Sensors Hal

First results on a sensor bio-inspired by electric fish
Noël Servagent, Brahim Jawad, Stéphane Bouvier, Frédéric Boyer, Alexis

Girin, Francesco Gomez, Vincent Lebastard, Pol-Bernard Gossiaux
To cite this version:

Noël Servagent, Brahim Jawad, Stéphane Bouvier, Frédéric Boyer, Alexis Girin, et al.. First
results on a sensor bio-inspired by electric fish. 2012.
HAL Id: hal-00695453

https://hal.archives-ouvertes.fr/hal-00695453
Submitted on 9 May 2012
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
First results on a sensor bio-inspired by electric fish
Noël Servagent1 , Brahim Jawad1,2 , Stéphane Bouvier1 , Frédéric Boyer2 ,
Alexis Girin2 , Francesco Gomez1 , Vincent Lebastard2 , and Pol-Bernard Gossiaux1 .
1
SUBATECH UMR CNRS 6457,
Laboratoire de Physique Subatomique et des Technologies Associées
University of Nantes - IN2P3/CNRS - Ecole des Mines de Nantes
4 rue Alfred Kastler, 44072 Nantes Cedex 03, France
2
Ecole des Mines de Nantes, IRCCYN UMR CNRS 6597
1, rue de la Noë, BP 92101, 44321 NANTES Cedex 03, France
8 mai 2012
Résumé
This article presents the first results of a work which aims at designing an active sensor
inspired by the electric fish. Its interest is its potential for robotics underwater navigation
and exploration tasks in conditions where vision and sonar would meet difficulty. It could
also be used as a complementary omnidirectional, short range sense to vision and sonar.
Combined with a well defined engine geometry, this sensor can be modeled analytically. In
this article, we focus on a particular measurement mode where one electrode of the sensor
acts as a current emitter and the others as current receivers. In spite of the high sensitivity
required by electric sense, the first results show that we can obtain a detection range of the
order of the sensor length, which suggests that this sensor principle could be used in future
for robotics obstacle avoidance.
1 Introduction
Underwater robot navigation is essentially based on the use of vision and sonar. While
the range of vision decreases with the rarefaction of available light energy, sonar escapes this
constraint, since, as any active sense, the energy supporting the measured information (here
acoustic) is produced by the sensor itself. Exploited by swimming animals practicing echolo-
cation, this advantage makes sonar a solution ideally suited to underwater robot navigation
in dark open spaces. On the other hand, in highly confined spaces, such as narrow pipes and
tunnels, caves and shallow waters, the reverberation of multiple echoes from the obstacles can
make the interpretation of acoustic measurements difficult. Furthermore, when waters contain a
lot of floating particles, diffraction can jam the sonar signals. Thus, in the navigation of complex
environments immersed in turbid waters, an alternative sense would be of great assistance to
vision and sonar. In fact, nature has already invented such a sense, the electric sense, which is
used for navigation and communication by several families of fish. These electric fish are prin-
cipally nocturnal and live in the particle rich waters of the equatorial forests [1]. These fish are
able to navigate in the complete darkness of their natural habitat using self generated electrical
fields. They emit electric signals into the environment, which in turn are perceived using an
array of electroreceptor organs in their skin. This ability is named electrolocation since it is
based on the same principle as echolocation with the electric field replacing the acoustic waves.
One of the best studied electric fish in terms of object perception is the Mormyrid Gnathonemus
1
ANGELS(231845)
petersii (see Fig. 1). The active electrolocation of this fish, has a range of one body length [2].
It is based on the emission of a dipole-shaped electric field (see Figure 1) named the carrier, by
polarization of the body relative to an emitter-organ in the tail named the Electric Organ of
Discharge (or EOD) [3]. The high internal conductivity of the fish’s body relative to the surroun-
ding environment focuses the emitted field lines [4], obliging them to flow through its specialized
electro-sensitive skin. Discrete electro-receptors distributed all over the epidermis capture an
instantaneous 3-dimensional electric image of the near environment [5], comparing the trans-
epidermal electric measurements that would be expected in the absence of any obstacles with
the sensory image that is actually measured. As a result, electric fish are able to detect, localize
and recognize the shape of objects in their vicinity [6, 7]. Moreover, analyzing the influence of
the environment on the amplitude and phase modulation of the carrier, the fish can perceive the
electric properties of the materials, as their conductivity, and capacity, which appear as ”electric
colors” [8, 9]. In short, electric fish can perceive any close object electrically contrasted with
respect to water. This short range omnidirectional sense is well adapted to navigating complex
environments encumbered by many obstacles such as the roots of the trees of flooded tropical
forests which are the natural habitat of electric fish. Hence, electric sense could have a narrow
but relevant niche in under-water robotics. Furthermore, it could be used as a complementary
short range sense for low level reflex navigation, or for some specific task related to the electric
properties of the environment, such as seeking a conductive object among insulating obstacles,
or avoiding a hot submarine spring (conductivity being dependent of temperature).
Figure 1 – The electric field is distorted by the presence of an object (for instance, an insulating
cylinder pushes the field lines away whereas a conductive cube funnels them).
Though potentially interesting, this mode of sensory perception, has been hitherto little
studied by engineers [10]. It is only recently that Mc. Iver et Al. [11, 12] have begun to implement
the electric sense in robotics. They were the first to construct a detector and its supporting
algorithms using only electric sense. Though successful, the proposed approach could not be
directly applied to a realistic 3D underwater vehicle since the sensor was only a set of (four)
point electrodes with no volume. The measurement mode proposed in [11, 12] can be named
U-U mode, the first letter standing for the type of imposed electrical quantity (voltage (U) or
currents (I)) and the second for the type of measured electrical quantity (U or I), since a voltage
2
ANGELS(231845)
is applied on a first pair of electrodes and measured on the second. The European project Angels
[13] is currently developing sensors inspired by electric fish with the objective of implementing
the resulting device on a realistic 3D underwater vehicle. The Angels solution offers different
working modes differentiated by imposing or measuring current and/or voltage, and can shape
the electric field around a vehicle of non negligible volume, using a real-time reconfiguration
capability. The purpose of this article is to describe this sensor operating in one particular basic
mode, its U-I mode, where a voltage is imposed and the resulting currents are measured. In this
preliminary feasibility study, we restrict our investigations to the case of resistive phenomena in
laboratory conditions. This allows us to assess the intrinsic capabilities of this new sensor, while
postponing the results of its implementation on a real underwater robot to a future article. As
we will see, the recovering of the fish range requires a sensor accuracy of 1/1000. Based on basic
electronics, the proposed solution achieves a first step towards the implementation of electric
sense in robotics.
The article is structured as follows : the second section is devoted to the U-I mode principle,
its exploitation for electrolocation and the modeling aspect of the sensor. In the third section we
detail the principle of the electrolocation measurement. In the fourth section we show how we
adapted our apparatus to the electric properties of tap water. In the fifth section the electronics
are fully detailed in relation to the desired task. We then illustrate the test bed in the sixth
section. In the seventh section we focus on the experimental results. Special attention is paid
to measurement accuracy and its sensitivity to temperature and conductivity. We also present
some comparisons between the model and the experiment for specific scenes. Future projects
that will use our electrolocation sensor are outlined in section eight. Section nine concludes.
2 The physics of the sensor

2.1 Principle of the measurement
Taking inspiration from the African fishes of the Mormyride family, we designed a device
capable of producing a dipolar basal field. In this case, a set of electrodes (named receivers)
are put under a controlled voltage with respect to a single ground emitter mimicking the EOD.
This creates a current loop around the sensor. Such a current loop polarizes the objects in the
surrounding scene which in turn produces perturbations of the measured currents flowing toward
the receivers.
2.2 The Modeling of the Electric Perception by a Finite Volume Sensor

Based on the principle stated in the previous subsection, our sensor is a set of groups of
metallic electrodes placed on the insulating boundaries of a cylindrical body (Fig. 2). On each
group of electrodes, we can impose a potential with respect to a common grounding group playing
the role of emitter, while measuring the currents flowing across the individual electrodes.
Our test probes are all based on a common geometry : a cylinder of diameter 20 mm with
spherical extremities (see Fig. 2). This slender shape was chosen since it offers the possibility to
approximate the electric measurements by simple expansions with respect to the sensor aspect
ratio r/L (r is the radius of the cylindrical shape and L is its length). The insulating parts are
made of PVC and the electrodes of 316 stainless steel. Several versions have been built with
various numbers of groups of electrodes and various segmentations of the electrodes.
As the sensors are slender, we can consider the current emitter and receivers (of ring or
hemispherical shape) as an array of effective spherical - electrodes arrayed along a line joining
their centers. Such a model is named ”poly-spherical model” in [14]. Furthermore, let us assume
that these spherical electrodes are sufficiently distant (with respect to their size) to prevent any
3
ANGELS(231845)
Figure 2 – Electrolocation test probes. Left : Dipolar sensor. Middle : 16-electrode ; each group
is composed of four electrodes. Right : Dipolar sensor illustrating mechanical components and
the conductivity meter on top of the sensor.
mutual perturbation between them. The vector of voltages U between the emitting electrode
and the grounding electrodes (i.e. the current receivers) is linked to the vector of currents I
measured on the same electrodes by Ohm’s law [14] :
I = CU = (C0 + δC) U (1)
where C and C0 are respectively the conductance matrix of the external scene with and without
object and δC is the perturbation of conductance induced by the presence of an object within the
reach of the sensor. Furthermore, we have C0 = γS0 and δC = γδS with γ the water conductivity,
S0 a matrix which depends on the sensor geometry only, and δS a matrix depending both on the
geometry of the sensor and objects boundaries. This poly-spherical model has been applied to
the prediction of the perturbations of large objects such as the insulating walls of a tank in [14],
and to small homogeneous objects in [15]. The response of such an object in a quasi-uniform
electric field E0 is chiefly described by the induced dipolar momentum p [16] whose value is
p = P · E0 , (2)
where P is the polarizability tensor that encodes the geometry of the object as well as its
intrinsic electric properties. Because the model consists of electrodes of spherical geometry it is
straightforward to find the associated conductance perturbation δC in the presence of objects
of polarizability P :
δC = C0 P T KPC0 , (3)
where P is a current
P conservation operator which ensures that the sensor is in electrokinetic
equilibrium, i.e. Ii = 0. K is a tensor encoding the electric and geometric properties of the
object with respect to the sensor :
1 ri · (P · rj )
Ki,j = − , (4)
4πγ | ri |3 | rj |3
where ri is the distance vector from the center of the ith (spherical) electrode to the center of
charge of the object. Note that the electronics hereafter will propose an in-line measurement
of γ which will allow the separation (4) of the influence of water conductivity from that of
the boundaries geometry. In particular, S0 can be evaluated once for all through preliminary
measurements of I without obstacles, or calculated with a numerical code capable of modeling
arbitrary geometries as the Boundary Elements Method [17, 18, 19].
4
ANGELS(231845)
Figure 3 – a) The sensor, here composed of two electrodes ; the arrows represent current lines
in the external conductive medium and the open circle represents the loading resistance. b) The
electromagnetic model. c) The equivalent electric circuit.
The prediction of lateral currents (currents measured in specific sectors of the previously
considered electrodes) is out of reach of the poly-spherical model described above and requires
a finer modeling, presented in [15] for a cylindrical geometry of the sensor.
3 The Principle of Electrolocation Measurement

3.1 From the Physical Model to the Equivalent Electrical Circuit
To understand the principle of measurement by electrolocation let us start by defining an
equivalent electric model of our sensor. For the sake of simplicity, let us focus on a simple
slender sensor composed of only two electrodes. This sensor can be considered as two equivalent
spherical electrodes of opposite charge ±Q with a different electrical potential and separated by
a distance L sufficiently greater than the radius of the electrodes (see section 2.2 and Fig. 3b).
The sensor and its surroundings are equivalent to a simple electric circuit (see Fig. 3c) with an
sinusoidale generator U in series with 3 resistances : the internal resistance Rg of the generator,
the resistance Rext offered by the external scene to the sensor and the internal resistance Rr
of the receiving electronics. Because the internal resistance of the generator and also of the
receiving electronics are negligible in comparison with the measured Rext for the U-I mode, the
circuit can be simplified by removing the resistances Rg and Rt from the circuit in Fig. 3c. Then
the principle of measurement by electrolocation is simply to find an accurate value for Rext .
From the theoretical view point, Rext is the inverse of the conductance matrix C = C0 + δC
with δC given by (1) in the case of a single object, and which reduces to a single scalar for a
sensor with two electrodes.
3.2 Consequences of Current Flowing in Water Medium

One important aspect is the effect of the electrical current on the water. Because tap water
is an ionic substance composed of different kinds of elements, chemical reactions can occur when
a direct electric current (DC) is imposed [20]. This led us to choose an alternating sinusoidal
signal with no DC component. As a result, the principle of electrolocation measurement is the
measurement of a signal amplitude modulated by the environment (see Fig. 4).
5
ANGELS(231845)
Figure 4 – Principle of the measurement : a sinusoidal signal U0 is imposed to avoid any

electrolysis. As a result the electrolocation sensitivity is a modulation of a current or voltage
(here a current I) on the electrodes
3.3 Accuracy Requirements

To estimate the accuracy required for detection of large obstacles, let us consider a dipolar
sensor like the one depicted in Fig. 2a and 3a submerged in a tank with no object. We can
deduce (section II of [14]) an expression of the relative resistance perturbation for a dipolar
sensor perpendicularly facing a wall made of material whose conductivity is much lower than
the conductivity of water (referred hereafter as an ”insulating wall”) :
δRext (r/L)
= , (5)
Rext e e + 2d)
4d(1 + d)(1 e
where Rext is the resistance with no object, δRext is the additive resistance induced by the wall,
r is the radius of the electrodes, L is the length of the sensor and de = d/L with d the distance
between the front electrode and the wall. This expression (5) is valid provided the polarization
on each of the electrodes under the influence of the wall can be neglected. This is indeed the case
when the closest distance between the wall and the sensor (d) is much larger than r. Defining
an electrolocation distance range of d = L with an electrode radius of r = 1 cm and a length
L = 20 cm, one obtains from eq. (5) that
δRext
≃ 0.2% , (6)
Rext
which corresponds to the minimal sensitivity mandatory to detect the wall at such a distance d.
To achieve this accuracy we adopted in this preliminary feasibility study the following methods :
– Use of a band pass filter to eliminate high and low frequency noise.
– Simultaneous measurement of the amplitude of the imposed voltage signal and the measu-
red currents to take into account amplitude variations of the sinusoidal generator device.
– A statistical measurement of 1000 values to estimate a mean value and standard deviation
reflecting the quality of the measurement (see section 7.1).
In a further step, the synchronization of emission and reception can be avoided, for instance by
using a RMS to DC converter on each of these signals.
6
ANGELS(231845)
4 Adaptation to the Electrical Characteristics of the Probe in

Fresh Water
The electrical conductivity γ of a material, defined as the inverse of the resistivity ρ, is
obtained experimentally by measuring the resistance R of a volume of section A and length l of
this material and then using the following formula :
l
γ= . (7)
RA
4.1 Temperature Response of water

Besides the fact that the water conductivity is dependent on the concentration of ions, it is
also dependent on temperature and frequency [21, 22]. For our purpose, we will resort to the
following simplified formula as for the temperature dependence :
γT 1 ∂γ
γ T′ = with β = ≃ 1 to 3% per ◦ C,
1 + β (T − T ′) γ ∂T
where γ T′ (resp. γ T ) is the conductivity at temperature T ′ (resp. T ) and β is the tempera-

ture compensation slope of the material. Conductivity measurements carried out on the water
supply in our laboratory showed a dispersion at constant temperature (25◦ C) that ranged from
360 µS/cm to 410 µS/cm. This dispersion is due to changes in chemical composition. The measu-
rements of conductivity as a function of temperature are presented in Fig. 5. They were obtained
using a conductivity meter WTW Cond 197i. The curve of the temperature compensation slope
shows discontinuities related to the low resolution of our conductivity meter (±1 µS/cm). Ho-
wever the average value appears clearly to be about β = 2.5% per ◦ C, which experimentally
confirms the dependence of the conductivity on the temperature of the water. Note that in
order to offer a measurement of conductivity independent of temperature, conduc-
tivity meter manufacturers offer to extrapolate the measurement to a reference
temperature 1 . The temperature compensation of the conductivity is either linear
or non-linear under the standard NE 27888. For our work, we have switched this
feature off.
4.2 Frequency Response of Water

The electrical signals emitted by fish for electrolocation are of two distinct types : sinusoidal
signals (in all Gymnotiforms except Gymnotus carapo and Electrophorus electricus) and pulse
signals (in the Mormyridae). Fishes belonging to the same species do not generate both types
of signals. In order to simplify the specifications of the signal processing electronics (amplifier,
current / voltage, filter, etc.), our preliminary study of the electrical sense has been focused
on the sinusoidal voltage type. To understand the electrical behavior of our laboratory’s water
supply coupled to the sensor, we measured the complex impedance Z of a dipolar hemispherical
probe electrode immersed in several tanks of volume equal to 25 l (small tank), 150 l (medium)
and 1000 l which corresponds to the (large) tank where experiments are carried out (see Fig.
4). All these tanks (small, medium and large) are filled with tap water of conductivity γ equal
to 0.0432 S/m, 0.0406 S/m and 0.0432 S/m respectively. In all cases, the results presented in
Fig. 6 can be qualitatively understood as the combination of three dominant features : a) the
presence of a shallow minimum for the capacitance of water for frequencies between 5 × 103 Hz
and 5 × 104 Hz, well established in literature [23], b) an increase in the contact impedance
1. In the case of WTW Cond 197i conductivity meter, two reference temperatures 20 ◦ C or 25 ◦ C are available
7
ANGELS(231845)
Figure 5 – Top : water conductivity versus temperature ; bottom : temperature compensation

slope versus temperature
650
600
550
500
450 2 3 4 5
10 10 10 10
20
large, γ = 0.0432 S/m
medium, γ = 0.0406 S/m
10 small, γ = 0.0432 S/m
−10 2 3 4 5
10 10 10 10
Figure 6 – Complex impedance of our dipolar probe immersed in a 220-liter tank of water.
between the electrodes and the water at low frequencies (ν . 103 Hz), in particular its capacitive
component [24], and c) the impedance of the coaxial cable, that manifests itself mostly at high
frequencies (ν & 105 Hz). Finally, in the band where the capacitance gets its minimum value, Z
can be approximated by a pure resistive component. This feature being observed on any of the
tanks, it is independent (probably beyond a certain minimum value) of the size of the tank. Thus,
to be in agreement with the resistive model of the article (section II.B), the frequency chosen for
our study in sinusoidal operating mode is in the intermediate band 5 × 103 Hz . ν . 5 × 104 Hz
(see Fig. 6) where the immersed probe can be considered to be purely resistive.
8
ANGELS(231845)
5 The Electronics
5.1 Basic principle
In a general way, we can define an electrolocation sensor as a set of N electrodes attached
to an insulating body. To open the field of study as widely as possible, Angels is developing a
sensor on which one can either, for each electrode, impose the voltage and measure the current or
impose the current (possibly 0 in passive measurement mode) and measure the voltage. However,
this article focuses on the particular case where voltages are set and current measured. Our first
electronic prototype based on this principle has been achieved. A simplified configuration of the
overall configuration corresponds to a sensor in which all electrodes measure currents (Ik ) and
are set to the same voltage (virtual ground), except one distinguished electrode which is set to
an imposed voltage U1 (see Fig. 7). This electrode is located in one tip of the prob like the EOD
of the fish is in its tail.
Figure 7 – Electrolocation sensors in U − I mode. Here the electrode 1 is the current emitter
5.2 Signal Generator

The final aim of our research project is to design on-board sensors for autonomous under-
water vehicles (see section 8). However in this preliminary feasibility study, we used an external
sinusoidal signal generator. In this remote configuration, a laboratory Low Frequency Generator
is coupled to an operational amplifier mounted in follower mode. The frequency of the sine wave
is f = 22.5 kHz. For the embedded generator, an integrated circuit function generator EXAR
XR2206 has been chosen. A prototype circuit has been built and tested (see Fig. 8). The sinusoi-
dal signal generation is validated and the circuit is implemented on the Angels multi-electrode
sensor.
Figure 8 – Prototype circuit of embedded generator and test result.
9
ANGELS(231845)
5.3 Receiving Electronics

The receiving electronics are illustrated in Fig. 9. One analogue channel consists of a current-
voltage amplifier associated with a second order band pass filter. The measuring electrodes are
connected to the current input of the preamplifier which is a virtual ground (see section V.A). We
name Rext,1−k the external resistance of the scene between the electrode imposing voltage 1 and
the measuring electrode nk. In the case of the external signal generator, the cables transmitting
the electrical signal to the electrodes are relatively long (5 m) and we had to take their resistances
r1 and rk into account in the theoretical study (see Fig. 9). The resonant frequency of the filter
Figure 9 – Schematic of receiving electronic in U − I mode (1 channel).
device (Rauch) was chosen so as to equal the frequency of the sine wave. In these conditions, the
transfer function of the filter is equal to Tfilter = − 12 , and the dependence between the voltage
generator VIN and the filter output VOUT is given by :
VOUT GR1
T = = , (8)
VIN 2(Rext,1−k + r1 + rk )
where G is the output gain of the filter. We express the external resistance of the scene as a linear
function of the voltage ratio VIN /VOUT . These voltage amplitudes are acquired experimentally
by an analogue to digital converter triggered to perform the conversion at the maximum and
at the minimum of the sensed signal, assuming the environment is purely resistive. In these
conditions, one can measure the external resistances through :
VIN GR1
Rext,1−k = A + B, with A = ; B = −r1 − rk . (9)
VOUT 2
Due to the fact that the measuring electrode is at potential zero, the application of Ohm’s
law allows one to find that the current flowing toward the electrode k is given by :
VOUT
Ik = 2 . (10)
GR1
Fig. 10 shows an 8-channel 4-layer PCB implementation of the receiving electronics.
To achieve a measurement of good accuracy, the coefficients A and B are obtained by way of
calibration. This calibration consists in connecting a set of standard resistors of Rcal (tolerance :
±0.01%) with a low temperature coefficient (TCR : ±0.6ppm/◦ C) to the receiving electronic
input. For each of these standard resistors, a population of 10000 measurements is acquired for
10
ANGELS(231845)
Figure 10 – Receiving electronics in U − I mode (8 channels).
statistical study 2 . A linear regression using the least squares values is applied to the average
values of each population to determine the coefficients A and B. Fig. 11 shows a calibration curve
for one channel. The displayed residuals are less than 0.01% in magnitude and well scattered
around zero, indicating the absence of any significant non-linear component. When the tolerance
of calibration resistors is combined with the amplitude of the regression residuals, a minimal
accuracy of our measurement device of about ±0.02% is deduced. This accuracy is compatible
with the expected one of section 3.3.
Mesured resistance
0.1
error (%)
−0.1
300 400 500 600 700 800 900 1000 1100 1200
Reference resistance (Ω)
Mesured resistance
0.01
error (%)
0
−0.01
300 400 500 600 700 800 900 1000 1100 1200
Mesured resistance
1000
(Ω)
500
300 400 500 600 700 800 900 1000 1100 1200
Figure 11 – Calibration of one channel of the receiving electronics (A = 2457.8Ω, B =

−5.9169Ω). The error bars represent ± the standard deviation of a population of 10000 measu-
rements.
With several measuring electrodes, the same method can be applied in order to determine the
currents flowing toward each of the receivers, except that Rext,(1)−k is not the external resistance
of the scene but the inverse of C1,k , where C is the conductance matrix described in section 2.2,
and all (Rext,1−k ) + rk branches are connected in parallel to the emitting electrode.
6 The Experimental Bench

6.1 Tank and Robot
In order to test our electrolocation sensors in controlled and repeatable conditions, an au-
tomated test bench consisting of a one cubic meter tank and a three-axis Cartesian robot, has
been built (see Fig. 12). This gantry was fixed on the top of the tank and allowed the probe to be
2. with a resulting standard deviation of 0.02% – see section 7.1 as well – which indicates that the statistical
error performed on each average is ≈ √0.02%
10000
and thus negligible with respect to the intrinsic tolerance of the
calibration resistors
11
ANGELS(231845)
(a) 3D modeling (b) experimental set-up
Figure 12 – Electrolocation test bench.
positioned in translation along its X and Y axes with a precision of 1/10 mm. The orientation
in the horizontal plane was adjusted to absolute precision of 0.023 ◦ using an absolute rotation
stage (see Fig. 13). All probes tested were positioned in the tank using a rigid glass epoxy fiber
tube (⊘14 mm) whose height is adjustable. This isolating tube allows the passage of electri-
cal cables dedicated to the signals coming from the electrodes to readout electronics (analogue
chain + ADC board) without compromising the measurements. Its vertical orientation is ensu-
red using a micrometric adjustment base. Our test also allows dynamic trajectory management
at a significant range of speeds relative to the tank size. The maximum speed available was 300
mm/s (≃ 1 km/h) for both translations and 80 ◦ /s (13.5 rot/min) for rotation.
(a) CAD (b) experimental set-up
Figure 13 – Stage-keeping probe.
This test-bed allows one to separate the problem of underwater locomotion (of our future
robots) from that of the perception, and to study the intrinsic capabilities of the electric sense
before its implementation on a real autonomous robot.
6.2 Computer Monitoring of the Robot

We chose a three axes robot computer control using simulink software with the dSpace
system. The entire structure of the system is given in Fig. 14.
The electrolocation signals coming from the electrodes of the probe are first processed by
analogue chain (amplification and filtering), then converted by a 16-bit ADC, DS2004 card with
a resolution of 0.3 mV/bit to 0.15 mV/bit (range from ±5 V to ±10 V). This card allows us to
convert all 16 channels at the same time at a maximum speed of 1.25 MHz.
6.3 The Objects

To test the measurability of objects of simple geometry using electric sense, a set of test
objects were fabricated with conductive and insulating materials (see Figures 15 and 16). We
fabricated objects of similar density to water in order to be able to keep them under water
12
ANGELS(231845)
Figure 14 – Computer monitoring of the test bench.
(a) (b) (c)
Figure 15 – Test objects : a)insulator cube (side= 40 mm), b) conductive cube (side= 40 mm),
c) conductive sphere (⊘ = 40 mm)
easily when using weakly invasive bonds such as nylon filaments or non-conductive rods of small
diameter (1 mm) (see Fig. 16).
(a) mounting of a (b) ellipsoid set (c) mounted

conductive ellipsoid (conductive and conductive ellip-
insulating) soid
Figure 16 – Ellipsoid object.
7 Experimental Results
7.1 Characterization of Electronic Noise of the experimental test bed
A characterization of the total noise of the analogue electronic chain has been performed.
The Fig. 17 shows the experimental set-up.
Two types of external resistor Rext were connected to our electronic device in order to
characterize the level of noise in the conditions of our experiments. The first type are the standard
13
ANGELS(231845)
Figure 17 – Schematic of experimental set-up for measuring noise
resistors already used for the calibration process (see section V.C) whose resistance is known
accurately, while the second type is the probe itself, immersed in the tank filled with tap water
and placed in radial contact to the wall at the level of its center with the motors of the gantry
first switched on and then switched off.
On Fig. 18a we represent the distribution of N = 10000 measurements, that is to say N
successive single ADC conversions, each of them lasting 15 ms. One observes an average value
located at 0.04 Ω from the expected value Rcal = 500 Ω given by the manufacturer (i.e. relative
error 0.008%). This is compatible with an intrinsic accuracy for Rcal of 0.01%. Thus, 0.02%
defines the precision of the sensor, which is compatible with the requirements of section 3.3.
This noise could have several origins such as filtering defects or amplitude fluctuations in the
power supply, or electromagnetic compatibility issues between the various electronic components
or even between the electronic boards [25, 26]. The electronics components themselves could even
be responsible for this noise, as the chosen electronics are not of ”low noise” type.
For the second characterization, the cable length from the probe to the electronics was
L = 5 m. With motors switched off (see Fig. 18b), the distribution of the measurements shows
a Gaussian noise, which is fairly similar to that of the standard resistor. This shows that both
the extra cable and the water environment are low sources of noise compared with the elec-
tronics. When the motors of the gantry are switched on (see Fig. 18c), the distribution of the
measurements, highlights a non-Gaussian noise, which is explained by the fact that the motor
supply generates non-random noise. We note that some measurements seem very distant from
the average value (some 10 Ω). The width at half height of this distribution is 0.7 Ω, that is
1.9 times higher than that obtained for the standard resistor. From these characterizations we
conclude that the dominant sources of noise are the electronics and the motors of the gantry.
7.2 Drift Test in Temperature of Electronics

We have already presented an experimental measurement dispersion of ±0.02% (see sect.
5.3) at room temperature. As such a small value is required in order to detect objects of low
electrical disturbance, we have assessed its stability with respect to other physical parameters.
Temperature is a significant factor in measurement stability, therefore the receiving electronics
were tested at various temperatures. For this purpose, the circuit was placed in a stove with a
standard input resistor 500 Ω ± 0.6 ppm/◦ C located outside.
The outside temperature was 20 ◦ C ± 1◦ C throughout all the duration of the experiment,
implying a negligible drift of the resistor value (±0.3 mΩ). The measurements have been consi-
dered only after stabilization reflecting the uniformity of temperature in the stove. By restricting
the temperature maximal value to 55 ◦ C, the results of this test show a resistance deviation of
the order of ±0.005%, as illustrated in Fig. 19. As this value is well below the required accuracy
of ±0.02% we conclude that our electronics is robust enough against temperature fluctuations
and necessitates no temperature correction.
14
ANGELS(231845)
500 500
450 450
400 400
350 350
300 300
(Number)
(Number)
250
250
0.625 Ω
200 0.36 Ω 200
150 150
100 100
50 50
0 0
499 499.2 499.4 499.6 499.8 500 500.2 500.4 500.6 500.8 501 564.5 565 565.5 566 566.5 567 567.5 568
Mean=500.0404 std=0.17911 N=10000 (Ω) Mean=566.2581 std=0.31273 N=10000 (Ω)
(a) standard resistor Rcal = 500 Ω (motors (b) probe resistance in water with motors
off) off
2000
1800
1600
1400
1200
(Number)
1000
0.7 Ω
800
600
400
200
0
560 562 564 566 568 570 572
Mean=566.151 std=1.4371 N=10000 (Ω)
(c) probe resistance in water with motors

on
Figure 18 – Histogram : distribution of resistance measurements obtained with U −I electronics

for a population of N = 10000 individual measurements ; plain line : “best” gaussian fit, of same
average and standard deviation as the distribution.
7.3 Measurements with Test Probes

7.3.1 Wall detection
The first electrolocation results were obtained using the probe with two electrodes described
in section 2.2. This probe has a cylindrical geometry of 2r = 20 mm and an overall length of
L = 220 mm. It consists of two stainless steel hemispherical electrodes separated by an insulating
tube made of PVC. The gantry was programmed to move the probe from one wall of the tank
to the other in the X-axis while the probe remained centrally placed with respect to the Y - and
Z-axes throughout. The walls are made of glass and are considered as pure insulators as they
have very low conductivity relative to tap water. The dimensions of the water surrounding the
probe are 971.5 mm × 971.5 mm × 975 mm and the conductivity measured is 356 µS/cm (typical
value). The walls are nicely detected by a significant increase of the resistance, i.e. a significant
decrease of the measured current (see Fig. 20) as the probe approaches the wall.
For a probe starting in the center of the tank (X = 0), a natural way to set a limit of
detectability (or range) is to require that Rext (X) = (1 + 0.02%) Rext (X = 0), where 0.02%
corresponds to the precision of our device. This leads to X ≈ 110 mm, that is a distance between
the wall and the front electrode of d = 266 mm ≈ 1.2 times the probe length (d ≈ 1.2L).
This first study shows that the electrical sense is a moderate-range mode of perception for
large objects like walls. More precisely, the wall was detected at a distance of about one probe
15
ANGELS(231845)
Figure 19 – Temperature drift of the U-I receiving electronics ; the error bars represent the
standard deviation obtained with a population of 1000 measurements for each temperature.
Figure 20 – External resistance measurements for a wall to wall crossing of the sensor.
length, which is similar to the range of active electric sense of fish.
16
ANGELS(231845)
(a) conductive sphere (b) insulating sphere
Figure 21 – External resistance measurements for a wall to wall crossing with 40 mm spherical
object placed at 50 mm. a) Water conductivity γ = 356 µS/cm. b) Water conductivity γ =
359 µS/cm.
(a) conductive cube
Figure 22 – External resistance measurements for a wall to wall crossing with a cubic object
(a = 40 mm placed at 50 mm). Water conductivity γ = 359 µS/cm.
7.3.2 Object detection

To show the sensitivity of our electrolocation device to the presence of objects of different
conductive types and shapes, we present in Figures 21 and 22 the external resistance measure-
ments for a probe, an electronic device and a trajectory unchanged with respect to those of the
previous subsection (a tank with no objects).
Four objects were used. One was a conductive cube of side = 40 mm. Two were spheres
(⊘ = 40 mm), one insulating ; one conductive.
Passing next to the objects, the resistance measurement increases if the object is insulating
and decreases if the object is conductive (see Figures 21 and 22). Two peaks appear during the
movement, which correspond to the positions of the objects at the minimal distance from one of
the two electrodes. Analysis of the curves in Figures 21 and 22 shows that the conductive nature
of the objects is easily identifiable by observing the increase or the decrease of the external
17
ANGELS(231845)
resistance as the probe passes close to the objects. Despite the small distance between the probe
center and the object center (50 mm), their presence results in a visibility average which remains
low, ranging from 0.3% to 0.9%. The measurements taken under the same conditions as above
(Fig. 22a) but for an object distance of 100 mm from the probe reveals visibility even lower at
0.03% which is one order of magnitude lower than for a 50 mm distance (see Fig. 23). In addition
one can also observe the sensitivity of the method for the shape of the object. We recover the
fact that for a same object geometry, the influence of a conductive material is two times higher
than for an insulating material while for a same material, the measurement is proportional to
the object volume at the leading order [6]. As a consequence, in the case of the cube, since its
volume includes that of the sphere, the measurement peak of Fig. 22 is higher than that of the
sphere (Fig. 21).
Figure 23 – External resistance measurement for a wall to wall crossing with cubic object
a = 40 mm placed at 100 mm.
This study has demonstrated the ability of a two electrodes sensor to detect small size
objects relative to its length. Their conductive difference was easily highlighted. However, it is
confirmed that the electric sense is of short range for this case (between one-third and one half of
the total length). Furthermore, the response of each of the sensor’s electrodes is maximum when
the object faces it while it rapidly decreases when the object moves away along the sensor. A
natural consequence of this feature is that the two-electrode sensor loses contact with the object
when it is located between the electrodes. This induces a small blind-spot in the detection,
visible between the two peaks in Figs 21 and 22. This drawback can be overcome by adding
more electrodes to the design. On Fig. 24, we show the currents for the 4-electrode sensor of
Fig 2b passing by a conductive cubic object in the same conditions as in Fig. 22. As the sensor
moves forward, the conductive cube first faces the head receiver electrode (X ≃ −0.1 m), whose
external resistance is hence lowered ; next (X ≃ −0.03 m), the cube faces the body electrode
closest to the head, whose external resistance is in turn reduced. For X ≃ 0.1 m, the cube faces
the emitting tail electrode with an overall decrease of all equivalent resistances. For the body
electrode located closest to the tail, we do not observe a doubly-peaked structure but a single
minimum for X ≃ −0.07 m. In conclusion, one observes a good mapping between the object
position along the sensor and the ordering of the current peaks for the various electrodes which
allows the contact with the object to be maintained.
18
ANGELS(231845)
950
940
930
Equivalent water resistance (Ω)
−0.2 −0.1 0 0.1 0.2 0.3

Probe center X position
940
930
920
−0.2 −0.1 0 0.1 0.2 0.3
774
772
770
768
−0.2 −0.1 0 0.1 0.2 0.3
Figure 24 – External resistance measurements by a 4-electrodes sensor for a wall to wall cros-
sing with a conductive cubic object a = 40 mm placed at 50 mm. Top, middle and bottom
panel correspond to the resistance measured respectively by the “head” electrode, by the body
electrode located closest to the head, and by the body electrode located closest to the tail.
7.4 Influence of water Conductivity

The external resistance offered by the environment Rext is inversely proportional to the
conductivity γ. The changes in these two quantities are related as follows :
∆Rext ∆γ
=− . (11)
Rext γ
Conductivity variations related to the chemical composition of water can be controlled throu-
ghout the experiment by avoiding external contamination. The conductivity is also strongly de-
pendent on the temperature, so that the external resistance varies as a function of temperature
as follows :
1 ∂Rext 1 ∂γ
=− = −β , (12)
Rext ∂T γ ∂T
where β (=1 to 3%/◦ C) is the temperature compensation slope of the water and T is the tempe-
rature. We have shown the experimental influence of temperature on the external resistance by
placing the probe motionless in the middle of the 1 m3 aquarium and performing measurements
over time thus allowing for natural daily variations of temperature. Despite the large volume of
water, the relative deviation of resistance reaches 0.2% (∆Rext = 1 Ω) in 5000 s (see Fig. 25),
that means, for a temperature compensation slope of 2.5%/◦ C, a cooling of about 0.08 ◦ C.
Another experiment showed the influence of temperature on external resistance measure-
ment. The probe crossed the tank and returned at low speed (see Fig. 26). In the middle of
19
ANGELS(231845)
Figure 25 – External resistance versus time due to natural convection cooling of the 1 m3 tank.
the tank, the curve below shows that the relative deviation of resistance between the outward
and return travels is 0.06% (∆Rext = −0.3 Ω). For a temperature compensation slope of about
2.5%/◦ C, we deduce an increase of the water temperature of about 0.025 ◦ C. Note that such a
resistance deviation of 0.06% cannot be attributed to the temperature drift of the electronics
since it stood at 0.005% (which is an order of magnitude lower) for a larger range [22 − 55 ◦ C].
To express the impact of this low temperature increase in terms of positioning accuracy, let
us consider the positions Xret for which the resistance on the return crossing admits the same
value as the one for Xfor = 0 (corresponding to the aquarium center) on the forward travel. The
absolute difference |Xret − Xfor | quantifies the lack of repeatability due to temperature increase
and is found to be of the order of 150 mm (see Fig. 26), which is unacceptable.
Figure 26 – Influence of temperature during a forward and return crossing of the probe.
7.5 Correcting for water Conductivity Variation

The problem of the influence of variations in water conductivity on the electric measurements
can be solved by directly measuring γ. In this perspective, an independent conductivity sensor
will be implemented on the various prototypes of electrolocation (probes and robots). It will
be similar to the electrolocation sensor, but much smaller and hidden in a dead zone of the
sensor (robot) boundaries in order to be indifferent to the variations of the external environment
geometry. Assuming a uniform conductivity of the water around the sensor, the resistance Rcond
provided by this independent conductivity meter can be used to normalize the external resistance
offered by the scene Rext = (γS)−1 (with S = S0 + δS defined in section II.A), according to the
20
ANGELS(231845)
formula :
Rcond,0 1
Rext,N = Rext = S −1 , (13)
Rcond γ0
where Rcond,0 and γ0 correspond respectively to known reference values for the resistance and
conductivity of water. To test the feasibility of this conductivity correction, we have built a
bi-electrode probe with a conductivity measuring module (see Fig. 27). This module attached
to the fiber-glass tube of the probe consists of two cylindrical electrodes containing a space that
is continuous with exterior through two holes of ⊘ = 4 mm. These openings were oriented in
the direction of the probe movement to facilitate the renewal of water and to obtain realistic
conductivity measurements.
(a) probe structure (b) achievement
Figure 27 – Bi-electrode probe with conductivity correction cell.
The experimental results were achieved by modifying the water conductivity (varying the
temperature). The curves in Fig. 28 show uncorrected external resistance measurements and
corrected values (normalized) for the ambient conditions of temperature and pressure (1 atm,
20 ◦ C). To demonstrate the sensitivity of the corrected conductivity to the external environment,
an object was placed in contact beside the motionless probe on the temperature range 5 −
24 ◦ C. The corrected measurements appear to be independent of the temperature or conductivity
variation. The presence of an object is detected, which clearly demonstrates that the correction
prevents the probe from losing its electrolocation ability.
Figure 28 – Correction of water conductivity (see text for details).
21
ANGELS(231845)
Equivalent water resistance (Ω)

4.3
4.25
4.2
0.6%
4.15
4.1
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
Probe center X position (x971.5mm)
Figure 29 – External resistance measurements for a wall to wall crossing with 40 mm spherical
object placed at 50 mm in sea water (conductivity γ = 52.4mS/cm).
7.6 Influence of the salinity of the water

Regarding future application to underwater robotics, the influence of the salinity of on the
measurement has also to be studied. In this perspective, we have performed first experimental
tests in order to assess the influence of the salinity of water on the sensor performance. First,
let us remark that for a given geometry of the scene, and in a range of frequencies where the
immersed probe has a pure resistive behavior (see section IV.B), γ does not influence the topology
of current lines around the sensor, since it only appears as a prefactor γ −1 in the expression of
Rext (see (4)), whatever kind of contrasted objects are present in the system. For this reason,
patterns observed in Figs. 20 to 24 are preserved, even if the magnitude of Rext diminishes
with greater salinity. This, is illustrated on the Figure 29 obtained in the same conditions as
those of Figure 21.a but with a water 150 hundred times more salty (i.e. water conductivity
is shifted from 356µS/cm to 52.4mS/cm, which corresponds to the conductivity of sea water).
On this plot, we recover the same response as that of Figure 21.a, but shifted of two orders
of magnitude. This has been obtained by shifting the generator frequency f from 25.5 to 17.5
kHz and by re-calibrating A and B in (9) with a set of standard resistors with low resistances
(ranking from Rcal ≃ 2Ω to 7Ω, see Figure 30). Since now the tolerance on the calibrating
resistances is about ±0.05% the plot of Figure 30.b shows that the accuracy of the sensor in
sea water is about ±0.1% which has to be compared to the ±0.02% in fresh water. However,
this reduction of accuracy still ensures a detection range about the sensor length. This slight
deterioration of the sensor performance is due to the influence of noise which increases with the
decrease of Rext . This is illustrated on Figure 31, which displays the ex-situ noise distribution
(on Fig. 31.a with a calibration resistor of Rcal ≃ 4.04Ω), the same distribution but in sea water
with gantry’s motors switched off (Figure 31.b), and with the motors switched on (Figure 31.c).
The increase of noise is essentially due to the decrease of the voltage generator VIN in (9). If
one want to maintain exactly the same range as in the fresh water, a direct way would be to
redesign the electronics with a generator capable of generating a high current for a reasonable
voltage. Thus, on a robot we could have two electronics, one for fresh water, the other for sea
water, with the ability to switch from ones to the others. Another solution consists in changing
the geometry of the sensor in order to increase its external resistance (the So matrix of section
II.A) in the absence of any object. This is the case of the design of the robot presented in section
8, where the size of electrodes is dramatically reduced with respect to the area of the insulating
22
ANGELS(231845)
boundaries.
Mesured resistance
0.5
error (%)
0
−0.5
2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
Mesured resistance
error (%)
0.05
0
−0.05
2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
resistance (Ω)
Mesured
6
4
2
2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
Figure 30 – Calibration of one channel of the receiving electronics.
7.7 Comparison with the Model

Using the analytical model in section II, which is, unlike the BEM, usable in real-time,
we then compared the predictions of the model with the experimental data in the case of a
simple scene. The two-electrode probe made a wall to wall movement, passing within 50 mm
of an insulating sphere of 40 mm diameter. In Fig. 32, we have presented experimental results
(circles) as well as theoretical predictions (plain and dashed curves) for γ = 359 µS/cm±1 µS/cm,
accordingly to the precision available on the conductivity measurements.
The model turns out to be in rather good agreement with the data, although one observes
a systematic deviation of the order of 1.5% which exceeds the error made on conductivity mea-
surements. The remaining (small) discrepancy could be due to effects not taken into account in
the physical modeling, such as the contact impedance between the electrodes and the conduc-
tive medium [27]. Very good agreement with the experimental data is obtained by adopting an
effective value of the conductivity γ of 354 µS/cm. This result suggests that such a sensor can be
used for real-time underwater electrolocation based on simple analytical models as in a Kalman
filtering based approach [28, 29].
8 Future Robot Project

Beyond these first results obtained on the electrolocation testbed, we are actually building
a first reconfigurable underwater robot equipped with this electrolocation sensor. This robot
is planned in the context of the European project Angels [13] whose goal is the design of an
eel-like robot composed of nine connected rigid modules equipped with electric sense. Thanks
to its ability to change its morphology, this robot would be well adapted to exploration missions
or object recognition in confined spaces. When all the modules are joined together, the robot
will swim like an eel to minimize energy consumption. When detached, each module will benefit
from the same electronics as our slender probes. An image of one module of the robot is depicted
in Fig. 33. It consists of an insulating shell on which are fixed small hemispherical electrodes. As
regards locomotion, it is achieved by two axial propellers fixed to the top and the bottom of the
module while a third (lateral) propeller is located inside a transverse hole crossing the module
from side to side.
23
ANGELS(231845)
To relate the sensor range to the dynamic performance of the modules, one considers the
simple scenario where a module goes forward perpendicularly to a wall with a maximal thrust :
ρ
T+ = Cd Avc2 , (14)
2
and tries to avoid collision by inverting the thrust. In (14), ρ is the volume mass of water, Cd
the dimensionless axial drag coefficient of the module, A is the area offered by the module to
the flow, while vc is the maximal cruising forward velocity. In the case of the robot of Fig. 33,
the shape is quasi-ellipsoidal with a length L = 0.25m, a height of 0.11m and a width of 0.055m,
while the propellers are such that T + = 0.3N . Thus (14) gives vc ≃ 0.3m/s with a Reynolds
number Re ≃ 5.104 . Then, introducing (14), into the axial dynamic balance of one module,
allows one to evaluate the minimal counterthrust T − required to reach the wall with a null
velocity, while starting from a maximal velocity vmax :
2
λvmax
T− = , (15)
e(2λd/m) − 1
where d denotes the sensor range, m (= 1.3kg), is the virtual (solid + added) axial mass of the
module while λ = (1/2)ρACd (= 3kg/m). Thus, d (= L = 0.3m) and T − (= T + = 0.3N, the
propellers are symmetric) being imposed by the intrinsic capabilities of the actuators and the
sensor, (15) allows evaluating vmax . In the case of the project’s robot, we have vmax ≃ 0.5m/s
(1.8km/h) which is higher than the (maximal) cruising speed given by (14). In fact, the minimal
range of detection beyond which collision is avoided is d = L/2. As a result, though there are
other delays, as those introduced by the propellers dynamics, the range of the sensor should not
limit too much the intrinsic dynamic performance of the actuators.
9 Conclusion
We have presented the first results obtained on a practicable electrolocation system devoted
to robotics. With electrodes arrayed on a realistic slender body, the sensor surpasses the idea-
lized point electrode device of [12]. Measurement accuracy was challenging as object detection
requires measurement of a very low relative perturbation (of the order of 1/1000), but this was
achieved in our work on the electronics. As a consequence, our first experimental results based
on current measurement showed a good sensitivity to the presence of objects. Finally, the pro-
posed electronics allow reproducing the range of the fish. Thus, in the perspective of underwater
robotics, being of one length of the body (in fact the maximum length of a dipole attached to
the body), the expected range is one meter for a one meter length vehicle, ten meters for a
ten meters length vehicle and so on (according to (5)). We have dealt with the problem of the
perturbation caused by variations in the fluid conductivity by introducing an embarked conduc-
tivity meter for real-time and local measurement of this quantity and will transpose this solution
to our future robot. However, the sensitivity of the sensor to other sources of conductivity va-
riations (than electric colors of materials) can also be a useful complement to other sensors,
in order to detect gradients of other nature, as chemical or thermic ones. As regards control
aspects, several control techniques have been experimented on the sensor to demonstrate its
exploitability for robotics applications and will be presented in future articles [28, 29, 30]. Some
of them use models while others are model-free approaches known as reactive approaches [31].
These reactive approaches are well adapted to obstacle avoidance or object following depending
on their electric conductivity with respect to the ambient medium. Finally, several extensions
of the sensor are currently in progress. First, other measurement modes as the active or passive
voltage measurements named I − U or 0 − U modes (“0” meaning that there is no current
24
ANGELS(231845)
emission from the sensor), or the passive current measurement mode 0 − I, are now being ex-
plored, with the ability to combine them and even to switch between them. Second, we are also
developing electronics allowing to switch between emitters and receivers (a feature referred to
as “reconfigurability”) in order to shape the basal electric field in real-time. To date, the U − I
sensor here presented, is in course of implementation on the modules of the Angels project.
First experiments show that thanks to a reactive controller, an autonomous module can follow
a conductive target. As regards submarine robotics, the article also addresses the problem of
electrolocation in salty water. In this context in spite of the decrease of the external resistance
of two orders of magnitude, the electronics proposed in the article still works with a new cali-
bration. Thus, for an underwater vehicle, we could combine two circuits, one designed for salty
and the other for fresh waters with the possibility to switch from one to the other. Finally, while
the article considers the case of a slender geometry, other designs, with more bulky shapes and
smaller electrodes tend to compensate the high conductivity of salty waters. Lastly, in spite of
these encouraging results, many things remain to do before equipping an operational underwater
vehicle with this new sensor. In this perspective, the exploitation of the individual advantages
of the different measurement modes (U − I, I − I, 0 − I, U − U ...), seems a promising way to
improve the performance of electric perception as a whole. For instance, I − I mode could offer
the advantage to make electrolocation independent of the water conductivity, while U − U mode
could be better to increase the number of electro-receptors.
Acknowledgment
The ANGELS project is funded by the European Commission, Information Society and Me-
dia, Future and Emerging Technologies (FET) contract number : 231845. The authors gratefully
acknowledge the reviewers’ comments, and thank Douglas Carnall for his careful proofreading
of this manuscript.
Références
[1] H. Lissmann and K. Machin, “The mechanism of object location in gymnarchus niloticus
and similar fish,” The Journal of Experimental Biology, no. 35, pp. 451–486, 1958.
[2] K. Shieh, W. Wilson, M. Winslow, D. McBride, and C. Hopkins, “Short-range orientation
in electric fish - an experimental study of passive electrolocation.” Journal of Experimental
Biology, vol. 199, pp. 2383–2393, 1996.
[3] B.Rasnow and J. M. Bower, “The electric organ discharges of the gymnotiform fishes,”
Journal of Comparative Physiology A, vol. 178, no. 3, pp. 383–396, 1996.
[4] D. Babineau, A. Longtin, and J. Lewis, “Modeling the electric field of weakly electric fish,”
The Journal of Experimental biology, vol. 209, pp. 3636–3651, 2006.
[5] M. MacIver, N. Sharabash, and M. Nelson, “Prey-capture behavior in gymnotid electric
fish : motion analysis and effects of water conductivity,” The Journal of Experimental Bio-
logy, vol. 204, pp. 543–557, 2001.
[6] B.Rasnow, “The effects of simple objects on the electric field of apteronotus,” Journal of
Comparative Physiology A, vol. 178, no. 3, pp. 397–411, 1996.
[7] A. Caputi, R. Budelli, and C. Bell, “The electric image in weakly electric fish : physical
images of resistive objects in gnathonemus petersii,” Journal of Experimental Biology, vol.
201, no. 14, pp. 2115–2128, 1998.
[8] G. V. D. Emde, S. Schwarz, L. Gomez, R. Budelli, and K. Grant, “Electric fish measure
distance in the dark,” Letters to Nature, Nature, vol. 395, pp. 890–894, 1998.
25
ANGELS(231845)
[9] G. Emde, “Active electrolocation of objects in weakly electric fish,” The Journal of Expe-
rimental Biology, vol. 202, pp. 1205–1215, 1999.
[10] M. MacIver and J. Solberg, “Towards a biorobotic electrosensory system,” Autonomous
robots, vol. 11, pp. 263–266, 2001.
[11] J. Solberg, K. Lynch, and M. MacIver, “Active electrolocation for underwater target loca-
lization,” The International Journal of Robotics Research, vol. 27, pp. 529–548, 2008.
[12] ——, “Robotic electrolocation : Active underwater target localization,” International
Conference on Robotics and Automation, pp. 4879–4886, 2007.
[13] The ANGELS’ project, http ://www.theangelsproject.eu.
[14] B. Jawad, P. Gossiaux, F. Boyer, V. Lebastard, F.Gomez, N.Servagent, S.Bouvier, A.Girin,
and M.Porez, “Sensor model for the navigation of underwater vehicles by the electric sense,”
in Robotics and Biomimetics ROBIO IEEE, 2010, pp. 879–884.
[15] F. Boyer, P. Gossiaux, B. Jawad, V. Lebastard, and M.Porez., “Model for a sensor inspired
by electric fish,” IEEE transactions on robotics, vol. 28, no. 2, pp. 492–505, 2012.
[16] J. Jackson, Classical Electrodynamics. John Wiley & Sons, New York, 1962.
[17] L. Gaul, M. Kögl, and M. Wagner, Boundary element methods for engineers and scientists.
Springer-Verlag, 2003.
[18] Y.Liu, Fast multipole boundary element method. Cambridge university press, 2009.
[19] M. Porez, V. Lebastard, A. J. Ijspeert, and F. Boyer, “Multy-physics model of an electric
fish-like robot : Numerical aspects anbd application to obstacle avoidance,” in IEEE/RSJ
Int. Conf. on Intelligent Robots and Systems, 2011, pp. 1901–1906.
[20] J.-P. Diard and B. L. G. C. Montella, Cinétique électrochimique. Hermann (Paris), 1996.
[21] L. A. Klein and C. T. Swift, “An improved model for the dielectric constant of sea water
at microwave frequencies,” IEEE Trans. Ant. Prop., vol. AP-25, pp. 104–111, 1977.
[22] W. Ellison, A. Balana, G. Delbos, K. Lamkaouchi, L. Eymard, C. Guillou, and C. Prigent,
“New permittivity measurements of seawater,” Radio Science, vol. 33, pp. 639–648, 1998.
[23] D. P. Fernandez, Y. Mulev, A. Goodwin, and J. L. Sengers, “A database for the static
dielectric constant of water and steam,” J. Phys. Chem. Ref. Data, vol. 25, no. 1, pp.
1291–1306, 1995.
[24] T. Vilhunen, J. P. Kaipio, P. J. Vauhkonenand, T. Savolainen, and M. Vauhkonen, “Simul-
taneous reconstruction of electrode contact impedances and internal electrical properties :
I. theory,” Meas. Sci. Technol., vol. 13, pp. 1848–1854, 2002.
[25] A. Charoy, Parasites et perturbations des électroniques. Dunod, 1996.
[26] G. Vasilescu, Bruits et signaux parasites. Dunod, 1999.
[27] E. Somersalo, M. Cheney, and D. Isaacson, “Existence and uniqueness for electrode models
for electric current computed tomography,” SIAM J. Appl. Math., vol. 52, pp. 1023–1040,
1992.
[28] V. Lebastard, C. Chevallereau, A. Amrouche, B. Jawad, A. Girin, F. Boyer, and P. Gos-
siaux, “Underwater robot navigation around a sphere using electrolocation sense and kalman
filter,” in IROS 2010 IEEE, 2010, pp. 4225–4230.
[29] V. Lebastard, C. Chevallereau, A. Girin, N. Servagent, P. Gossiaux, and F. Boyer., “Recons-
truction and navigation with electric sense based on kalman filtering,” The International
Journal of Robotics Research., 2011 (submitted).
26
ANGELS(231845)
[30] F. Boyer, V. Lebastard, C. Chevallereau, and N. Servagent, “Underwater reflex naviga-

tion in confined environment based on electric sense,” IEEE transactions on robotics, 2012
(submitted).
[31] R. Brooks, “Intelligence without representation,” Artificial Intelligence, vol. 47, pp. 139–
159, 1991.
27
ANGELS(231845)
250
200
150
(Number)
0.02 Ω
100
50
0
4 4.01 4.02 4.03 4.04 4.05 4.06 4.07 4.08 4.09 4.1
Mean=4.0548 std=0.010498 N=10000 (Ω)
(a) standard resistor Rcal = 4 Ω
500
400
300
(Number)
0.024 Ω
200
100
0
4.1 4.12 4.14 4.16 4.18 4.2 4.22
Mean=4.1614 std=0.012176 N=10000 (Ω)
(b) Probe resistance in sea water with motors off
1200
1000
800
(Number)
600 0.055 Ω
400
200
0
4 4.05 4.1 4.15 4.2 4.25 4.3 4.35
Mean=4.1655 std=0.033619 N=10000 (Ω)
(c) Probe resistance in sea water with motors on
Figure 31 – Histogram : distribution of resistance measurements obtained with U −I electronics

for a population of N = 10000 individual measurements in sea water ; plain line : “best” gaussian
fit, of same average and standard deviation as the distribution.
28
ANGELS(231845)
630
625
620
615
Resistance (Ohms)
610
605
Experiment
600
595
Model
590
585
Uncertainly of
the conductivity
580
−30 −20 −10 0 10 20 30
position of the sensor (cm)
Figure 32 – Comparison between the the experimental data (circles) and the model ; plain line
corresponds to the model evaluated with the measured conductivity value of γ = 359 µS/cm,
while dashed lines correspond to γ = 358 µS/cm and γ = 360 µS/cm, chosen to illustrate the
effect of the conductivity meter accuracy (1 µS/cm) on the model predictions.
(a) overview (b) View with electronic

clutter (green).
Figure 33 – CAD view of the ANGELS-v1 module, designed by the CRIM Lab of Scuola
Superiore Sant’Anna. The attachment device on top will be removed in the final version and
allows it to be studied on the gantry of the tank.
29

Sensors Hal

Uploaded by

Copyright:

Available Formats

Sensors Hal

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sensors Hal

Uploaded by

Copyright:

Available Formats

First results on a sensor bio-inspired by electric fish

Noël Servagent, Brahim Jawad, Stéphane Bouvier, Frédéric Boyer, Alexis

To cite this version:

HAL Id: hal-00695453

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

2 The physics of the sensor

2.2 The Modeling of the Electric Perception by a Finite Volume Sensor

I = CU = (C0 + δC) U (1)

3 The Principle of Electrolocation Measurement

3.2 Consequences of Current Flowing in Water Medium

Figure 4 – Principle of the measurement : a sinusoidal signal U0 is imposed to avoid any

3.3 Accuracy Requirements

4 Adaptation to the Electrical Characteristics of the Probe in

4.1 Temperature Response of water

where γ T′ (resp. γ T ) is the conductivity at temperature T ′ (resp. T ) and β is the tempera-

4.2 Frequency Response of Water

Figure 5 – Top : water conductivity versus temperature ; bottom : temperature compensation

5.2 Signal Generator

Figure 8 – Prototype circuit of embedded generator and test result.

5.3 Receiving Electronics

Figure 9 – Schematic of receiving electronic in U − I mode (1 channel).

Figure 10 – Receiving electronics in U − I mode (8 channels).

Figure 11 – Calibration of one channel of the receiving electronics (A = 2457.8Ω, B =

6 The Experimental Bench

(a) 3D modeling (b) experimental set-up

Figure 12 – Electrolocation test bench.

(a) CAD (b) experimental set-up

Figure 13 – Stage-keeping probe.

6.2 Computer Monitoring of the Robot

6.3 The Objects

Figure 14 – Computer monitoring of the test bench.

(a) (b) (c)

(a) mounting of a (b) ellipsoid set (c) mounted

Figure 16 – Ellipsoid object.

Figure 17 – Schematic of experimental set-up for measuring noise

7.2 Drift Test in Temperature of Electronics

200 0.36 Ω 200

(c) probe resistance in water with motors

Figure 18 – Histogram : distribution of resistance measurements obtained with U −I electronics

7.3 Measurements with Test Probes

length, which is similar to the range of active electric sense of fish.

(a) conductive sphere (b) insulating sphere

(a) conductive cube

7.3.2 Object detection

−0.2 −0.1 0 0.1 0.2 0.3

7.4 Influence of water Conductivity

7.5 Correcting for water Conductivity Variation

(a) probe structure (b) achievement

Figure 27 – Bi-electrode probe with conductivity correction cell.

Figure 28 – Correction of water conductivity (see text for details).

Equivalent water resistance (Ω)

7.6 Influence of the salinity of the water

Figure 30 – Calibration of one channel of the receiving electronics.

7.7 Comparison with the Model

8 Future Robot Project

[30] F. Boyer, V. Lebastard, C. Chevallereau, and N. Servagent, “Underwater reflex naviga-

(a) standard resistor Rcal = 4 Ω

(b) Probe resistance in sea water with motors off

(c) Probe resistance in sea water with motors on