before the

entire field has been explored.

To synchronize with other hosts, the navigation

path for each mobile host needs to be developed so

that the distances between the hosts are within the

long range of communication during the reconvening

session.

The overall sense of search progress of MRP will

be achieved at defined times and the hosts only need

to communicate with each other during the

reconvening session. However, robots may need to

move back and forth to the rally point more often,

which may be wasteful of energy, leading us to

consider a third strategy.

3.3 Mobile Integrator (MI)

Approach

The third strategy, which we call the mobile

integrator (MI), is designed to minimize unnecessary

movement. Only the robot who detects a target or

multiple targets will move toward and inform the

moving host robot, otherwise it will continue its own

searching task. The destination of the mobile

integrators are setup at the some preset points of the

searching area, and the host robots move

continuously and slowly throughout the search effort,

attempting to stay in the middle of the searching

crowd within each sub-team. The stop searching

command will be sent out by the host when the

searching task is over if the robots are within the

communication range, otherwise, the searching

robots will eventually stop at the preset points.

Notice that this strategy involves a tradeoff; there

will be less movement than in the previous strategy,

but at any particular time there may be less certainty

about the progress of a search and the location of the

robots as compared to the second strategy, in which

the robots synchronize periodically.

Compared with MRP method, communication cost

of MI method is higher, and the travel cost is lower.

Since movements usually consume much power than

communication, the overall power consumption of

MI should be less than MRP.

3.4 Mobile Integrator with Time-Out

(MITO) Approach

In MI approach, in the case when a searching robot

detects a target at a very early stage and then informs

the host, if the explicit communication between the

HIERARCHICAL MULTI-ROBOT COORDINATION - Aggregation Strategies Using Hybrid Communication

291

robot and the host is not available when the host

sends out the “game over” command, the robot may

search around for a long time before it finally

approaches the exit point. In order to save the energy

of the searching robot, we propose a fourth strategy,

which we call Mobile Integrator with Time-Out

(MITO), to minimize unnecessary movement after

the task is over.

The strategy is similar to the MI approach, except

that the searching robot moves toward the host for

more target information after a predefined time-out.

This time-out period may be set up according to the

size of the environment or the number of the targets.

With this time-out feature, the searching robot may

lessen the amount of unnecessary searching.

4 HOST POSITION ESTIMATION

It would be good if the searching robots could

estimate the position of the host of each sub-team

upon aggregation time. It is possible for the

searching robot to predict the host position at any

given time based on the initial planed path

information broadcasted by the hosts before

searching, with the assumption that the host robot

always moves at the same given speed.

To function effectively with an underlying

obstacle avoidance algorithm, the wavefront path

planner only transmits waypoints, not the entire path.

The wavefront planner finds the longest straight-line

distances that don't cross obstacles between cells that

are on the path. The endpoints of these straight lines

become sequential goal locations for the underlying

device driving the robot.

SP (Xs, Ys)

WP1

, Y

)

WP2

, Y

)

WP3

, Y

)

GP (Xg, Yg)

global coordinate

Figure 2: Initial planned path with three waypoints for host

robot at the entrance, where WP stands for waypoint, SP

stands for starting point, and GP stands for goal.

Assume that there are three waypoints in the initial

planed path for host robot, as shown in Fig. 2. The

time intervals between starting point to waypoint,

waypoint to waypoint, and waypoint to goal point

can be obtained by Equation (1) and the angles

between the x-axis of the global coordinate and

different waypoint phase can be obtained by

Equation (2).

vyyxxt

/)()(

−+−=Δ

vyyxxt /)()(

122

−+−=Δ

vyyxxt /)()(

233

−+−=Δ

vyyxxt /)()(

344

−+−=Δ (1)

arctg

−

θθ

arctg

−

θθ

(2)

Then the estimated position of the host robot at time t

can be obtained by the following equation.

sin)(,cos)(

when

θθ

vtytyvtxtx

+=+=

Δ≤

2121

sin)(,cos)(

twhen

θθ

vtytyvtxtx

+=+=

Δ≤≤Δ

3232

sin)(,cos)(

twhen

θθ

vtytyvtxtx

+=+=

Δ≤≤Δ

4343

sin)(,cos)(

twhen

θθ

vtytyvtxtx

+=+=

Δ≤≤Δ

(3)

Since it takes time for the searching robot to

catch up with the mobile host, it would not be

appropriate for the searching robot to set the host’s

current estimated location as the destination. Instead,

the searching robot has to predict the travel time to

the current host position from its current position,

and predict the host’s future location with this travel

time interval, and set up this host’s future location as

its new path destination.

The prediction of the time interval from the

searching robot to current location of the host, and

the estimation of future location of host can be

computed in a way similar to what is shown in Fig. 3.

If the environment dynamically changes, then the

above approach may not be able to obtain the

expected results. To minimize the accumulated

estimation error, the host would always inform all the

searching robots its current waypoint plan during

every aggregation time.

ICINCO 2006 - ROBOTICS AND AUTOMATION

292

5 A HYBRID COMMUNICATION

APPROACH

The above approach may not be able to obtain the

expected results since the initial path may be

modified due to the dynamic environmental changes,

such as some unexpected obstacles or mobile robots

on its way. A hybrid communication is proposed in

this section, where communication via vision is

applied to help in detecting and locating the host in

order to accelerate the reconnection of the radio

communication.

If the radio channel of a searching robot is broken

due to the weak radio signals or traffic jams, and the

host is still within the visual range of the searching

robot, the visual channel can detect and track the host

and guide the searching robot toward the host until

radio communication is reestablished. Sometimes,

even if the radio communication cannot be

reestablished at a very short distance, the visual

channel at least can prevent the searching robot

moving further from its teammates, so that once the

communication is available again, the robot can

exchange information immediately.

However, the vision system does not always help

in some environment, such as a highly object density

environment. Sometimes, for a very large scale

multi-robot system, the robot vision system might

often be blocked by other mobile robots if they are

not distributed far way. Under these situations, the

hybrid approach would not be faster (but would also

not be slower) than a pure radio communication

approach.

6 SIMULATION AND

EXPERIMENTAL RESULTS

6.1 Simulation Results of Hybrid

Communications

To evaluate the hybrid communication approach, a

simple proof-of-concept simulator was written using

C/C++ under Windows environment, where only two

robots are simulated: one is a lost robot and the other

is a networked robot. A city grid simulation

environment is setup, where the area is 16m by 16m

square with nine 4m by 4m square block evenly

distributed and 1m width streets in between. The lost

robot and the networked robot are distributed

randomly on the grid at their starting points. Then

both move at a speed of 1m/step to a preset

rendezvous point while searching for each other on

their way.

The simulation results with different radio ranges

are depicted in Fig. 3(b), using 100 runs for each

radio range. It can be seen that the recovery times

tends to decrease with increasing radio ranges. There

are diminishing returns once the radio coverage has

increased beyond a size where participants are likely

to connect to each other quickly.



It is noted that the scalability of the proposed

hybrid communication is limited because the chance

the robot field of view is being blocked by other

mobile robots increase dramatically with a very large

scale multi robot system.

0 1 2 3 4 5 6

-5

Radio radius (m)

Recovery Time (steps)

mean(radio-only)

mean(radio-vision)



Figure 3: Means (joined by lines) and standard deviation

values (unjoined points) of recovery times with different

radio radius when the vision radius is 15m.

6.2 Simulation Results of Aggregate

Strategies

To apply the proposed aggregate strategies to a large

scale multi-robot system, searching simulations using

10 robots are carried out. These 10 robots are

divided into two sub-teams, each sub-team has one

host and four searching robots. The searching area is

set up as an office building with 20 office rooms and

three targets are randomly distributed within these

office rooms. 100 target configurations are randomly

generated, and for each configuration, four

approaches, SRP, MRP, MI, and MITO, are

conducted. The power consumption for each robot is

calculated as

)(*)(*)(

tcktdktP

(4)

where d(t) denotes the travel distance, c(t) denotes

communication power consumption.

k and

k are

coefficients. The simulation results are shown in Fig

Aggregate Strategies

Average Searching Time (seconds)

SRP MRP MI MITO

Aggregate Strategies

Power Consumption

SRP MRP MI MITO

Figure 4: (a) average searching time comparison; (b)

average power consumption comparison.

HIERARCHICAL MULTI-ROBOT COORDINATION - Aggregation Strategies Using Hybrid Communication

293

The MITO approach outperforms other three

approaches in both average searching time and power

consumption. These simulation results demonstrate

that the proposed aggregate strategies are efficient

and scalable to a large scale multi-robot system.

6.3 Experimental Results of

Aggregate Strategies

The experiments are conducted in a small lab area

(6m x 8m). Three mobile robots are used: one

Pioneer 3DX equipped with a pan-tilt-zoom camera,

laser range finder, and 16 sonars, and two Centirbots

where each is equipped with a camera and 8 sonars.

The communication between the robots is wireless.

The radio range is setup as 1m, which can be easily

configured by exchanging the current location

information between the robots. When the distance

between each other is greater than 1m, the robots

assume that the communication failure happens;

otherwise, they are connected. Different color

cylinders are installed on top of each robot for robot

recognition using vision. The vision system can

detect the color cylinders anywhere inside the lab.

The moving speed is setup at 0.1m/sec for Pioneer

3DX and 0.05m/second for Centribots. Fig. 5

shows some snapshots of experiment using MI

strategy.

(a) Start from entrance (b) Dispersed searching

Figure 5: Snapshot of experiment using MI strategy.

The pioneer 3dx is first running around to build

the environment map and send this map to other

robots. Each robot can localize itself (Fox 1999) at

any time based on this map. And each robot also has

the navigation algorithm (Ulrich, 1998) installed to

move from one point to the destination point.

We assume that all of the robots are initially

connected through an ad hoc network and located at

the entrance, which is on left-bottom corner, and

eventually they reconvene at the left-top corner. The

period of reconvening of MRP is set at 2 minutes. A

random search approach is also conducted in the

experiment for performance comparison. Since the

MITO approach would have the same searching time

with MI approach, only MI approach is conducted on

the experiment.

As the searching performance of the MRP and the

MI strategies depends on the target distribution, four

different target distributions are manually designed in

Fig. 6, where blue stars are targets and color circles

are robots.

case 2 case 3 case 4case 1

Figure 6: Different target distributions.

15 runs for each strategy were carried out on each

configuration. To speed up the experiments, 20

minutes is set as the maximum searching time. Any

experiments which exceed 20 minutes are treated as

20 minutes long. The experimental results are

depicted in Fig. 7. The x-axis shows the 4 different

configurations of target distribution, whereas the y-

axis depicts the average searching time.

1234

Configurations

Average Searching Time

(Minutes)

SRP

MRP

RANDOM

Figure 7: Experimental results of three integration

strategies working on different target distributions.

From Fig. 7, it can be seen that, generally, the

searching times with proposed strategies have been

significantly reduced compared to those without any

strategy. The performance of MI overcomes the

other two strategies for all four target distributions.

When the targets can be detected on the early stage

of the searching, such as in case 1, the MI and MRP

have much better performance than the SRP due to

the mobility of their host, while the robots have to

wait until the rendezvous at a fixed point to learn of

the detection in SRP.

It is worth noting that although MRP may have

worse performance than SRP under some conditions

in the searching environment as in Fig. 6, the

mobility attributes of the MRP and the MI strategies

ICINCO 2006 - ROBOTICS AND AUTOMATION

294

would provide significant performance advantages

over SRP if the searching environment increases to a

large scale space. In a large scale space, the latency

caused by the SRP might create too much anxiety

back at the base. However, if a robot is abducted or

malfunctions, it is easier to detect with SRP and MRP,

while it would be difficult for the MI strategy since

there is no mandatory checkpoint, and the MITO

approach accommodates this drawback.

7 CONCLUSIONS

In this paper, four aggregation strategies are

presented for coordinating a team of robots with

limited communication power in a searching task. To

improve the efficiency of the searching procedure,

we distribute the robots in the environment as far as

possible to cover the whole area, aware we are

breaking the communication link, and let them

reconvene at some point to exchange information.

Our integration strategies have been implemented

and tested in experimental runs under different target

distribution environments using three real-world

mobile robots. Experimental results presented in this

paper suggest that our techniques can significantly

reduce the searching time with different degrees of

efficiency comparing to the randomly searching

approach. Our experiments suggest that MI has the

best search time performance compared to MRP and

SRP.

The future research topic will extend the searching

task in an unknown environment, where machine

learning techniques will be applied to learn the

environment and adaptively response to the

environment changes.

REFERENCES

Alpern, Steven; Gal, S. (2003) The Theory of Search

Games and Rendezvous. International Series in

Operations Research and Management Science, Vol.

55. Kluwer Academic Publishers,

Anderson, S., Simmons, R., & Goldberg, D. (2003).

Maintaining line of sight communications networks

between planetary rovers. Proceedings of the

Conference on Intelligent Robots and Systems (IROS).

Arkin, R.C. (1992). Cooperation without communication:

multiagent scheme-based robot navigation. J. Robotic

Systems, 9(3):351-364.

Arkin, R.C. & Diaz, J. (2002). Line-of-Sight constrained

exploration for reactive multiagent robot teams. in

AMC 7

International Workshop on Advanced Motion

Control.

Balch, T, & Arkin, R.C. (1994). Communication in

reactive multiagent robotic systems. Autonomous

Robots, 1(1), pp. 27-52.

Dias, M., Zinck, M., Zlot, R., and Stentz, A. (2004).

Robust Multirobot Coordinate in Dynamic

Environments., in Proceedings of IEEE International

Conference on Robotics and Automation, pp.3435 –

3442.

DOD (1992). US Army Survival Manual: FM 21-76. US

Department of Defense.

Grabowski, R., Navarro-Serment, L.E., Paredis, C., and

Khosla, P.K. (2000). Heterogeneous teams of modular

robots for mapping and exploration. J. Autonomous

Robot., vol. 8, no. 3, pp. 293-308.

Fox, D, Burgard, W, Dellaert, F, & Thrun, S. (1999).

Monte Carlo Localization: Efficient Position

Estimation for Mobile Robots. AAAI.

Hu, H, Kelly, I, Keating, D., & Vinagre, D. (1998).

Coordination of multiple mobile robots via

communication. Proceedings of SPIE’98 Mobile

Robots XIII Conference, pp. 94-103.

Kuniyoshi, Y, Rougeaux, S., Ishii, M., Kita, N., Sakane, S.

& Kakikura, M.. (1994). Cooperation by observation –

the framework and basic task patterns-. In Proc. IEEE

Int. Conf. Robotics and Automation, pp. 767-774.

Murphy, R. (2004). Human-Robot Interaction in Rescue

Robotics. IEEE Systems, Man and Cybernetics Part C:

Applications and Reviews, special issue on Human-

Robot Interaction, Vol. 34, No. 2.

Nickerson, J. V. (2005). A Concept of Communication

Distance and its Application to Six Situations in

Mobile Environments. IEEE Transactions on Mobile

Computing, Vo. 4, No.5, Sept./Oct. 2005, pp. 409-419.

Pimentel, B.S. & Campos, M. (2003). Cooperative

Communication in Ad Hoc Networked Mobile Robots.

Proceedings of 2003 IEEE/RSJ Intl. Conference on

Intelligent Robots and Systems, pp. 2876-2881.

Roy, N; Dudek, G. (2001). Collaborative robot exploration

and rendezvous: algorithm, performance bounds and

observations. J. Autonomous Robot., vol. 11, no. 2, pp.

117-136.

Stroupe, A.W., Ravichandran, R., & Balch, T. (2004).

Valued-based action selection for exploration and

mapping with robot teams., in Proc. IEEE Int. Conf.

Robot. Autom. (ICRA), 2004, PP. 4190-4197.

Sweeney, J., Brunette, T., Yang, Y., and Grupen, R. (2002).

Coordinated teams of reactive mobile platforms. in

Proc. of IEEE Int. Conf. on Robotics and Automation,

2002, pp. 299-304.

Ulam, P., & Arkin, R.C. (2004). When good comms go bad:

communication recovery form multi-robot teams. Proc.

2004 IEEE Int. Conf. on Robotics and Automation,

New Orleans, LA, May. 2004.

Ulrich, I & Borenstein J. (1998). VFH+: Reliable Obstacle

Avoidance for Fast Mobile Robots. in Proceedings of

the International Conference on Robotics and

Automation (ICRA’98), Belgium.

HIERARCHICAL MULTI-ROBOT COORDINATION - Aggregation Strategies Using Hybrid Communication

295