IEEE Robotics & Automation Magazine - June 2013 - 61

a corridor and a small room, a
kitchen, which present tiles on the
floor. As expected, the tiles induced
some bouncing on the robot body,
which consequently hindered the
mapping process. SAETTA was
guided by a joystick to perform less
regular trajectories than the ones
used in the IR-mapping test.
Figure 14 depicts the results
gathered in the various tests. As it
can be seen, each map effectively
represents the related plan.
Furthermore, several objects on the
floor have been represented: in particular, some boxes and furniture
were present during the tests.
These experiments show how
SAETTA can be used when exploiting sophisticated sensors. At the
same time, this kind of sensor can be
used in multirobot contexts: for
example, in [19], each robot exploits
a laser to extract features representing the teammates. Although different, the platforms used had an
almost identical processing power,
and, consequently, this kind of algorithms could be replicable on
SAETTA platforms.

INT

200 ms

Main Thread

Comm Thread

PIC Comm
6 ms

ZigBee
< 2 ms

Vision -IR
Collision
Avoidance
27 ms

RFID
Localization
Reader
Comm
< 3 ms

100 ms

Motion
Control
< 1 ms

100 ms
Reader
Comm
< 3 ms

ZigBee
< 2 ms

Gas Mapping
< 4 ms

100 ms

INT
PIC Comm
6 ms
t

Reader
Comm
< 3 ms
...

ZigBee
< 2 ms

...

Figure 15. Representation of the architecture used for gas source localization: the main
thread is composed, besides the standard communication with the PIC, by the collision
avoidance algorithm that uses both Webcam and IR data, a motion control based on a
behavioral scheme to impose speeds to actuators, and a gas mapping algorithm. In this case,
the execution of the vision algorithm being much shorter (200 ms) and strictly coupled with
the movement controller, it has been chosen as a subtask of the main thread. The other two
threads, one which manages the wireless ZigBee communication and the other in charge of
acquiring RFID data, run at 10 Hz and are decoupled from the others. The only contact points
with the main thread are the shared variables: for the former, the data gathered from other
robots or commands received by a PC, and, for the latter, the IDs of the RFID tags.

Gas Source Localization
In this section, the architecture used to localize a leakage of
gas in an indoor environment is described. To successfully
complete the task, the robot had to localize itself, avoiding
obstacles, and build a meaningful representation of the gas
dispersal. These experiments have been performed in a
smart home facility called the PEIS Home [20]. One of the
distinctive features of the PEIS Home is that its floor contains
a hexagonal grid of RFID tags, which can be individually
accessed by the robot while it navigates, and thus act as a
physical read/write memory [12]. This floor was used to
localize the robot by means of an RFID reader. During the
experiments, the robot avoided obstacles using the camera
and IR sensors. The former was used to execute the polly
algorithm [21] while the latter was in charge of recognizing
nearby obstacles suddenly appearing along the path. The
movement is determined by a simple behavioral algorithm:
SAETTA goes straight until an obstacle is detected; when this
happens, random self-rotations are executed until a freeway
is obtained. Gas concentration was detected by a 5521 MiCS
metal oxide semiconductor sensor from e2v. This sensor was
installed in place of the accelerometer and controlled by the
PIC with the standard sampling time (5 Hz). The used technique was the kernel distribution mapping plus variance
algorithm [22]: in its standard implementation, it takes about

(a)

(b)

Figure 16. Gas mapping experiment. (a) SAETTA equipment during
the navigation: within the triangle, the RFID reader is depicted:
dotted circle emphasizes the gas sensor while ellipses highlight
IR sensors. (b) Gas map gathered during the experiment: the final
error between the best hypothesis and the true location is less than
50 cm over an environment of 15 m2. (Photos courtesy of AASS
Centre, Orebro, and DIA, Rome.)

200 ms on SAETTA while using optimized code, as previously discussed, it takes, at most, 4 ms. The time reduction is
important for two reasons: it allows executing multiple tasks
june 2013

IEEE ROBOTICS & AUTOMATION MAGAZINE

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2013