LU et al.: FTMCS FOR ODOR SOURCE LOCALIZATION 5421
TABLE I
P
ARAMETERS OF THE P3-DX MOBILE ROBOTS
Fig. 2. Motion control architecture.
III. COORDINATING CONTROL AND TRACKING CONTROL
In this section, we will propose a motion control architecture
and describe each module in this architecture.
A. Motion Control Architecture
The proposed motion control architecture, which consists of
five modules, i.e., a sensing module, a soft sensor module, a
coordinating control module, a tracking control module, and an
actuation module, is illustrated in Fig. 2.
1) Module Description: For the sensing module, the input
data are from two aspects: 1) from the robot; and 2) from
the environment. Moreover, the sensing module enables trans-
ferring the real velocity, the real position of the robot, and
environmental information as the output data into the tracking
control module and the soft sensor module. For the coordinating
control module, the input data are the position of the odor
source, the virtual positions of other robots, and the virtual
velocities of other robots. Moreover, the virtual velocity and
virtual position of the current robot as the output data of the
module are transferred into the tracking control module. For the
tracking control module, the input data are the virtual velocity,
the virtual position, the real position, and the r eal velocity,
whereas the output data are the applied torque for the left
wheel and that for the right wheel of robots. In the proposed
motion control architecture, the soft sensor can be used to
“measure” the position of the odor source. In order to run the
soft sensor, the input data, which are from the sensing module
and communication networks, are the positions of robots, the
odor concentration, and the wind velocity. In addition, the
output data are the positions of the odor source [14].
2) Specification of Mobile Robots: In order to enable the
motion control architecture to run at real mobile robots, the
TABLE II
S
PECIFICATIONS OF THE MOBILE ROBOT
specifications for mobile robots need to be described. Specif-
ically, four kinds of information, including wind velocities,
odor concentrations, robot positions, and robot velocities, need
to be detected. Correspondingly, the required detection r eso-
lutions are given in Table II. For the wind sensor, in order
to satisfy the resolutions, we can use WindSonic produced by
Gill Instruments Ltd. as the wind sensor. For the concentration
sensor, we can use e2v MiCS-5524 as the concentration sensor
to detect carbon monoxide, hydrocarbons, and volatile organic
compounds. The Differential Global Positioning System and an
encoder can be used to get the position and velocity with the
corresponding resolutions. For the actuator, the Pittman brush
direct current motors can be used. Moreover, the soft sensor,
the coordinating control algorithm, and the tracking control
algorithm can run at the hardware environment with Pentium M
1.4 GHz with 512 MB of random access memory. The operation
system is Linux. In order to satisfy the aforementioned require-
ments, the mainboard PCM-9584 can be employed.
B. Potential Function
The potential functions have been widely used f or robot
navigation and control. Various functions have been proposed
for different aims [4], [19], [23]. In this paper, we consider the
following potential function form as
J = f (x
i
− x
j
2
)
where f : R
2
× R
2
→ R denotes a map, x
i
− x
j
2
refers to
the distance between the ith robot and the jth robot, and ·
2
denotes the two-norm. Moreover, this kind of potential function
should have the following properties.
1) Potential function J has the zero minimum at x
i
−
x
j
2
= d, where d>0 is a predefined constant.
2) There exists a constant c>0 such that the potential
function satisfies
J ≤ c
N
i=1
∂J
∂x
i
2
2
.
It is obvious that we can always build a potential function
that possesses the aforementioned properties. For example,
the potential function proposed by Olfati-Saber [21] has the
aforementioned properties.
C. Coordinating Control
In the coordinating control module, we provide the following
finite-time coordinating control algorithm, which can enable the