IEEE Robotics & Automation Magazine - June 2013 - 58

INT

Main Thread
PIC Comm
SV

Motion
Control
Low Level Data
Processing

Comm
Thread
ZigBee
ZigBee

INT

Generic
Thread

...
...
...

scanner and builds maps related to
parts of real domestic environments.
Finally, a gas mapping task is presented: in this situation, SAETTA is
equipped with a large variety of sensors. Related to this experiment, a
description of the software architecture together with some technical
details is provided.

Consensus Algorithm
In this scenario, three robots were
used. Each agent, having a prior
Motion
ZigBee
knowledge of its own state with
Control
respect to a common frame, repeat...
Low Level Data
edly broadcasted its position to the
...
Processing
others. Applying the consensus fil...
ter, the agents would meet to the
t
mean of the initial positions if the
dynamics of the robots were linear.
Because of nonholonomic conFigure 9. High-level program structure: on the left, a vertical time line is depicted. The
first block, the main thread, is constituted by the periodic execution of subtasks as the
straints, the input generated by this
communication with the PIC (PIC Comm block), motion control, and low-level data processing.
control law was then filtered by
The period of this thread is given by the interrupt (INT), which rises every 200 ms. Other parallel
a Cartesian controller [14]. As
periodic threads can run at the same time: their execution can be shorter than the control cycle
(e.g., Comm Thread) or longer (Generic Thread) than that. The communication between threads
depicted in Figure 10, this controller
is performed by shared variable, SV.
overcomes the drawback: the trajectories of the robots converge to a
common point, which obviously is
not the mean of the initial states. In this test, each robot
140
traveled on an average for 78.8 cm: the maximum linear
120
speed was 7.54 cm/s while the rotational one was 0.98 rad/s.
100
Each robot broadcasted its state with a frequency of 10 Hz
(double with respect to the control cycle to prevent
80
data losses): on an average, each robot sent 174 packets. In
60
Table 2, it can be seen that the number of lost packets are
40
negligible as well as the computational load: the algorithm
20
takes about 3 ms per iteration, and this time can be further0
more reduced avoiding the use of trigonometric functions
−20
by lookup tables and an integer implementation. Also, the
−20 0
20 40 60 80 100 120 140
cm
odometry has proven to be very precise. Robots arrived to a
(a)
common point without having feedback about their positions by external sensors, e.g., cameras. As mentioned previously, several tests revealed errors smaller than 1% over long
straight paths (about 12 m) as well as small angular displacements over self-rotations.
PIC Comm

ZigBee

Table 2. Consensus experiment data.
(b)

Figure 10. (a) Robot trajectories during consensus test: black
triangles represent initial configurations. Axes units are expressed
in centimeters. (b) Stroboscopic view of the experiment. (Photo
courtesy of DIA, Rome.)

IEEE ROBOTICS & AUTOMATION MAGAZINE

june 2013

Robot 1

Robot 2

Robot 3

74.4527

79.4467

82.6943

Maximum speed (cm/s) 7.54

7.36

7.25

mean speed (cm/s)

3.15

3.85

3.99

Maximum data loss
respect to one target

Path (cm)

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