sensors

Article
Design and Characterization of a Three-Axis Hall
Effect-Based Soft Skin Sensor
Tito Pradhono Tomo 1, *, Sophon Somlor 1 , Alexander Schmitz 1 , Lorenzo Jamone 2 ,
Weijie Huang 1 , Harris Kristanto 1 and Shigeki Sugano 1
1 Department of Modern Mechanical Engineering, School of Creative Science and Engineering,
Waseda University, 3-4-1 Okubo, Shinjuku-ku, Tokyo 169-8555, Japan;
sophon@sugano.mech.waseda.ac.jp (S.S.); schmitz@aoni.waseda.jp (A.S.);
kou498431413@fuji.waseda.jp (W.H.); harris.kristanto@fuji.waseda.jp (H.K.); sugano@waseda.jp (S.S.)
2 Instituto de Sistemas e Robótica, Instituto Superior Técnico, Lisbon 1049-001, Portugal;
ljamone@isr.tecnico.ulisboa.pt
* Correspondence: tito@toki.waseda.jp; Tel.: +81-3-5286-3264; Fax: +81-3-5272-0948

Academic Editors: Yajing Shen, Lianqing Liu, Ning Xi, Wen Jung Li and Xin Zhao
Received: 12 February 2016; Accepted: 31 March 2016; Published: 7 April 2016

Abstract: This paper presents an easy means to produce a 3-axis Hall effect–based skin sensor for
robotic applications. It uses an off-the-shelf chip and is physically small and provides digital output.
Furthermore, the sensor has a soft exterior for safe interactions with the environment; in particular it
uses soft silicone with about an 8 mm thickness. Tests were performed to evaluate the drift due to
temperature changes, and a compensation using the integral temperature sensor was implemented.
Furthermore, the hysteresis and the crosstalk between the 3-axis measurements were evaluated.
The sensor is able to detect minimal forces of about 1 gf. The sensor was calibrated and results with
total forces up to 1450 gf in the normal and tangential directions of the sensor are presented. The test
revealed that the sensor is able to measure the different components of the force vector.

Keywords: tactile; skin; sensor; magnetic

1. Introduction
Tactile sensors are crucial for a safe and robust interaction of a robot with its environment.
They provide the most direct measurements of the contact forces during planned and accidental
contacts. In order to appropriately react to contact forces, it is beneficial if the force sensors are
distributed in the robot skin and measure not only normal forces, but also shear forces. Moreover, soft
skin is important for softening impact forces and also for robust object grasping. Several studies have
been conducted in recent years to develop a distributed tactile sensor for robot skin [1]. For example,
a soft capacitive-type sensor that could measure force in 3-axis was introduced in [2]. Although this
sensor had high sensitivity and could measure normal and shear forces, each sensor required space of
approximately 13 mm ˆ 13 mm (from the center of one sensor to another). Moreover, the manufacturing
required a lot of manual work.
In this paper we present a Hall effect–based tactile sensor that can measure three-directional force
data in limited space. It is both physically small and easy to produce. We evaluate the feasibility
of measuring 3-axis force while maintaining a soft exterior for safe interactions. While in previous
work [3] we presented preliminary results obtained with the sensor, the current paper provides a
detailed sensor characterization. In particular, drift due to temperature changes and the hysteresis in
the sensor measurements are analyzed. Furthermore, a calibration to counteract crosstalk between the
axes and to extract more precise force vector measurements is provided. Moreover, the design of a
printed circuit board (PCB) is presented to show how 16 chips (each measuring 3-axis force) can be
placed on a 20 ˆ 23–mm-sized PCB.

Sensors 2016, 16, 491; doi:10.3390/s16040491 www.mdpi.com/journal/sensors

Sensors 2016, 16, 491 2 of 11

The rest of this paper is organized as follows. In Section 2 we provide a review of related tactile
sensors. Section 3 describes the sensor principle and the production process. Section 4 presents the
experimental procedure that was used to evaluate the sensor and Section 5 shows the results. Section 6
shows a PCB for distributed sensing, and Section 7 draws conclusions and presents future work.

2. Related Works
There are many sensing principles which can be employed for tactile sensing [4]. This review
will focus on sensors that are capable of measuring normal as well as shear forces in robotic skin.
Touchence [5] sells a thin, small-sized 3-axis tactile sensor based on MEMS piezoelectric elements,
but the sensor is rigid and the necessary additional electronics are bigger than the sensor itself.
A small-sized optical sensor that can measure both normal and shear forces was already proposed
more than 10 years ago in [6], but to the best of the authors’ knowledge it has not been integrated in a
robotic system yet. Optical 3-axis sensors have also been integrated into soft sensor flesh [7], and those
sensors have recently been made commercially available by Touchence as well. A smaller (10 mm
wide and 8 mm high) but stiffer optical sensor that can measure the force vector is currently available
from OptoForce [8]. An implementation of four tri-axial sensors for a robotic fingertip is presented
in [9], and 3-axis F/T (force/torque) sensors were also integrated in the soft skin of the robot Macra [8],
both based on strain-gauges. A capacitive 3-axis sensor embedded in soft silicone is presented in [2].
The idea of using Hall effect sensors and magnets to measure a tactile response was originally
proposed in [10,11], where only preliminary prototypes were presented, and then this was not
investigated anymore until recently [12–16]. In [14], Hall effect–based tactile sensors are integrated on
a robot hand and used in object classification experiments; a full characterization of the sensor and
more real world experiments are reported in [15]. However, the sensor is based on one-dimensional
(1D) Hall effect sensing, and therefore only normal forces can be measured. The work in [12] instead
proposes a three-dimensional (3D) sensor, with a design similar to the one we present; however,
no accurate characterization is reported, and therefore it is not easy to evaluate the quality of the
measurements. The work described in [13,16] is more mature, and it has been successfully applied to
real robotic scenarios; however, even though many simulation analyses are presented, this work also
lacks a complete characterization of the real sensor. Moreover, the design they propose (with a rubber
dome and four Hall effect sensors) imposes constraints on the minimum size of the whole system.
A magnetic-based tactile sensor for fingertips has been commercially produced [17]. However, the
output signal from this sensor has to be amplified first before it can be read by a microcontroller.
The amplifier has a rather big size, meaning that a lot of space is required for integrating this device
into a robot. Instead, the current paper uses a single small-sized chip with digital output and the
surface of the sensor is flat. Only a magnet embedded in soft silicone is required in addition in order to
measure 3-axial forces.

3. Sensor Description
The sensor described in this paper is easy to produce and requires only a few tools. This section
describes the sensor structure as well as the production process.

3.1. Sensor Concept

The force vector can be detected by measuring a magnetic field change. To achieve that, a
magnetometer (MLX90393) from Melexis [18] is used. A single MLX90393 chip is capable of providing
3-axis magnetic data and temperature data through the I2C fast mode protocol (four wires). The chip
is mounted on a printed circuit board (PCB). We embedded the chip below a soft material, specifically
silicone rubber, and implanted a small magnet above it as shown in Figure 1. The soft material acts
as a compliant layer, and also transmits the force applied on the top surface. As a result, the small
magnet will be displaced from its initial position when force is applied, causing a magnetic field
Sensors 2016, 16, 491 3 of 10
Sensors 2016, 16, 491 3 of 11
magnetic field change. For the experiments in this paper, a PCB with one single chip is used.
Sensors 2016, 16, 491
Section
3 of 10
VI of this paper shows how multiple chips can be implemented on one PCB.
change. For the
magnetic experiments
field change. For in
thethis paper, a in
experiments PCBthiswith one
paper, single
a PCB chip
with oneissingle
used.chip
Section 6 of
is used. this paper
Section
shows how multiple chips can be implemented on one PCB.
VI of this paper shows how multiple chips can be implemented on one PCB.

Figure
Figure 1. Conceptual
1.
FigureConceptual design
1. Conceptual design(left)
design (left) and
(left)and the
and the molding
the molding process(right).
process
molding process (right).
(a)(a)
(right). (a) Liquid
Liquid
Liquid silicone
silicone
silicone rubber
rubber
rubber was
was was
poured
poured intointo
into
poured the the
the molding
molding cast;
cast;
molding (b)
(b)
cast; AAsmall
(b)A small magnet
smallmagnet wasplaced
was
magnet was placedinside
placed inside
inside the
thethe hole;
hole;
hole; (c) More
(c)
(c) More More liquid
liquid
liquid silicone
silicone
silicone
rubber
rubber waswas
rubber
was poured
poured above
poured the
above
above the
the first layer
firstlayer
first to
layerto cover
tocover thesmall
cover the
the smallmagnet.
small magnet.
magnet.

3.2. Soft Outer

3.2. Soft Layer
3.2. Soft Outer
Outer Layer
Layer
The The
The
following
following
following
steps
steps
steps areare
are
requiredfor
required
required
for the
for the
the molding
molding
molding
process.
process.
process.
First,
First,
First,
an MLX90393
an an MLX90393
MLX90393
chipchip
is placed at
is placed
chip is placed at
at the
the middle of a molding cast, supported with four guidance points and double-sided tape. The chip
the middle
middle of a
of a molding molding cast, supported with four guidance points and double-sided tape. The chip
is covered by liquidcast,
siliconesupported with Supersoft
rubber (Ecoflex four guidance points and
from Smooth-On, double-sided
shore hardness 00-30). tape.Please
The chip
is covered
is covered by liquid
byoptimal silicone
liquidmaterial rubber
siliconeselection (Ecoflex
rubber and (EcoflexSupersoft
Supersoftfrom Smooth-On,
from Smooth-On, shore hardness
shore hardness 00-30). Please
note that optimization of the thickness of the compliant layer is not00-30).
note
Please that
thenote
optimal
focus that material
optimal
of this
selection
paper.material
and
selection
To distribute
optimization
and optimization
the magnetic
of the
field evenly,
thickness
ofthe
theposition of
thickness the compliant
of magnet
of the the compliant layer
should be
is not
layer is
the focus
centeredof this
abovepaper.
the To
chip. distribute
A guidance the
lid magnetic
(Figure 1a) field
is used evenly,
to createthea position
hole for
not the focus of this paper. To distribute the magnetic field evenly, the position of the magnet should be of
placingthe a magnet
small should
magnet be
centered above
in theabove
centered the
center.the chip.
Thechip.
magnet A guidance
A guidance lid (Figure
for the current
lid (Figure 1a) is
is used
implementation
1a) used to
is ato create
create aa hole
Neodymium hole for
for placing
magnet coated with
placing aa small magnet
nickel
small magnet
in the center.
(Nd-Fe-B) The
with magnet
a dimensionfor the
of current
2 mm × implementation
2.5 mm × 1.7 mm. is a
AfterNeodymium
the
in the center. The magnet for the current implementation is a Neodymium magnet coated with nickelfirst layer magnet
of coated
silicone has with
cured, nickel
(Nd-Fe-B) withisaa placed
the magnet
(Nd-Fe-B) with dimension
dimension insideofthe2 hole,
of mm ˆ
2 mm × 2.5
and mm
2.5more × 1.7
mmliquid
ˆ mm.
mm. After
1.7silicone rubber
After the
theisfirst
usedlayer
first of
to cover
layer of silicone
silicone has has cured,
it. The silicone
cured,
the layer above
magnet is the PCB
placed is 8 mm
inside thethick
hole,overall,
and with liquid
more the small magnetrubber
silicone covered isby approximately
used to cover it.2The
mm silicone
of
the magnet is placed inside the hole, and more liquid silicone rubber is used to cover it. The silicone
layersilicone,
layer above the
above
as inPCB
the Figure
PCB is is 882.mm
In our experiments
mm thick
thick overall,
overall, with
there wassmall
with the
a good bond between
the small magnet
magnet covered
covered by
the
byfirst and second layer
approximately
approximately 22 mm
mm ofof
of silicone. The silicone covers an area of 55 mm × 55 mm.
silicone, as
silicone, as in
in Figure
Figure 2.
2. In
In our
our experiments
experiments there
there was
was aa good
good bond
bond between
between the
the first
first and
and second layer
second layer
of silicone. The silicone covers an area of 55 mm × 55
of silicone. The silicone covers an area of 55 mm ˆ 55 mm. mm.

Figure 2. The prototype of the Hall effect–based skin sensor.

Figure 2. The prototype of the Hall effect–based skin sensor.

sensor.
Sensors 2016, 16, 491 44 of
of 11
10

4. Experimental Setup
4. Experimental Setup
Three experiments were conducted to understand the characteristics of the sensor. This section
Three
explains theexperiments were
experimental conducted
setups to understand
and procedures the characteristics
that were used during the oftests.
the sensor. This section
explains the experimental setups and procedures that were used during the tests.
4.1. Temperature Drift Test
4.1. Temperature Drift Test
This test studies the effect of thermal drift on the sensor reading. The skin sensor was placed
This test studies the effect of thermal drift on the sensor reading. The skin sensor was placed
inside an oven along with a Sparkfun TMP102, which is an I2C temperature sensor for measuring the
inside an oven along with a Sparkfun TMP102, which is an I2C temperature sensor for measuring the
temperature inside the oven during the test. The MLX90393 chip also includes a temperature sensor.
temperature inside the oven during the test. The MLX90393 chip also includes a temperature sensor.
The experiment started with the room temperature of 27 ˝°C, then the sensor was heated up until the
The experiment started with the room temperature of 27 C, then the sensor was heated up until the
skin sensor reached 40 °C. Afterwards, the oven was turned off and the door of the oven was opened
skin sensor reached 40 ˝ C. Afterwards, the oven was turned off and the door of the oven was opened
to let the temperature drop to around 30 ˝°C. The temperature value and the skin sensor’s readouts
to let the temperature drop to around 30 C. The temperature value and the skin sensor’s readouts
were recorded using Arduino Due, stored in an SD (secure digital) card.
were recorded using Arduino Due, stored in an SD (secure digital) card.
4.2. Hysteresis Test
4.2. Hysteresis Test
A viscoelastic material such as silicone can introduce hysteresis in the sensor’s force
A viscoelastic material such as silicone can introduce hysteresis in the sensor’s force measurements.
measurements. To evaluate the hysteresis, the skin sensor was placed on top of an acrylic platform
To evaluate the hysteresis, the skin sensor was placed on top of an acrylic platform table tilted 45 degrees
table tilted 45 degrees in the y-axis direction. The sensor was pushed for 5 min with a load of 1450 gf
in the y-axis direction. The sensor was pushed for 5 min with a load of 1450 gf (the maximum load that
(the maximum load that our experimental setup can achieve). To perform this, a voice coil motor
our experimental setup can achieve). To perform this, a voice coil motor (VM5050-190 from Geeplus),
(VM5050-190 from Geeplus), a linear bushing, an aluminum shaft adapter, a six-axis force/torque
a linear bushing, an aluminum shaft adapter, a six-axis force/torque (F/T) sensor (Nano1.5/1.5 from
(F/T) sensor (Nano1.5/1.5 from BL Autotech) for monitoring the pushing force during the experiment,
BL Autotech) for monitoring the pushing force during the experiment, and a 30 ˆ 30 mm acrylic push
and a 30 × 30 mm acrylic push plate, which is used to push on the proposed sensor, were utilized.
plate, which is used to push on the proposed sensor, were utilized. The configuration for this test can
The configuration for this test can be seen in Figure 3. Two microcontrollers were required due to the
be seen in Figure 3. Two microcontrollers were required due to the different input voltage of our sensor
different input voltage of our sensor (3.3 V) and the F/T sensor (5 V). The data from both sensors were
(3.3 V) and the F/T sensor (5 V). The data from both sensors were recorded into the SD cards installed
recorded into the SD cards installed on Arduino Due and Uno with the synchronized sampling rate
on Arduino Due and Uno with the synchronized sampling rate of 100 Hz in our experiments (the
of 100 Hz in our experiments (the maximum sampling rate of the Hall effect sensor is about 240 Hz
maximum sampling rate of the Hall effect sensor is about 240 Hz to measure all three axes). Finally, the
to measure all three axes). Finally, the voice coil motor applied no force on the sensor for one minute,
voice coil motor applied no force on the sensor for one minute, and afterwards the acrylic push plate
and afterwards the acrylic push plate was retracted. The silicone is sticky, and therefore a force in the
was retracted. The silicone is sticky, and therefore a force in the minus z direction is recorded when the
minus z direction is recorded when the acrylic push plate is removed.
acrylic push plate is removed.

Figure 3. Experiment
Experiment setup
setup used
used in
in this paper. The
The right
right side shows the addition of the adjustable
angle tilt stage and the angle push plate, in particular the 30-degree setup where the stage is adjusted
to 30 degrees and the 30-degree push plate is used.

4.3. Load
Load Test
To calibrate
calibratethe sensor
the andand
sensor to evaluate its capability
to evaluate of tri-axial
its capability of force measurement,
tri-axial two experiments
force measurement, two
were conducted.
experiments wereThe first experiment
conducted. was
The first a normal force
experiment was atest whereforce
normal multiple magnitudes
test where of normal
multiple force
magnitudes
Sensors 2016, 16, 491 5 of 11

of normal force were applied on the sensor’s top surface. In the second experiment, the sensor was
pushed with both normal and shear force in different angles and with different force magnitudes.
The configuration for this experiment was similar to the hysteresis test. In the normal force
experiment, the skin sensor was mounted directly on the flat and sturdy X-Y table. In the shear force
experiment, the sensor was mounted on an adjustable angle tilt stage (Figure 3). This acrylic stage
was fixed to the X-Y table. The angle can be adjusted in three different positions that are 15, 30, and
45 degrees, and corresponding acrylic push plates with the same angles were used.
In both the normal and shear force experiments, the sensor was pushed by the voice coil motor
with a stepwise force and its magnitude was increased every 10 seconds. The applied force was ranging
from approximately 70 gf to 1450 gf. A Savitzky-Golay filter was utilized to filter all data (polynomial
order = 4, frame size = 21). We performed the normal force test and the shear force tests with 15, 30,
and 45 degree, in four directions (+/´ x/y direction).

5. Results and Discussion

5.1. Thermal Drift Evaluation

From Figure 4 it can be clearly seen that the Hall effect sensor measurements change with the
changing temperature, even though the temperature change measured by the chip was slower than
the external temperature. A possible explanation for the change in the Hall effect measurements is an
expansion of the silicone packaging with higher temperature. The test also revealed that the sensor
reading in the z-axis was the most affected by the temperature change, but also the x-axis and y-axis
measurements slightly changed; a closer look reveals that the y-axis is more affected than the x-axis.
This is in accordance with the results presented in Section 5.3, which show that changes in the z-axis
also affect the x-axis and y-axis, which might be due to a slightly misaligned magnet. The graph shows
that the sensor changes are proportional to the temperature change measured by the chip, meaning
that temperature compensation can possibly be performed.
A linear regression was conducted to find the coefficient k for calibrating the sensor’s outputs.
We selected a Huber robust model for this task. The temperature compensation was calculated
as follows:
Si,T “ Si ´ k i ˆ ∆ST (1)

where:

‚ i is each axis of the skin sensor (x, y and z).

‚ Si and Si,T are the skin sensor readout and compensated value, respectively.
‚ ∆ST is the temperature change measured by the MLX90393 built-in temperature sensor.

To evaluate the temperature compensation performance, another test was conducted. This time,
the temperature was raised to 35 ˝ C. Figure 5 shows a comparison between the sensor’s readout before
and after the compensation was applied. A moving average of the temperature was used for the
compensation. After being compensated, the z-axis maximum value was around 600 digits (over a
full scale of 65,500 digits), which corresponds to 120 g in our experiments (contact size 30 ˆ 30 mm).
Further improvements are likely possible by employing a high-pass filter. Furthermore, small steps
can be seen on Sy and Sz . The cause of those is not clear to the authors and will be further investigated
in future work.
Sensors 2016, 16, 491 6 of 11
Sensors 2016, 16, 491 6 of 10
Sensors 2016, 16, 491 6 of 10

60 10000
60 ST 10000
ST
External temp
External temp
Sx
S
Syx

(digits)
S

readout(digits)
(°C)

Szy
Temperature(°C)

Sz
Temperature

Sensorreadout
40
40

Sensor
0
0

20
200 500 1000 1500 2000 2500 3000
0 500 1000 1500(s)
Time 2000 2500 3000
Time (s)

Figure 4.
Figure Theeffect
4. The effectof
oftemperature
temperature changes
changes on
on the
the sensor
sensor measurement.
measurement.
Figure 4. The effect of temperature changes on the sensor measurement.

Sensor before temperature compensation Sensor after temperature compensation

Sensor before temperature compensation Sensor after temperature compensation
35 3000 35
35 ST 3000 35 ST
ST ST
34 2500 34 Sx,T
34 Sx 2500 34
(digits)

Sx Sx,T

(digits)
readout(digits)

readout(digits)
Sy Sy,T
(°C)

33
(°C)

33 2000 Sy,T
Sy
Temperature(°C)

33
Temperature(°C)

33 2000
Sz Sz,T
Sz 32 Sz,T
Sensorreadout
Temperature

32 1500
Temperature

Sensorreadout
32 1500 32
31 1000 31
31 1000 31
Sensor

Sensor
600
30 500 30 600
400
30 500 30 400
200
29 0 29 0200
29 0 29 0
28 -500 28
280 200 400 600 800 1000-500 280 200 400 600 800 1000
0 200 400 600 800 1000 0 200 400 600 800 1000
Time (s) Time (s)
Time (s) Time (s)

Figure 5. A comparison between the sensor’s readout

readoutbefore and after temperature compensation was
Figure
Figure 5.5. A comparison between the sensor’s readout before and after temperature compensation was
applied.
applied. The right figure also shows the moving average of the sensor temperature, which was used
applied. The
The right
right figure
figure also
also shows
shows the
the moving
moving average
average of
of the
the sensor
sensor temperature,
temperature, which
which was
was used
used
for
for calibrating
calibrating the
the sensor.
sensor.
for calibrating the sensor.

5.2. Hysteresis Evaluation

5.2. Hysteresis Evaluation
As expected, there is hysteresis in the sensor measurements. In the result presented in Figure 6,
As expected,
expected, there
thereisishysteresis
hysteresisininthe
thesensor
sensormeasurements.
measurements.InInthe theresult
resultpresented
presented inin
Figure 6, 6,
Figure it
it took about one minute for the sensor to reach its quasi-static state, both while loading and
took about
it took one one
about minute for the
minute forsensor to reach
the sensor toits quasi-static
reach state, both
its quasi-static while
state, loading
both while and unloading
loading and
unloading the sensor. The x-axis, which was not loaded, showed only a minor drift. Please note that
the sensor. the
unloading Thesensor.
x-axis, The
which was which
x-axis, not loaded, showed
was not only
loaded, a minor
showed drift.
only Pleasedrift.
a minor note Please
that annote
optimal
that
an optimal soft material selection was not the focus of this study, and the hysteresis can probably be
soft materialsoft
an optimal selection was
material not the focus
selection was notof this study,of
the focus and the
this hysteresis
study, canhysteresis
and the probably becanreduced with
probably be
reduced with different materials, for example closed-cell foams.
reduced with different materials, for example closed-cell foams.
different materials, for example closed-cell foams.
 Sensors2016,
Sensors 16,491
2016,16, 491 77 of
of 10
11

Hysteresis characteristics of skin sensor

5000

Sensor readout (digits)

0

-5000

-10000
Sx
-15000 Sy
Sz
-20000
0 50 100 150 200 250 300 350 400
Time (s)

Hysteresischaracteristics
Figure6.6.Hysteresis
Figure characteristicsof
ofthe
theskin
skinsensor.
sensor.

5.3.
5.3.Load
LoadTests
Tests and
and Calibration
Calibration
The
The results from fromthe theload
load testtest
cancan be seen
be seen in Figure
in Figure 7. Figure7. Figure
7a,b show7a,b the
show Hallthe Hall
effect andeffect and
reference
reference
force sensorforce sensorrespectively,
readout, readout, respectively,
when only when a normalonly a normal
force forceEven
is applied. is applied.
thoughEven onlythough
normal only
force
normal force the
was applied, wasHall applied,
effectthe Hallalso
sensor effect sensora also
detected detected
magnetic fieldachange
magnetic field
in the change
y-axis. in the effect
A related y-axis.is
A related
also effectinisSection
described also described
5.1, and we insuspect
Sectionthat 5.1,the
and we suspect
orientation thatsmall
of the the orientation
magnet wasof notthe small
perfectly
magnet
alignedwas withnottheperfectly
sensor and aligned
caused with theFurthermore,
this. sensor and caused this. is
as silicone Furthermore, as silicone is forces
soft but incompressible, soft butin
incompressible,
one axis can cause forces in one axis
movements can cause
in another movements
direction, as theinsilicone
anothermovesdirection,
awayasfrom the silicone moves
the pressure.
awayFigure
from the7c pressure.
shows the Hall effect sensor response when a 45-degree shear force is applied in the
Figure
y-axis. Even7cthough
shows no theforce
Hall was
effectapplied
sensor in response when
the x-axis, a 45-degree
as can be also seenshearfromforcetheis measurements
applied in the y- of
axis. Even though
the reference sensor noinforce was
Figure 7d,applied
the Hall ineffect
the x-axis,
sensorasalso
candetected
be also seena smallfrom the measurements
magnetic field changeof in
the
the reference sensor Furthermore,
x-axis direction. in Figure 7d, the the y-axis
Hall effect sensorsensor
Hall effect also detected
measurements a small aremagnetic
bigger than fieldthe
change
z-axis
in the x-axis direction.
measurements, Furthermore,
even though the force theiny-axis Hall effect
the z-axis sensor Due
was bigger. measurements are bigger
to this cross-talk and than the
different
z-axis measurements,
magnitude of the skin even sensorthough
response, the aforce in the was
calibration z-axis was bigger. Due to this cross-talk and
performed.
different
We magnitude
used different of the skin sensor
models response,
to calculate a calibration
the x, y and z forces was from
performed.
the Hall effect sensor values,
withWe theused different models
measurements of thetosix-axis
calculateF/T thesensor
x, y and z forces
as the from the
reference. To Hall effectthe
calculate sensor values,
parameters
with thecalibration,
for the measurements of thefrom
datasets six-axisall F/T
anglessensor
were asused.
the reference.
For the To calculatepurpose,
evaluation the parameterswe used fornew
the
calibration, datasets from all angles were used. For the evaluation
datasets that were not used for the calibration. Robust Huber regression was used (MATLAB functionpurpose, we used new datasets
that were not used
LinearModel.fit); forsquares
least the calibration.
regression Robust
performed Huber
nearlyregression
the samewas as theused
robust(MATLAB function
Huber regression.
LinearModel.fit);
A quadratic model least squares regression
performed better than performed nearlyas
a linear model, thecan
same as theby
be seen robust
higher Huber regression.
R-square values
Ainquadratic
Table 1. model
Also, aperformed
neural network better than
(one ahidden
linear model, as can
layer with 20be seen by
hidden higher
units) was R-square
trainedvalues
with thein
Table
same 1. Also, adata
training neuralthatnetwork
we used (one hidden layer with
for approximating 20 hidden of
the parameters units) was trained
the linear with the
or quadratic same
equation.
training
The neural data that weperformed
network used for approximating
better for the test thecase
parameters
with only ofnormal
the linear or quadratic
force, but overallequation. The
the quadratic
neural
equationnetwork performed
performed best. better for the testresult
The calibration case with
usingonlythe normal
quadratic force, but overall
model is shown theinquadratic
Figure 8.
equation
The graphs performed
show the best. The calibration
comparison between resulttheusing
forcethe quadratic
detected using model
the isF/Tshown
sensor in Figure
and the8.forceThe
graphs show the comparison between the force detected using the F/T
calculated using the skin sensor. A good correspondence between the measurements can be observed, sensor and the force calculated
using the biggest
with the skin sensor. A goodincorrespondence
differences the normal loadbetween measurements can be observed, with the
case for the z-axis.
biggestAsdifferences in the normal
a final evaluation, load case for
the minimum the z-axis.
detectable load was evaluated. Rubber weights with a
As a of
diameter final
aboutevaluation,
1 cm were theplaced
minimum on thedetectable
sensor and load
it waswasfound
evaluated. Rubber
that a force weights
of about 1 gfwith a
in the
diameter of aboutsensor
z-axis produced 1 cm were placed on that
measurements the sensor and than
are higher it wasthe found that anoise
observed forceof ofthe
about 1 gf as
sensor, in can
the z-be
axis
seenproduced
in Figure sensor
9. measurements that are higher than the observed noise of the sensor, as can be
seen in Figure 9.
 Sensors 2016, 16, 491 8 of 11
Sensors 2016, 16, 491 8 of 10

Sensors 2016, 16, 491skin sensor

Uncalibrated 6-axis F/T sensor
Uncalibrated skin sensor 6-axis F/T sensor8 of 10
80002016, 16, 491
Sensors 1500
5000 1000 8 of 10
Sx
Uncalibrated skin sensor Fx F/T sensor
6-axis Uncalibrated skin sensor 6-axis F/T sensor
6000
8000 1000
1500 5000 1000
S Fy 500
Uncalibrated
Syx skin sensor 6-axis
Fx F/T sensor 0 Uncalibrated skin sensor 6-axis F/T sensor

(digits)
t (d igits)

4000
6000
8000 Sz 1500 5000
500
1000 F 1000
S 500
FFxy
z
Sxy 0 0

(digits)
igits)

2000
4000

(gf) (gf)
-5000

(gf) (gf)
6000
SSz 1000

readout
0
500 F
t (dou

Fyz 500
0
y 0 -5000

(digits)
igits)

(gf) Force

(gf) Force
read

2000
4000 Sz -5000
5000 F

readout
-500
t (dou

z -10000
-2000 0
readSensor

Sensor
20000 -500
ForceForce

ForceForce
S
read

-5000 x -1000

readout
-1000 0 Fx
ou

-4000 -500 -10000

-2000
0 SSy -500
Sensor

Sensor
-15000 -1000 F
-500
-1500
-1000
x -1500 Fyx
-6000
-4000
-2000 -10000
SSS
Sensor

Sensor
-15000 zxy -1000 F
FFxy
z
-8000 -1000
-2000
-1500 -20000 -1500
-2000
-60000
-4000 20 40 60 80 0 20SS 40 60 80
0 20 40 60 80 0 20 40 60 80
Time (s) Time (s)
-15000 z
yTime (s) -1500 FFzTime (s)
-8000 -1500 y
-6000 -2000 -20000 -2000
0 (a)
20 40 60 80 0 (b)20 40 60 80 0 (c) S
20 z
40 60 80 0 (d)F20 40
z Time (s)
60 80
Time (s) Time (s) Time (s)
-8000 -2000 -20000 -2000
0
Figure 20
(a)The
7. 40 sensor’s
60 80 readout (a) 0 and
20
(b)the 40 corresponding
60 80 0 from
force 20 40
(c) F/T 60 80 (b) when
sensor 0 only
20 40
(d)normal 60 80
Figure 7.TimeThe(s) sensor’s readout (a) and the(s)corresponding force from
Time TimeF/T
(s) sensor (b) when onlyTime normal
(s)
force
forceisis(a)
applied; TheThesensor’s
sensor’sreadout (b) the(c) and
andthe the corresponding (c) force from F/T sensor (d)
(d) when
Figure 7.applied;
The sensor’s readout readout
(a) and (c)corresponding corresponding
force from forcesensor
F/T from (b) F/T sensor
when only (d) when
normal
45-degree
45-degree shear
shear force
force isisapplied
applied in
inthe
they-direction.
y-direction.
force is7.applied;
Figure The readout
The sensor’s sensor’s (a) readout
and the (c)corresponding
and the corresponding
force from F/T force from(b)
sensor F/Twhensensoronly (d)normal
when
45-degree
force issensor
Calibrated
shear
applied; force is applied
The sensor’s
using quadratic
in the
regression readout
y-direction.
(c) and
Calibrated the
sensor corresponding
using quadratic regressionforce from F/T sensor (d) when
Calibrated sensor using quadratic regression
500 500
45-degree shear force is applied in the y-direction. 500
Calibrated sensor using quadratic regression Calibrated sensor using quadratic regression
Calibrated sensor using quadratic regression
500
0 500
0 0
500
Calibrated sensor using quadratic regression Calibrated sensor using quadratic regression
Calibrated sensor using quadratic regression
F F F
500 x 5000 x x
Force (gf)
Force (gf)

Force (gf)
-5000 -500 5000
-500
Fy F Fy
y
F F F
Fx 0 Fzx Fx
(gf) (gf)
(gf) (gf)

0
Force (gf)

-500
-1000 z -500
-1000 -1000
-5000 z
F F F
SFxy SFxy SF y
Force Force

Force Force

x,c x,c x,c

x
Force (gf)

-500 F -500 F -500 F

-1000
-1500 SFyz -1000
-1500 SFyz -1000
-1500 z
SFy,c
y,c y,c y
S S S
S x,c
SFzx,c F
Sz x,c Fz,c
-1000
-1500 z,c -1000 z,c -1000
-1500
-2000 S
z
-2000 S -1500
-2000 S
0 S y,c20 40 60 0 S y,c20 40 60 0 S y,c20 40 60
x,c x,c x,c Time (s)
S Time (s) S Time (s) Sz,c
-1500
-2000 S z,c -1500 S z,c -1500
-2000 S
y,c -2000 y,c
0 20 (a) 40 60 0
y,c
20 (b) 40 60 0 20 (c) 40 60
S Time (s) S Sz,c Time (s)
z,c z,c Time (s) -2000
-2000 -2000
0 0 20 40 60
Figure 8.20(a)(a)
Time
40
Normal
(s)
60
force; (b) 45-degree 0 20 (b)
(y-axis 40 60 (c) 45-degree (x-axis
direction);
Time (s)
(c)
direction)
Time (s) force
calibration result. Sx,c, Sy,c and Sz,c are the calibrated
(a)Normal skin sensor measurements.
(b) (c)
Figure 8. (a) force; (b) 45-degree (y-axis direction); (c) 45-degree (x-axis direction) force
Figure 8. (a) Normal force; (b) 45-degree (y-axis direction); (c) 45-degree (x-axis direction) force
calibration
Figure
calibration result.
8. (a) Normal
result. Sx,c, Sy,c and Sz,c are the calibrated skin sensor measurements.
Sx,c ,force; (b)Sz,c
Sy,c and 45-degree (y-axis direction); (c)measurements.
45-degree (x-axis direction) force
4 are the calibrated skin sensor
calibration result. Sx,c, Sy,c and Sz,c are the calibrated skin sensor measurements.
4

24
(gf) (gf)
(gf) Force

02
ForceForce

-20
1 2 3 4 0
Time (s)
-2
0 1 2 3 4
Figure 9. The z-axis calibrated sensor
-2 measurements (S(s)
Time z,c) when a weight of 1 g (contact area of about
0 1
1 cm2) is placed on the sensor at a time of around 2 s.2The
Time
3
calibration is4 slightly incorrect in this case,
Figure 9. The z-axis calibrated sensor measurements (S(s)
z,c) when a weight of 1 g (contact area of about
showing a measurement of 2 gf. This might be partially due to the different contact area than that
1 cm2) is
Figure placed on the sensor at a time of around 2 s.(SThe calibration is slightly incorrectarea
in this case,
Figure 9.The
during 9.
the z-axis
z-axiscalibrated
calibration.
The sensor
sensoritmeasurements
Nevertheless,
calibrated can be seen that
measurements (Sz,c
the) when
sensor
z,c ) when
aaweight
can ofof11gdetect
already
weight g(contact
a 1 garea
(contact ofofabout
weight. about
showing
11cm a measurement of 2 gf. This might be partially due to the different contact area than that
cm2) )isisplaced
placedon onthe
thesensor
sensoratataatime
timeof ofaround
around22s.s.The
Thecalibration
calibrationisisslightly
slightlyincorrect
incorrectininthis
thiscase,
2
case,
during
showing the calibration. Nevertheless, it can be seen that the sensor can already detect a 1 g weight.
showingaameasurement of
of22gf.
Table 1. R-squared
measurement This
value
gf. might
Thisfor be
bepartially
the normal
might forcedue
partially dueto
and tothe
shear different
theforce contact
contactarea
experiments.
different areathan
thanthat
that
during the calibration. Nevertheless, it can be seen that the sensor can already detect
during the calibration. Nevertheless, it can be seen that the sensor can already detect a 1 g weight. a 1 g weight.
Table 1. R-squared
Linear value for
+ Huber the normal
Quadratic + force
Huber and shear force experiments.
Feedforward Neural Network
Normal Force 0.8634
Table 1.Linear
R-squared
+ Huber 0.8925
value forQuadratic
the normal+force
Huberand shear 0.9368
force experiments.
Feedforward Neural Network
Shear 45—y 0.8634 0.9418 0.8275
Normal Force Linear 0.8634
+ Huber Quadratic 0.8925
+ Huber Feedforward0.9368 Neural Network
Shear 45—x 0.9272 0.9744 0.9644
Shear 45—y
Normal Force 0.8634
0.8634 0.9418
0.8925 0.8275
0.9368
Shear 45—y
Shear 45—x 0.9272
0.8634 0.9744
0.9418 0.9644
0.8275
Shear 45—x 0.9272 0.9744 0.9644
Sensors 2016, 16, 491 9 of 11

Table 1. R-squared value for the normal force and shear force experiments.

Linear + Huber Quadratic + Huber Feedforward Neural Network

Normal Force 0.8634 0.8925 0.9368
Shear 45—y 0.8634 0.9418 0.8275
Shear 45—x 0.9272 0.9744 0.9644
Sensors 2016, 16, 491 9 of 10

6. Design of PCB for Distributed Sensors

6. Design of PCB for Distributed Sensors
A custom PCB (Figure 10) has been designed to integrate 16 MLX90393 chips for detecting
A custom PCB (Figure 10) has been designed to integrate 16 MLX90393 chips for detecting
distributed force vectors. The PCB has two layers, is only 20 ˆ 23 mm big and has four I2C outputs.
distributed force vectors. The PCB has two layers, is only 20 × 23 mm big and has four I2C outputs.
The distance of the center of one chip to the next one is 4.7 mm. After the PCB has been produced, in
The distance of the center of one chip to the next one is 4.7 mm. After the PCB has been produced, in
future work the crosstalk between the sensors will be evaluated.
future work the crosstalk between the sensors will be evaluated.

Figure 10. A custom PCB for integrating 16 MLX90393 chips.

7. Conclusions
Conclusions
This paper
paper presented
presentedthe
thedesign
designand characterization
and characterization of aofHall effect–based
a Hall soft soft
effect–based skin skin
sensor. The
sensor.
temperature
The test test
temperature shows thatthat
shows thethe
skin sensor’s
skin readout
sensor’s readoutin in
thethez-axis
z-axisdirection
directionwas
wasthe
the one
one mostly
temperature changes.
affected by temperature changes. Using
Using aa built-in
built-in temperature
temperature sensor, drift compensation was
performed. Next, the hysteresis was evaluated. Indeed
performed. Indeed there
there is
is hysteresis
hysteresis inin the
thesensor
sensormeasurements,
measurements,
which is to be expected due to the use of silicone, which is a viscoelastic material. After the calibration
which
when applying varying amounts of normal and shear force, the tests showed that the
of the sensor, when
sensor can measure the components of the force vector. Furthermore, the design of a PCB with 16
3-axis force measurements in limited space is presented. As future work, the crosstalk between those
16 sensors
sensors in
inclose
closeproximity
proximitywill bebe
will evaluated. Furthermore,
evaluated. Furthermore, a spherical magnet
a spherical will be
magnet used
will be in order
used in
to avoid
order possible
to avoid spurious
possible signals
spurious on irrelevant
signals axes.
on irrelevant TheThe
axes. useuseofofdifferent
differentmaterials
materials or
or different
potential means
thicknesses of the packaging will be studied as potential means to to reduce
reduce the
the hysteresis.
hysteresis.

Acknowledgments: This research was partially supported by the JSPS Grant-in-Aid for Scientific Research
Research (S)
No. 25220005, JSPS
No. 25220005, JSPS Grant-in-Aid
Grant-in-Aid for
for Young
Young Scientists
Scientists (B)
(B) No.
No. 15K21443,
15K21443, NEDO
NEDO project
project No.
No. 15657422,
15657422, Research
Research
Institute for Science and Engineering of Waseda University, the Program for Leading Graduate Schools, “Graduate
Institute for Science and Engineering of Waseda University, the Program for Leading Graduate Schools,
Program for Embodiment Informatics” of the Ministry of Education, Culture, Sports, Science and Technology, and
“Graduate
Project Program[PIEF-GA-2013-628315].
LIMOMAN for Embodiment Informatics” of the Ministry of Education, Culture, Sports, Science and
Technology, and Project LIMOMAN [PIEF-GA-2013-628315].

Author Contributions: T.P.T. implemented the sensor and performed the experiments; S.So. assisted with
implementing the sensor, testing the sensor, and designed the test setup; A.S., L.J. and S.Su. conceived and
designed the experiments and provided support. T.P.T. and A.S. wrote the paper. W.H. and H.K. helped in
building and testing the sensor.

Conflicts of Interest: The authors declare no conflict of interest.

Sensors 2016, 16, 491 10 of 11

Author Contributions: T.P.T. implemented the sensor and performed the experiments; S.So. assisted with
implementing the sensor, testing the sensor, and designed the test setup; A.S., L.J. and S.Su. conceived and
designed the experiments and provided support. T.P.T. and A.S. wrote the paper. W.H. and H.K. helped in
building and testing the sensor.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript:

PCB Printed circuit board

MEMS Microelectromechanical system
F/T Force/torque
SD Secure digital

References
1. Dahiya, R.S.; Mittendorfer, P.; Valle, M.; Cheng, G.; Lumelsky, V.J. Directions toward effective utilization of
tactile skin: A review. IEEE Sens. J. 2013, 13, 4121–4138. [CrossRef]
2. Somlor, S.; Hartanto, R.S.; Schmitz, A.; Sugano, S. A Novel Tri-axial Capacitive-Type Skin Sensor. Adv. Robot.
2015, 29, 1375–1391. [CrossRef]
3. Tomo, T.P.; Somlor, S.; Jamone, L.; Schmitz, A.; Hashimoto, S.; Sugano, S. Development of a Hall-Effect Based
Skin Sensor. In Proceedings of the IEEE Sensors Conference, Busan, Korea, 1–4 November 2015.
4. Dahiya, R.S.; Metta, G.; Valle, M.; Sandini, G. Tactile Sensing—From Humans to Humanoids.
IEEE Trans. Robot. 2010, 26, 1–20. [CrossRef]
5. ShokacPot™ ShokacCube™, Innovation unique ‘Soft’ tactile sensor. Available online:
http://www.touchence.jp (accessed on 6 April 2016).
6. Yamada, K.; Goto, K.; Nakajima, Y.; Koshida, N.; Shinoda, H. A sensor skin using wire-free tactile sensing
elements based on optical connection. In Proceedings of the 41st SICE Annual Conference SICE, Osaka,
Japan, 5–7 August 2002; pp. 131–134.
7. Yoshikai, T.; Hayashi, M.; Ishizaka, Y.; Fukushima, H.; Kadowaki, A.; Sagisaka, T.; Kobayashi, K.;
Kumagai, I.; Inaba, M. Development of robots with soft sensor flesh for achieving close interaction behavior.
Adv. Artif. Intell. 2012. [CrossRef]
8. 3D Force Sensor (OMD) Datasheet. Available online: http://optoforce.com/ (accessed on 6 April 2016).
9. Oddo, C.M.; Controzzi, M.; Beccai, L.; Cipriani, C.; Carrozza, M.C. Roughness encoding for discrimination
of surfaces in artificial active-touch. IEEE Trans. Robot. 2011, 27, 522–533. [CrossRef]
10. Clark, J.J. A Magnetic Field Based Compliance Matching Sensor for High Resolution, High Compliance Tactile
Sensing. In Proceedings of the 1988 IEEE International Conference Robotics and Automation, Philadelphia,
PA, USA, 24–29 April 1988.
11. Nowlin, W.C. Experimental Results on Bayesian Algorithms for Interpreting Compliant Tactile Sensing Data.
In Proceedings of the 1991 IEEE International Conference Robotics and Automation, Sacramento, CA, USA,
9–11 April 1991.
12. Ledermann, C.; Wirges, S.; Oertel, D.; Mende, M.; Woern, H. Tactile sensor on a magnetic basis using
novel 3D Hall sensor-First prototypes and results. In Proceedings of the 17th International Conference on
Intelligent Engineering Systems (INES), San Jose, CA, USA, 19–21 June 2013; pp. 55–60.
13. Youssefian, S.; Rahbar, N.; Torres-Jara, E. Contact Behavior of Soft Spherical Tactile Sensors. IEEE Sens. J.
2014, 14, 1435–1442. [CrossRef]
14. Jamone, L.; Metta, G.; Nori, F.; Sandini, G. James: A Humanoid Robot Acting over an Unstructured World.
In Proceedings of the 6th IEEE-RAS International Conference on Humanoid Robots, Genova, Italy, 4–6
December 2006; pp. 143–150.
15. Jamone, L.; Natale, L.; Metta, G.; Sandini, G. Highly sensitive soft tactile sensors for an anthropomorphic
robotic hand. IEEE Sens. J. 2015, 15, 4226–4233. [CrossRef]
Sensors 2016, 16, 491 11 of 11

16. Torres-Jara, E.; Vasilescu, I.; Coral, R. A soft touch: Compliant tactile sensors for sensitive manipulation.
Available online: https://www.researchgate.net/publication/37992348_A_soft_touch_Compliant_Tactile_
Sensors_for_Sensitive_Manipulation (accessed on 6 April 2016).
17. Magnetic Based Tactile Sensor (MTS). Available online: http://bl-autotec.co.jp/ (accessed on 6 April 2016).
18. Melexis, MLX90393 Micropower Triaxis® Magnetometer Datasheet. Available online: http://www.melexis.com/
Position–Speed-Sensors/Triaxis%C2%AE-Hall-ICs/Triaxis%C2%AE-Micropower-Magnetometer-829.aspx
(accessed on 6 April 2016).

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons by Attribution
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).