IEEE Robotics & Automation Magazine - September 2011 - 111

*
Table 3. Memory usage of the routing tables (kb).
Group

Routing
tables

FPB Bloom
filters

10.52 10.52 10.52 10.52

EEB Bloom
filters

3
48

Percentage

100

9.449 9.452 9.483 9.462

21.9

(a)

(b)

(c)

(d)

F (N)

Figure 7. (a) Initial position of the snake. (b) Obstacles
appearing before the snake. Snake passing the obstacle area
(c) before and (d) after.

r (N ⋅m)

Integrated Experiments
A practical WiME system has been implemented. Altogether, 30 wireless visual sensors are mounted on the ceiling in a building with a topology shown in Figure 2.

Software packages and communication protocols were
developed for wheelchair navigation using the proposed
algorithms. The only processor onboard the wheelchair is
an 8-bit processor Atmega 128L, which is used to link the
WiME network wirelessly through the IEEE 802.15.4
protocol and to drive the two differential wheels via two
pulsewidth modulation (PWM) signals. With such a lowperformance processor onboard, the high-level functions,
such as localization and path planning in the tiered architecture [1], cannot be implemented. However, the distributed environment intelligence developed in this article can
control such a low intelligence wheelchair with superior
mobility (see the video MovieWiME.wmv). A communication layer was developed on top of the IEEE 802.15.4
short-frame protocol for the coordination of wireless sensors and wheelchairs. There are five sets of commands
defined, as shown in Figure 9:

v (m/s)

is 0.8 m/s. The robot with a mass of 0.56 kg has its maximum
driving force Fmax ¼ 4:4 N, smax ¼ 2:0 Nm, and friction
coefficient lmax ¼ 0:6. The parameters are selected empirically. In general, we adjust the maximum speed, force, and
torque to produce fast and smooth tracking.
Figure 7 shows four real-time images captured by the
four eyes during the experiment. The four eyes arranged in
a square have IDs of 30, 40, 50, and 60 from the top-right
anticlockwise. In Figure 7(a), the robot is controlled by Eye
30 (top right) and is headed toward the control area of Eye
60 (bottom right). The light gray circles represent a
dynamic snake track for the robot to follow. The green
circles indicate the l window used for predictive control.
The white rectangular blobs are the dynamic obstacles. As
one can see in Figure 7(a), dynamic obstacles exist in the
view of eyes 30, 40, and 50 but not in Eye 60. After 0.5 s,
shown in Figure 7(b), a triple obstacle is detected by Eye 60.
Although the robot is controlled by Eye 30, which cannot
see the new obstacle, the snake is updated to an obstacle-free
path through internal data exchange with Eye 60. Robot
control privilege is handed over from eyes 30 to 60 in Figure
7(c), where the robot is passing the triple obstacle area. Figure
7(d) shows the robot after it has successfully passed the triple
obstacle returning to its original snake track.
The corresponding control signals and robot velocity
are shown in Figure 8. It shows that predictive control
keeps the driving force and steering torque within the
admissible range. The alternating signs of the steering torque indicate the feedback regulation of the predictive control required to follow the snake track, even though the
predictive control is designed in an open-loop format for
each single prediction in the l window. The robot speed is
about 0.76 m/s in Figure 7(a). Because there is no obstacle
in the view of Eye 60, the path generated is a straight line.
When the obstacle is detected by Eye 60, the snake track
changes as in Figure 7(b). The robot maintains its high
speed when crossing the overlap area between eyes 30 and
60. When the robot approaches the triple obstacle after
2.3 s, it decelerates to avoid skidding [Figure7(c)]. It accelerates again after passing the triple obstacle in Figure 7(d).

5
4
3
2
1
0
-1
-2
-3
-4
-5

0.5

1.5

19.7

2
2.5
Time (s)

3.5

Figure 8. A robot's control and velocity.

SEPTEMBER 2011

IEEE ROBOTICS & AUTOMATION MAGAZINE

111

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2011