IEEE Robotics & Automation Magazine - December 2022 - 64

of the electric field intensity for the three architectures separately,
as shown in the right column of Figure 3. When such
architectures are used in a specific localization task for the
robot, their performance can be analyzed according to the
uniformity of the distribution of the electric field, as we have
discussed previously.
Specific to the localization tasks in the section " Localization
Experiments, " we talk a little more about the performance
of the three architectures. In the static scenario, the
localization may fail since the information obtained by the
electric receiver is partial and limited, which is very obvious
in Architecture 3. On the other hand, in the towing scenario,
the localization can be successful since the robot is continuously
acquiring new electric measurements. The experimental
results under Architecture 3 also reflect this. The towing
scenario can be equivalent to the displacement caused by the
external flow or the external force. This means that our biomimetic
electric sense-based localization method can be
applied to an unknown flow environment, which is very
meaningful for practical applications. Besides, the two types
of scenarios without self-motion reflect that Architecture 3
may be a poor architecture, while Architectures 1 and 2 perform
better. This phenomenon implies that the localization
results are better when adjacent emitters are orthogonal.
The previous discussion may provide a guide for us to construct
electric-emitter architectures for different working tasks.
Real-World Applications
Based on the previous evaluations and discussions, we now try
to conclude the appropriate real-world application environments
for our proposed electric sense-based scheme. First, the
hardware of our scheme can be customized for different sizes
of the robots/vehicles or for different scales of working spaces
since the detectable range of a single emitter can be easily and
flexibly changed by adjusting the physical parameters of the
emitter, such as the distance between its two electrodes.
Second, our localization scheme can be used in a largescale
2D (respectively, 3D) environment by deploying a 2D
(respectively, 3D) emitter architecture, especially considering
the good property that the architecture is allowed to be built
in a sparse manner. Specifically, we now discuss the issue of
emitter density. On one hand, according to our previous evaluation
of the maximum detectable range (2.32-2.76 m) of a
single emitter, the emitters can be placed at intervals of about
4.64-5.52 m to make sure the robots realize a continuous realtime
localization. On the other hand, if a lower emitter density
is still needed, the emitters can also be placed sparsely (e.g.,
at intervals larger than 5.52 m and even up to 10-20 m) since
the convergence of our localization method in such a sparse
situation has been verified by Experiment 4 (the kidnapped
robot problem). Once the robot is within the detectable range
of the emitters, the estimation can converge quickly, so the
robot doesn't need to always stay in the detectable range. We
want to claim that, although our evaluations of the maximum
detectable range and further the densities of the emitter array
are all in a 2D environment, it is similar to the situation in a
64 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
3D environment since the electric field generated by the emitter
is 3D. Note also that the emitter densities are evaluated
based on the maximum detectable range, which is determined
by the physical parameters of the hardware used in our
experiments, and the maximum detectable range can easily
be increased by adjusting these parameters and thus further
decreasing the density of the emitter array.
Third, by integrating other sensors (e.g., the pressure sensors
and the IMU sensors), which are also low in cost, the
working scope of our localization scheme could be further
improved at low cost. Concretely, one can equip the robot
with a pressure sensor (e.g., CPS131 from Consensic, Inc. or
MS5803-01BA from TE Connectivity Ltd., used frequently in
our previous robots [21]). The pressure sensor can obtain the
ground truth of the robot's depth by calibrating the current
atmospheric pressure, which may be more reliable than using
an extra camera to calculate the depth. Based on the pressure
sensor, the change in depth can be obtained at millimeter-level
accuracy. Besides, the robot's pitch and roll can be obtained
through the IMU. When the robot works in the coverage of
the 2D or 3D emitter architecture, the pressure sensor and the
IMU are integrated with the electric sense for perception and
localization. When the robot is out of the coverage of the
emitter architecture, the pressure sensor and the IMU keep
working. Once the robot is back within the architecture's coverage,
the electric sense works, and the estimation will be corrected
quickly.
In summary, although the experiments conducted in this
article are in a 2D setting, and the experimental pool is not
large, our proposed electric sense-based scheme is easily
extended for large-scale 3D working environments. We also
mention that, as far as the authors are aware, few studies have
considered the electric sense for free-swimming robots even
in two dimensions.
Conclusion and Future Work
In this article, for free-swimming small underwater robots in a
large-scale environment, we propose a novel electric sensebased
localization scheme, including a hardware solution and
three model-based perception methods. First, our hardware
solution is composed of an ac-based electric receiver furnished
on the robot and ac-based electric emitters fixed on the seabed.
The robot uses the measurements of its receiver to estimate its
relative position and orientation to the emitters on the seabed.
Second, we construct distributed electric-emitter architectures
to enable localization in a large-scale environment. Third, we
design three localization methods by selectively fusing the
dynamic model, IMU, and electric sense to explore the contributing
factors of the localization methods. Finally, to fully verify
our proposed electric sense-based scheme, we conduct four
types of localization experiments, including the very complicated
scenario of the kidnapped robot problem. Three typical electric-emitter
architectures are compared in these experiments,
and we find that architectures with orthogonal features are stable
in four localization scenarios. Our study provides a novel
solution to the localization of autonomous underwater vehicles
IEEE Robotics & Automation Magazine - December 2022

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