IEEE Robotics & Automation Magazine - June 2013 - 57

Software
Relying on a Linux OS, the software architecture has been
developed in a more structured manner than the PIC code. In
particular, it uses threads, to a great extent, to separate the different kinds of activities and to leave the researcher free to
choose the used synchronization mechanisms. All the threads
communicate by means of shared variables (see Figure 9).
The critical activities reside in the main thread that is periodically triggered by an interrupt deriving from the interface
with the PIC, and its time period is 200 ms. It contains the
routines to communicate with the PIC and compute the control law for robot motion.
Each additional device has its own thread. One is used
for the ZigBee communication module (or the wireless
fidelity one), another for the radio frequency identification
(RFID) reader, and so on. In general, each of these threads is
periodic with a period suitable for the specific application.
However, if needed, interprocess communication techniques [13] can be used to synchronize the various threads
with the main one. Note that each of them does not consume resources when it is idle. Obviously, the over-all computational load should not exceed the CPU capabilities with
respect to the time constraints regarding the main cycle.
Experimental Tests
In this section, some experiments are described to depict the
performances of the platform. In each test, the robot was

1.5

rad

1
0.5
0
−0.5

60
s

100

120

Figure 7. Heading acquired by the magnetometer: continuous
line (blue) reports the straight motion affected by iron beams, the
dotted (red) one is relative to a self-rotation of r/2 . Signals are
represented in radians units over a temporal scale (seconds).

asked to execute some well-known tasks in robotics. The first
experiment is devoted to characterize the interaction between
multiple units: robots have to meet at a common point using
the consensus algorithm [11]. In this scenario, both the communication and odometry performances were tested.
The second experiment is related to the IR sensors: a
robot explores a small area to build a map of the surrounding environment. In the third, a particle filter (PF)-based
localization is shown. Typically, this is a time-consuming
algorithm. However, it can also be used on small platforms,
thanks to an efficient implementation made possible by the
use of the compass and good robot odometry.
Although all the previous experiments are focused on testing transducer and odometry capabilities, the next two experiments are devoted to describing how SAETTA performs in
more difficult scenarios. In the fourth experiment, SAETTA is
equipped with a Webcam that is properly calibrated to extract
distance information about surrounding neighbors: in particular, a leader-follower experiment is described. In the fifth
experiment, SAETTA is equipped with a short-range laser

150
100

Degrees

High-Level Tier
The high-level tier is implemented on a Linux-embedded
board, the NETUS from ACME Systems srl. It has an
AT91SAM9G20 Atmel processor at 400 MHz and is equipped
with 64 MB of RAM. This board furnishes five serial ports and
two USB lines that allow SAETTA to be equipped with a wide
range of peripherals even at the same time (the only limitation
is related to the capability of batteries). The communication
with the external world is provided both by a standard wireless
device and ZigBee transceiver. The former is used when a large
bandwidth is required. The latter is preferred whenever the data
rate is within 125 kb/s as this kind of communication reduces
the power consumption. The ZigBee communication is implemented by an Xbee chip from Max Stream (www.sparkfun.
com/datasheets/Wireless/Zigbee/XBee-Manual.pdf ).
Nominally, this chip has 20 and 100 m range in indoor and outdoor environments, respectively. It provides several options that
are useful in robotic tasks: for example, it optionally handles the
acknowledgment of communication (such as Transmission
Control Protocol/Internet Protocol) or not [similar to user
datagram protocol (UDP)]. Furthermore, to save power, it can
broadcast the packets only to specific subsets of other nodes.
These options (that are completely managed by the module
without further computational load for the processor) find a
direct correspondence in multirobot applications, where the
communication graph has more [11] or less [12] importance,
requiring a strictly connected net or a sporadic exchange of
information without acknowledgments.

50
0
−50

−100
−150
40

80
s

100

120

Figure 8. Magnetometer measurements compared with bearing
achieved by odometry: continuous line (blue) is the odometry of the
robot, dotted line (red) is the magnetometer reading. The signals
are expressed in degree units over a temporal scale (seconds). The
maximum error is about 7°.

june 2013

IEEE ROBOTICS & AUTOMATION MAGAZINE

http://www.sparkfun

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