Instrumentation & Measurement Magazine 24-5 - 30

matrix comprises 16 photoreflectors NJL5908AR with a spatial
resolution equal to 3 mm and a distance from PCB edges equal
to 2 mm, by obtaining a total size for the PCB equal to 13 mm.
A single A/D converter (AD7490) with an on-board microcontroller
(PIC16F1824) have been used to complete the board. A
standard serial interface with a sampling rate of 500 Hz is used
to interrogate the sensor. As described, the deformable layer
has been designed with a flat top side, while the bottom side
presents 16 square cells with a reflective surface of 2 mm side
and intermediate black walls with 1 mm side. The edges of the
deformable layer are 2 mm thick, so that the total size of the sensitive
flat area is 15 × 15 mm2
. The rigid grid has been realized
in ABS via 3D printing technology, by reducing its thickness to
the minimum functional value (0.8 mm). A suitable aluminum
case realized via 3D printing technology has been designed
to house the sensor by maintaining a minimum thickness. All
layers have been bonded among them and into the case, by obtaining
a final prototype with a total thickness for the tip less
than 6 mm. The used silicone has a shore hardness of 26 A, corresponding
to a maximum applicable normal force up to 30 N,
with a resolution of 0.3 N and about 10% of hysteresis.
During last two years, within the European projects REFILLS
and REMODEL, a completely new design for the PCB
board has been developed in order to increase the measurement
performance of the sensor, by modifying the LED driving and
the phototransistor signals acquisition [10]. The selected optical
components are the same (NJL5908AR) and also the spatial
distribution (a 5 × 5 matrix with a distance among taxels equal
to 3.55 mm), by obtaining a sensitive area equal to about 22 ×
22 mm2
below. Domed cups with different curvature radii allow different
performance in terms of reconstructed forces and stability
of contact area to be obtained. Fig. 1 reports the timeline of the
tactile sensing technology evolution through the years and the
European projects, providing pictures and the main features of
the prototypes described in this section.
Sensor Measurements and
Characterization
All described prototypes, in rest condition, measure for each
i-th taxel a positive offset voltage vi
(0). If contact with external
objects occurs, the taxel shows a voltage variation
Δvi
(t) = vi(t)−vi(0), related to the deformable layer deformations.
The voltage measurements Δvi correspond to the tactile
map available as sensor output. From a graphical point of
view, in order to report Δvi
sentations can be used: a first representation which reports
the voltage variation as a red circle with radius proportional
to the acquired Δvi
matrix of cells whose colors change with Δvi
; and a second representation which uses a
, according to a de.
However, while in the previous version LEDs were
parallel driven with a voltage supply, in this design the LEDs
are in series and current driven through an adjustable current
source (LM334) in order to increase emitted light stability, by
reducing its drift due to temperature and power supply noise.
Additionally, analog buffers have been introduced to decouple
the photoreflector voltage signals from the A/D converter
stage. They have been realized by using the low power operational
amplifiers ADA4691 and used to directly connect signals
to a PIC microcontroller (PIC16F19175) with integrated 12-bit
A/D channels. This solution (without a separate A/D device
with SPI interface) allowed a simplification of the interrogation
firmware and an enhancement of the signal-to-noise ratio.
All components are integrated into a single PCB, by obtaining
a fully integrated sensor with a programmable device used for
data acquisition. In addition to a standard serial interface with
a sampling frequency equal to 500 Hz, an alternative interface
(e.g., SPI, I2C, Wireless) can be implemented. The rigid grid has
been re-designed in order to exploit PCB edges for the alignment
with the optoelectronic components and for the bonding.
The deformable pad can be realized with different shapes on the
basis of the application scenario. A flat surface can be well exploited
if the object shape recognition is requested for the task
implementation. Instead, if the robotic task needs information
about contact forces and moments, it is possible to reconstruct
them from the tactile image by using a domed deformable layer
with a suitably trained neural network, as better described
30
fined colorbar. This tactile map can be directly used for object
shape recognition or elaborated to estimated derived quantities
such as contact force and torque components, to exploit in
feedback control systems. Fig. 2 reports some examples of typical
voltage signals and tactile maps used for the reconstruction
of a grasped wire shape and a pen. The figure shows both possible
map representations for different prototypes. The sensor
measurement performance mainly depends on the mechanical
properties of the material and the shape of the deformable
layer and by the operating points and optical characteristics
of the optoelectronic devices. With the aim to characterize the
properties of proposed tactile sensor, since tactile commercial
solutions are not available as reference sensors, a 6-axis force/
torque sensor (manufactured by ATI) has been used as reference.
This paper reports the characteristics of the most recent
version presented above, assembled with the reference sensor
as reported in Fig. 2. For example, the hysteresis can be
evaluated by comparing the voltage variation Δvi
with respect to normal force fz
for a taxel
measured by the reference sensor.
From the data, reported in Fig. 2, a hysteresis mean error
of about 10% has been computed, also by considering different
velocity in force application. By using the characterization
setup, additional properties can be easily derived, such as the
repeatability in terms of Δvi
obtained by applying the same
force profile, which presents a maximum error less than 7%.
The time response has been evaluated by applying a force with
a step change to the sensor: the obtained value is less than 1ms.
Moreover, by computing the Power Spectral Density of a typical
voltage signal, it is possible to evaluate that the noise level
is below the signal level of about four orders of magnitude.
The contact interaction between the sensor pad and an item
is not described only by a tactile map, but also by friction forces
and torques. Such quantities are of paramount importance
during a grasp; indeed, the friction measurement can be used
to conveniently choose the grasp force, both to firmly grasp an
object and to allow controlled sliding.
IEEE Instrumentation & Measurement Magazine
August 2021
amplitudes, two possible repre
Instrumentation & Measurement Magazine 24-5

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 24-5
