IEEE Robotics & Automation Magazine - March 2012 - 54

CO2 Concentration (% by Volume)

0.40
Passive
Semiactive
Active

0.35
0.30
0.25
0.20
0.15
0.10
0.05
0

50
Time (s)

100

Figure 2. Comparison of the gas transport approaches:
measured CO2 concentration at stable environmental conditions
with microdrone in flight.

position was chosen for all experiments approximately 1 m
downwind from the fan in the height of the gas flow. A second Dr€ager X-am 5600 gas detector was used as a reference
system to provide reference measurements. In these experiments, we used a CO2 infrared sensor [22].
One experiment for each gas transport approach was performed. At the beginning of each trial, the gas concentration
was adjusted so that the reference sensor measured a stable
value of 0:5% by volume. Afterward, the microdrone equipped
with the gas-sensitive payload and one of the three gas transport mechanisms was flown to the measuring point to perform
CO2 measurements in flight for approximately 100 s.
Results
The results of the experiments are shown in Figure 2. Clear
differences can be seen between the different gas transport
approaches. In contrast to measurements in a large volume
of the same gas concentration, none of the approaches is
capable to measure the reference gas concentration of 0:5%
by volume. The highest measured concentrations (peaks) lay
around 0:32 (passive), 0:30 (semiactive), and 0:39 (active) %
by volume, which is 64, 60, and 78% of the reference
measurements. The averaged measurement results after the
sensor responded were 0:18 Æ 0:02 (passive), 0:26 Æ 0:01

Figure 3. A microdrone flying below a visualized plume in a
wind tunnel at a flow speed of 2 msÀ1 . (Photo courtesy of BAM.)

IEEE ROBOTICS & AUTOMATION MAGAZINE

MARCH 2012

(semiactive), and 0:33 Æ 0:02 (active) % by volume, which is
36, 52, and 66% of the reference measurements.
The reason for the generally lower measurements compared with the reference sensor is the rotor movement of
the microdrone. The active gas transport method can avoid
this dilution effect best with the long carbon fiber tube.
However, the disadvantages of this method are the additional weight (approximately 76 g), the position of the tube
inlet, which strongly dictates the measurement results as
well as the enlarged drone size that is exposed to wind. A
tradeoff between the applicability and sensitivity is given by
the semiactive payload (approximately 26 g). Using the
semiactive gas transport, it turns out that flying rather below
the plume is advantageous. In Figure 3, it can be seen that
the plume in front of the microdrone is still intact, while the
rotors redirect the plume completely downward.
The semiactive gas transport approach was used in the
real-world experiments as it offers high applicability and
reasonable sensitivity.
Estimation of the Wind Vector
The local wind vector is important for many existing gas dispersion models [23] to characterize the dispersion properties
of the plume as well as for gas source localization, plume tracking [24], and GDM (the "Statistical Gas Distribution Modeling" section). Wind measurements are furthermore
important since high wind speeds and strong wind gusts in the
target area may also limit the use of the microdrone presented
in the "Gas-Sensitive Microdrone: Robotic Platform" section,
which can only resist wind speeds of up to 8 msÀ1 .
The response of many gas sensors is caused by direct
interaction with the chemical compound and thus represents
only a small area around the sensor surface. Additionally, a
single gas sensor does not provide directional information.
To mitigate these limitations, directional information in the
form of wind vector is useful. Consequentially, the onboard
measurement of the wind vector in real time is crucial.
In the following section, a new approach introduced by Bartholmai and Neumann [21] is described and validated, which
estimates the wind vector based on the existing measurement
data of the microdrone's onboard sensors (IMU). This
approach makes additional anemometric sensors superfluous
and is, to the best of the authors' knowledge, unique.
Theory
The wind vector estimation presented in this section is
based on the wind triangle (Figure 4). The wind triangle is
commonly used in navigation and describes the relationship between the flight vector~
v, ground vector ~
w, and wind
vector ~
u. Here, we can consider the 2-D case since the
knowledge of the gravity vector is available. Two of the
three vectors or four of the six parameters of the wind triangle (flight speed j~
vj, ground speed j~
wj, wind speed j~
uj,
drift angle a, and the angles b and c) are needed to derive
the remaining parameters. However, only the ground vector is directly given by the GPS receiver of the microdrone.

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2012