IEEE Robotics & Automation Magazine - December 2022 - 58

Furthermore, to evaluate the three typical emitter architectures,
considering the limiting size of the pool, we use the
simple forms of their configurations, as shown in Figure 3
(left column). Specifically, each of the three architectures contains
two emitters. The position and orientation of emitter 1
for all three architectures are (1.5 m, 0.5 m, 0˚). The position
and orientation of emitter 2 are (1.5 m, 0.5 m, 90˚), (1.5 m,
1.5 m, 90˚), and (1.5 m, 1.5 m, 0˚), for Architecture 1, Architecture
2, and Architecture 3, respectively.
Localization Experiments
Now we conduct the localization experiments for the small
underwater robot in four scenarios to fully demonstrate the
effectiveness of our proposed hardware solution and localization
methods. Specifically, in the first two scenarios, we separately
conduct localization experiments for a stationary robot
(Experiment 1) and a robot under towing (Experiment 2);
1.5
2
0.5
1
10
20
30
40
123456789
(b)
10
15
20
5
12 34 56 78 9
Labels of the Poses
(c)
Figure 7. Experiment 1: Localization of a stationary robot under the " electric sense +
IMU " method. (a) The robot's nine poses, where the arrow direction represents the
robot's orientation and the start of the arrow corresponds to the robot's position. (b)
The average position errors for the nine poses under each emitter architecture. For
Architecture 3, three of the average position errors (for poses 1, 3, and 5) are out of
range of the figure. (c) The average orientation errors for the nine poses under each
architecture. Archi.: Architecture.
58 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
X (m)
(a)
236
69.1
92.3
Architecture 1
Architecture 2
Architecture 3
that is the robot is set to be without self-motion (its control
output
u = ). Then, in the last two scenarios, we conduct
localization experiments for an autonomous swimming robot
0 ,
^hu ! including the position tracking problem (Experiment
3) and the kidnapped robot problem, which is much
more complicated (Experiment 4).
Experiment 1: Stationary Robot
In this scenario, we fix the pose of the robot and then test our
electric sense-based localization scheme. Since the dynamic
model of the robot does not work
^hu = the " electric
0 ,
sense + IMU + dynamic-model " method is equivalent to the
4
5
6
3
2
1
7
0.51 1.5 2
2.5
3
8
9
" electric sense + IMU " method, and the " IMU + dynamicmodel "
method is invalid. Therefore, in this scenario, we only
need to verify the localization ability of the " electric sense +
IMU " method. Note that this is a global localization problem
since the robot's initial position is unknown to the robot. To
this end, the initial positions of the
particles, which are used in the SIR
algorithm in the electric sense module,
are randomly distributed throughout
the experimental pool;
is, x ! [, ] m and y ! [, ] m.
that
() 03i
() 02i
Meanwhile, the angle of the IMU
equipped on the robot is calibrated
according to the coordinate system
before the experiment starts, thereby
giving a rough actual value as the initial
orientation of the robot. As shown
in Figure 7(a), the robot is fixed at
nine poses [x(m), y(m), ψ(°)] of [0.5,
0.4, 0], [0.5, 0.4, 45], [0.5, 0.4, 90], [2.6,
1.6, 180], [2.6,1.6, 225], [2.6, 1.6, 270],
[2.0, 0.5, 270], [2.0, 0.5, 135], and [2.0,
0.5, 0], respectively. The robot's actual
pose is recognized by the overhead
camera as the ground truth. The position
error is defined by the Euclidean
IEEE Robotics & Automation Magazine - December 2022

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2022
