Advanced Robotics 24 (2010) 1–23
brill.nl/ar
Full paper
Distributed Service-Based Cooperation in Aerial/Ground
Robot Teams Applied to Fire Detection and
Extinguishing Missions
Antidio Viguria
a,∗
, Ivan Maza
b
and Anibal Ollero
a,b
a
Center for Advanced Aerospace Technologies (CATEC), 41309 Seville, Spain
b
Robotics, Vision and Control Group, University of Seville, 41092 Seville, Spain
Received 3 October 2008; accepted 4 April 2009
Abstract
This paper presents a system for the coordination of aerial and ground robots for applications such as sur-
veillance and intervention in emergency management. The overall system architecture is described. An
important part for the coordination between robots is the task allocation strategy. A distributed market-based
algorithm, called S +T, has been developed to solve the multi-robot task allocation problem in applications
that require cooperation among the robots to accomplish all the tasks. Using this algorithm, robots can pro-
vide transport and communication relay services dynamically to other robots during the missions. Moreover,
the paper presents a demonstration with a team of heterogeneous robots (aerial and ground) cooperating in
a mission of fire detection and extinguishing.
© Koninklijke Brill NV, Leiden and The Robotics Society of Japan, 2010
Keywords
Distributed task allocation, market-based coordination, auctions, multi-robot, heterogeneous teams
1. Introduction
Surveillance and intervention in disaster scenarios are very valuable missions for
robot teams. In fact, many missions are almost impossible to accomplish with
only one robot and the cooperation of heterogeneous robots is particularly use-
ful. This is the case of fire detection and extinguishing. Fire detection may require
the patrolling of robots, such as unmanned aerial vehicles, with infrared and visual
cameras or specialized fire sensors in terrains that cannot be traversed with ground
robots [3]. These aerial robots are usually very constrained in the payload required
for fire extinguishing. Then, the intervention of ground robots is needed for the ex-
*
To whom correspondence should be addressed. E-mail: aviguria@catec.aero
© Koninklijke Brill NV, Leiden and The Robotics Society of Japan, 2010 DOI:10.1163/016918609X12585524300339
2 A. Viguria et al. / Advanced Robotics 24 (2010) 1–23
tinguishing phase [4]. Furthermore, small low-cost aerial robots are also constrained
in time of flight and autonomy. Thus, they usually need to be transported to the par-
ticular areas to be patrolled. This transportation service could also be performed
by suitable ground robots that can move near to the area to be patrolled carrying
autonomous helicopters. Afterwards, those aerial vehicles could take-off and pa-
trol the area searching for the fire, and could precisely localize it by means of their
onboard cameras and sensors. Also, it is important to consider that many disaster
scenarios, including fire detection and extinguishing, do not have the appropriate
communication infrastructure or that this infrastructure could be damaged. Then,
robots providing communication relay services are also very useful. In this sce-
nario, the team of robots requires innovative cooperation methods considering their
heterogeneity and, particularly, the role of the mentioned services.
An important issue in distributed multi-robot coordination is the multi-robot task
allocation (MRTA) problem that has received a lot of attention in the last decade.
It deals with the way that tasks are distributed among robots and requires us to
define some metrics to assess the relevance of assigning given tasks to one or an-
other robot. Different approaches have been used to solve this problem: centralized
[5, 6], hybrid [7, 8] and distributed [9, 10]. Within the distributed approaches, the
market-based approach [11] has become very popular since it offers a good compro-
mise between communication requirements and the quality of the solution. It can
be considered an intermediate solution between centralized and completely distrib-
uted since it makes decisions based on inter-agent communications transmitted at
different time instances. This type of algorithm is more fault-tolerant than a central-
ized approach and can obtain more efficient solutions than a completely distributed
approach.
Market-based approaches generally assume that each task can be executed com-
pletely by a single robot. However, this could not be the case, for example, in a
surveillance or disaster scenario, in which a task consisting of transmitting images
in real-time could require another robot to act as a communication relay. Our ap-
proach to solve this problem is based on the concept of service. If a robot cannot
execute a task by itself, it asks for help and, if possible, another robot will provide
the required service. Required services are generated dynamically and are necessary
to successfully complete their associated task.
It is widely accepted that one of the main advantages of multi-robot systems
with respect to a stand-alone robot is their capability to perform tasks that can be
impossible for a single robot. In this paper, a new task allocation protocol (S +T),
designed to exploit this characteristic and based on the concept of services, is de-
scribed. This protocol is based on a distributed market-based approach and could
be considered as an extension of the SIT algorithm [12].
A similar idea is presented in Ref. [13], where soft temporal constraints were
considered using master/slave relations, and also in Ref. [14], where the efficiency
of the solution is increased considering at the same time the decomposition and
allocation of complex tasks in a distributed manner. However, the potential execu-
A. Viguria et al. / Advanced Robotics 24 (2010) 1–23 3
tion loops associated with the relation between tasks and services that could lead
to deadlock situations were not addressed in these works. In this paper, these dead-
locks are solved by a novel distributed algorithm. Moreover, the parameters of our
algorithm can be adapted to give priority to either the execution time or the energy
consumption (i.e., the sum of the distances traveled by each of the robots) in the
mission.
The main contribution of this paper is the application of a novel task allocation
algorithm, which introduces the concept of services to enable the execution of tasks
that need the cooperation of more than one robot, to a fire detection and extinguish-
ing mission with heterogeneous robots that not only involves exploration, detection
and monitoring, but also actuation in order to extinguish the fire.
This work extends the results presented in SSRR 2006 [1] and ICRA 2008
[2] where only some preliminary experiments and the distributed task allocation
(S +T) algorithm were presented. In the present paper, we have integrated the S+T
algorithm within an architecture for heterogeneous robots, and a fire extinguishing
demonstration with aerial and ground robots is presented.
The paper is organized as follows. The overall architecture of the multi-robot
team is presented in Section 2, describing the different subsystems involved: the ro-
bot themselves, the communication system, and a monitoring and planning station.
The next section is focused on the distributed task allocation system, presenting our
novel algorithm called S + T and describing the concept of tasks versus services.
Also, a deadlock problem regarding the distributed execution of synchronized tasks
is stated and our solution described. The performance of the S +T algorithm and
its different characteristics are evaluated in simulation, and the results are shown
in Section 4 with missions consisting of visiting several waypoints. The whole in-
tegrated system has been tested with promising results in a demonstration with a
team of heterogeneous robots (aerial and ground) cooperating in a mission of fire
detection and extinguishing (described in Section 5). Finally, conclusions and future
work are discussed in Section 6.
2. Multi-robot Architecture Overview
A complete system architecture has been developed in order to make the integration
of heterogeneous robots easier. Disaster scenarios usually need various robots with
different characteristics. This architecture is designed to reuse the common compo-
nents to all the robots and minimize the time needed to incorporate a new robot to
the system. A global view of the system components is presented in Fig. 1, where
three main blocks can be identified:
• Monitoring and planning station (MPS). This provides means to the human op-
erator for preparing plans, sending missions and monitoring the execution. It
also encompasses the alarm monitoring station, which is in charge of perform-
ing autonomous cooperative perception processing [3], and specialized image
processing activities (such as fire detection) providing different alarms to the
4 A. Viguria et al. / Advanced Robotics 24 (2010) 1–23
Figure 1. Global architecture illustrated with the vehicles used in the demonstration described in
Section 5. The communication network is used for both the communication between the robots, and
the communication between the robot team and the MPS.
operator. Finally, all the information related to each mission is saved in a data-
base for mission debriefing purposes.
• Communication network. This is the support for every communication between
the different components of the system. It deals with task requests/status and
data transmissions, such as images or robot telemetry. It should be noted that the
robots currently integrated in the architecture are using the BBCS (BlackBoard
Communication System) developed by the Technical University of Berlin [15]
and tested in the COMETS Project [16] funded by the IST Programme of the
European Commission. It is a robust communication system implemented via a
distributed shared memory, the blackboard (BB), in which each network node
has a local copy of the BB portion it is accessing.
• Robot team. The software architecture of each robot (see Fig. 2) is based
on hierarchical layers, with the higher levels ‘decoupled’ from the particular
characteristics of the robot. Therefore, the high-level software modules can be
reused in different types of robots with minor changes. Three layers have been
developed: RAL, MML and RIL. Since this paper only deals with high-level
aspects of the robot team behavior, only the RAL (see Fig. 2) will be com-
mented on in this section. The MML layer manages the modules implemented
in the RIL layer with the different robot functionalities. For example, the MML
layer starts and stops the necessary modules (RIL) for each task, makes the
appropriate connections between the modules, and passes to them the correct
parameters.
Regarding the task allocation process, it should be pointed out that this architec-
ture supports two different modes of operation (see Fig. 2):
A. Viguria et al. / Advanced Robotics 24 (2010) 1–23 5
Figure 2. Robot team architecture (dashed lines correspond to the manual allocation mode and solid
lines to the autonomous allocation mode). Each robot has a layered architecture with three layers:
robot abstraction layer (RAL), mobile manager layer (MML) and robot implementation layer (RIL).
The RAL layer is the one that deals with the task allocation and synchronization.
• Manual allocation mode. The human operator allocates individual elementary
tasks and sequences of elementary tasks from the MPS to the robots. Each RAL
manages those tasks and reports the robot status during the mission execution.
• Autonomous allocation mode. A Distributed Task Allocation Module (DTAM)
in the RAL allows us to autonomously negotiate task allocation in a distributed
way. In this mode, the MPS should only provide a list of elementary tasks to be
executed by a group of robots.
The system architecture supports the use of both modes of operation during a
mission execution, allowing the operator, for example, to manually allocate a task
to a given robot, whereas the rest of tasks are being allocated autonomously.
When the manual allocation mode is used, the system can be considered cen-
tralized and it is very sensitive to any communication failure. On the other hand,
when the autonomous allocation mode is used, the proposed system is based on a
distributed architecture. Therefore, communication failures will not affect seriously
the operation of the robots, allowing the execution of most of the tasks. However, it
is true that the Distributed Task Allocation algorithm decreases its efficiency since
robots that are out of communication coverage or with communication problems
will not take part in the allocation process. In fact, a robot that experiences a com-
plete failure of its communication system can no longer be considered part of the
cooperative team and it should go back to the MPS for repairing.
