Abstract— This work presents an architecture that can help
to increase the reliability in groups of cooperative mobile
robots by taking advantage of analytical and sensor
redundancy. First, the design of the architecture is portrayed
and the faults to be detected are described. The different layers
of the system are then explained and analyzed, using
simulations to test their capabilities and limitations. Finally, the
architecture is implemented on a group of small mobile robots
to validate the results of the simulations.
I. INTRODUCTION
HE use of robotic systems in environments hostile to
human beings or in dangerous tasks, such as land mines
extraction and rescue operations, has increased over the last
years, making the work safer for human operators. But the
advantages of operating with robotic systems are cut back
when faults are taken into account, as they disable the robot
in some of its functions, or in worst cases, they make the
robot unable to work at all. Recent studies show that the
mean time between failures is less than 20 hours for field
robots [1], after which they require some sort of repair,
consuming time and resources. This implies that an
increment in reliability is needed, increasing the mean time
between failures or reducing the repairing time.
Two main methods have been used to increase the
reliability of robots. One method is to construct more robust
mobile robots, which can be done either by adding
mechanical or sensory redundancy to the robot. The other
method to increase the reliability is to add FDI systems,
which identify present problems and thus reduce the time
and effort needed for repairs [2]. As these systems are also
used in most fault tolerant control systems, as in [3], this
work is centred in the design and development of a FDI
architecture for cooperative mobile robots, which can obtain
information regarding the state of each robot for future
control decisions or repairs.
Several FDI methods have been developed for mobile
robots [4]. The core of the FDI algorithm used in this paper
is presented in [5], where only two possible faults are
analyzed: a reduction in the radius of one tire and a periodic
bump. In [6], the concept is further developed using a bank
of Kalman filters to determine faults on sensors and
actuators of a four wheeled robot. Another FDI method is

This work was supported by FONDECYT project no 1020741, “Fault
Detection and Identification in Nonlinear Time Variant Systems”.
R. A. Carrasco and A. Cipriano are with the Electrical Engineering
Department, Pontificia Universidad Católica de Chile, Santiago, Chile (e-
mail: {rax; aciprian}@ing.puc.cl).
presented in [7], where a bank of Kalman filters is combined
with a Markov model representation to identify the faults
through probability calculations. Other approaches to FDI in
mobile robots include the use of more advanced filtering
techniques to identify the faults, which incorporate the
nonlinear robot dynamics, as seen in [8].
Also aiming towards an improvement in the reliability,
several researchers have proposed a multiple robot
approach. In [9], the authors explain how the redundancy
present in cooperative mobile robots can be used to increase
the robustness of the group and thus improving the
efficiency. In the same line of study, the ALLIANCE
architecture presented in [10] shows a simple FDI system for
cooperative robots based on behavioural programming.
Regretfully, with exception of [11], which describes a
simple FDI method for cooperative manipulators, none of
works mentioned above contain an analysis on the FDI
limitations, nor of the efficiency of the isolation (number of
false positives or wrongly isolated faults), so no reliability
comparisons can be made between them.
This paper presents the basis for a FDI framework for
mobile robots, defining an ordered structure over which
future FDI methodologies can be developed and
implemented. Taking advantage of the different benefits that
single and multiple robots’ FDI mechanisms show, this work
shows a layered architecture for FDI on cooperative robots,
where the different layers can be implemented on the robot
system depending on the capabilities and resources present.
The idea behind this architecture is to combine methods
behind single and cooperative robot FDI systems to achieve
an architecture capable of detecting a wide range of faults.
The multiple layer approach allows to take advantage of
the different information, control, and redundancy levels that
exist within the control structure, designing each layer of the
FDI architecture according to the level of information
available at its corresponding level in the control structure.
This permits an efficient use of the available information.
Multiple layers have been used by other authors to
achieve FDI in different classes of robotic systems, adapting
the FDI system depending on the redundancy that exists,
[12]-[13], but the idea of having a cooperative layer within
the architecture has not been implemented yet.
Fig. 1 illustrates the framework used, describing the
control structure for each robot Ri, according to the level of
control, and the interaction with the different layers of the
FDI system.
The control structure has five main layers. First, the
Layered Architecture for Fault Detection and Isolation in
Cooperative Mobile Robots
Rodrigo A. Carrasco and Aldo Cipriano
T
Proceedings of the European Control Conference 2007
Kos, Greece, July 2-5, 2007
WeA11.4
ISBN: 978-960-89028-5-5 2950

physical layer consists of the body, sensors, and actuators
needed. The Actuator Control Layer controls the robot’s
hardware in order to follow a determined trajectory. Next,
the Navigation Control is dedicated to design the trajectories
needed to achieve the different objectives. The highest
control layer in a single robot is the Control of Objectives
Layer, which designates the tasks that must be done and
where the robot must go in order to do them. Finally, in
cooperative robots another layer is added: the Multirobot
Coordination Layer. This layer can be either centralized or
distributed among the robots, and is the one that designates
the objectives of each robot, to achieve their common goal.
This work is divided into five sections. First, section 2
presents a description of the robot system over which the
FDI architecture is implemented. Next, section 3 describes
the first layer of the architecture, presenting the method used
and an analysis of the fault detection capabilities and
limitations. Section 4 continues with the description of the
second layer of the architecture, indicating how the
cooperative robots approach is used. Section 5 then
describes the interaction between both layers. Finally,
section 6 shows the experimental results of each layer.
II. SYSTEM DESCRIPTION
The FDI architecture’s design is based on simulations, for
an analysis of its capabilities, being then implemented on a
group of small mobile robots, for validation.
The simulations were done in Matlab, using a the same
mathematical model described in [14], which has the
kinematic and dynamic equations for each robot. The
sensors readings are also simulated in the model by adding
noise to the measurements, using a 10% of the maximum
value of the measurement as standard deviation.
A group of homogeneous small mobile robots,
constructed at our university, is used to test each layer of the
architecture and validate the data obtained through
simulations. The group is composed by three robots, as the
one showed in Fig. 2. Each robot moves using two
independent actuated wheels, enabling differential steering.
They have two low cost microcontrollers for programming
and control purposes, and are equipped with optical
encoders on both wheels to achieve relative localization. The
robots are also equipped with a digital compass, to measure
the heading angle. Each robot has a frontal sonar and a low
resolution CMOS camera, for navigation purposes. For this
work, the camera is only used to recognize other robots,
which is done by identifying the red marker that each one
has on top.
Faults can be divided into two groups: those that can be
continuously monitored on a single robot, and those that can
be detected through cooperation between them. Although
some faults can actually be detected through both methods,
they are grouped were it is easier to detect them.
For the first layer of the FDI architecture, seven different
faults on sensors and actuators are considered: 1-2: slippage
of one of the wheels, 3-4: one of the wheels gets stuck, 5:
both wheels get stuck, and 6-7: one of the encoders gets
stuck (i.e.: the velocity of that wheel is read as zero). The
cooperative layer isolates faults on sensors that are
redundant in the robot team. This layer is designed to detect
four different faults: 1: additive fault on the sonar, 2: the
sonar gives a constant value, 3: additive fault on the
compass, 4: the compass gives a constant value.
III. CONTINUOUS FDI LAYER
A. Method Description
The use of multiple models has shown to be a good tool
for continuous monitoring of faults in mobile robots. As all
the faults this layer must detect can be modelled within a
Kalman filter, a bank of eight Kalman filters is used: one for
modelling normal operation (M0), and seven for modelling
the faults (M1-M7). The basic structure of each model Mi is
as follows:

1, , ,
1, 1, ,
M ; 0..7
k i i k i i k k i
i
k i i k i k i
X A X B U w
i
Z C X v
+
+ +
= + +⎧⎪ =⎨ = +⎪⎩
(1)

In (1), Xk,i is the state vector for the robot at time k using
model i, whereas Zk,i is the measurement. The matrices Ai,
Bi, and Ci are the state equation matrices for model i, and Uk
is the control input at time k. The process and measurements

Fig. 1. Control and FDI Structures for robot Ri.

Fig. 2. Mobile Robot used with marker.
WeA11.4
2951