Advanced Robotics 22 (2008) 829–849
`
`www.brill.nl/ar
`
`Full paper
`
`Biomimetic Tactile Sensor Array
`
`Nicholas Wettels a,∗
`, Veronica J. Santos a, Roland S. Johansson b and Gerald E. Loeb a
`a A. E. Mann Institute and Biomedical Engineering Department, University of Southern California,
`1042 Downey Way DRB-101, Los Angeles, CA 90089, USA
`b Physiology Section, Department of Integrative Medical Biology, Umeå University,
`901 87 Umeå, Sweden
`
`Received 3 August 2007; accepted 28 September 2007
`
`Abstract
`The performance of robotic and prosthetic hands in unstructured environments is severely limited by their
`having little or no tactile information compared to the rich tactile feedback of the human hand. We are
`developing a novel, robust tactile sensor array that mimics the mechanical properties and distributed touch
`receptors of the human fingertip. It consists of a rigid core surrounded by a weakly conductive fluid con-
`tained within an elastomeric skin. The sensor uses the deformable properties of the finger pad as part of
`the transduction process. Multiple electrodes are mounted on the surface of the rigid core and connected
`to impedance-measuring circuitry safely embedded within the core. External forces deform the fluid path
`around the electrodes, resulting in a distributed pattern of impedance changes containing information about
`those forces and the objects that applied them. Here we describe means to optimize the dynamic range of
`individual electrode sensors by texturing the inner surface of the silicone skin. Forces ranging from 0.1 to
`30 N produced impedances ranging from 5 to 1000 k . Spatial resolution (below 2 mm) and frequency
`response (above 50 Hz) appeared to be limited only by the viscoelastic properties of the silicone elastomeric
`skin.
` Koninklijke Brill NV, Leiden and The Robotics Society of Japan, 2008
`
`Keywords
`Biomimetic, electrode impedance, pressure sensor, haptics, tactile sensor
`
`1. Introduction
`
`Currently, robotic manipulanda excel in structured environments built around the
`robot, as in the automotive industry. The performance of robotic and prosthetic
`hands in unstructured environments, however, is severely limited by their having
`little or no tactile information compared to the rich tactile feedback of the human
`hand. Advancements in sensor hardware will undoubtedly cascade into equally
`
`* To whom correspondence should be addressed. E-mail: nick.wettels@gmail.com
`
` Koninklijke Brill NV, Leiden and The Robotics Society of Japan, 2008
`
`DOI:10 1163/156855308X314533
`
`CARDWARE EXHIBIT 2028, Page 1 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`830
`
`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`important advancements in controller design and grasp planning algorithms. Nu-
`merous fields of research would benefit from such advances: prosthetics [1], an-
`thropomorphic robotics [2], tele-operated and autonomous robotics [3], robotic and
`telesurgery [4, 5], telediagnostics and palpation [6].
`A wide variety of technologies have been applied to solve the tactile sensing
`problem in robotics and medicine [7]. Transduction mechanisms such as optics, ca-
`pacitance, piezoresistance, ultrasound, conductive polymers, etc., have all yielded
`viable solutions, but only for limited environments or applications. For example,
`most MEMS sensors provide good resolution and sensitivity, but lack the robustness
`for many applications outside the laboratory [8–10]. Beebe proposed a piezoresi-
`tive silicon-based MEMS sensor with a high tensile strength, but the sensor was
`limited by hysteresis and the inability to sense shear forces [11], which are just as
`important for grip control as normal forces. Conductive particles [12] suspended in
`elastomers can result in elastic materials whose resistivity changes with deforma-
`tion. A recent enhancement of such materials called quantum tunneling composites
`greatly increases sensitivity and dynamic range, but at the expense of mechanical
`hysteresis and simultaneous sensitivity to temperature and absorption of gases [12].
`Hysteresis alone may preclude the development of real-time grip control algorithms
`with short latencies similar to those observed in humans.
`The curved, deformable nature of biological fingertips provides mechanical fea-
`tures that are important for the manipulation of the wide variety of objects encoun-
`tered naturally. Multi-axis force sensing arrays have been fabricated using MEMS,
`but they are not suitable for mounting on such surfaces or for use in environments
`that include heavy loads, dust, fluids, sharp edges and wide temperature swings
`[9, 10]. If protective, skin-like elastic coverings are placed on top of sensor ar-
`rays, they desensitize the sensors and function as low-pass temporal and spatial
`filters with respect to incident stimuli [13]. Therefore, we considered it beneficial
`to make the cosmesis (skin) part of the transduction process rather than fighting it
`after the fact. This led to the approach of using a fluid in an elastomer as the trans-
`duction mechanism. Recently, flexible sensors have been developed to allow the
`mounting of the sensor on curved surfaces [14, 15]. However, these sensors can-
`not detect shear forces and their delicate electronic components are vulnerable to
`damage when mounted directly on gripping surfaces.
`Helsel et al. described an impedance-based sensor using a planar grid of gold–
`chromium electrodes, ethylene glycol-based conductive fluid and latex skin [16].
`Another fluid-filled sensor measured the impedance between electrodes on the inner
`surface of the elastomeric skin and on the surface of the core [17, 18]. These sensors
`were designed to detect deflection of the skin, but would saturate or otherwise fail at
`forces that caused contact with the core. Fingertips that employ electrorheological
`fluids for dynamic shape control use plates to apply high voltage fields. These plates
`can be used for capacitive sensing, but practical arrays have encountered problems
`including arcing [19, 20].
`
`CARDWARE EXHIBIT 2028, Page 2 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`831
`
`Here we describe a biomimetic tactile sensor that is sensitive to the wide range
`of normal and shear forces encountered in robotic and prosthetic applications. It is
`intrinsically simple, robust, and easy to manufacture and repair. This report focuses
`on a novel approach to extending the dynamic range by texturing the inner surface
`of the elastomeric skin. Preliminary results indicate promising spatial and temporal
`resolution.
`
`2. Methods
`
`2.1. Design Concept
`
`Our tactile sensor, modeled after the human digit, consists of a rigid central core sur-
`rounded by a weakly conductive fluid contained within a silicone elastomeric ‘skin’
`(Fig. 1) [21]. As with biological fingertips, our design incorporates the low-pass fil-
`ter effects of cosmetic, protective skin and fluid into the transduction process. The
`skin is resistant to wear, and possesses texture and tackiness similar to the properties
`that facilitate grip by biological fingertips. Electrodes are distributed along the sur-
`face of the rigid core and all sensitive components are safely embedded within the
`core. By applying an alternating current to each contact, one can measure the im-
`pedance of each volumetric flow path from a given contact to a reference electrode.
`External forces deform the fluid path around the electrodes, resulting in a distrib-
`uted pattern of impedance changes containing information about force magnitude,
`direction, point of contact and object shape.
`
`2.2. Prototype Fabrication and Theory of Transduction
`
`Fabrication and results from an early prototype are described in a previous confer-
`ence paper [22]. To create the rigid core of the array described here, we machined
`jeweler’s wax to create a negative mold of a shape similar to the distal phalanx
`of a human finger, with its tapered body and flat gripping surface. Individual gold
`contacts were formed by melting the end of a 5-µm Parylene-C insulated, 0.25-mm
`diameter gold wire into a ball and swaging to a 2.3-mm diameter, 0.5-mm thick
`disk. The contacts were tacked to the inside of the negative mold and the wire leads
`were soldered to a multi-pin electrical connector. A capillary tube was affixed in
`the mold for later use to inflate the fingertip with fluid. The mold was filled with
`liquid dental acrylic (Hygenic; Perm reline/repair resin) and cured to form a rigid
`finger core with electrodes on its surface. The sensor has no moving parts, and del-
`icate electrical components and wiring are protected in the high-strength rigid core
`(compressive strengths 10–100 MPa and tensile strength 1–10 MPa [23]).
`The choice of silicone elastomer for the skin depends on achieving mechanical
`properties and cosmetic appearance similar to normal skin. Candidate outer materi-
`als include Dragon Skin (Smooth-On Inc.; Shore A hardness of 10 and tear strength
`of 102 lb/in) and VST-30 (Factor II Shore A hardness of 23 and tear strength of
`100 lb/in). A higher durometer inner coating can be used to optimize mechanical
`properties, while a softer, outer coating will provide a cosmetic appearance and
`
`CARDWARE EXHIBIT 2028, Page 3 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`832
`
`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`Figure 1. (A) Mechanical drawing of the biomimetic tactile sensor showing a rigid core shaped like
`the distal phalanx with an internal, sealed compartment for electronics connected to sensing elec-
`trodes in contact with a weakly conductive fluid under a viscoelastic skin. (B) Mechanical drawing of
`the current core design with a sample electrode layout. (C) Current acrylic prototype core with skin
`and fingernail removed. The thin gold ground electrode, circular gold working electrodes and blue
`thermistor are visible on the surface of the core.
`
`feel. The coating of the elastomeric skin onto the rigid core enables easy repair
`of the most vulnerable part of any finger. The skin is easily replaced without any
`effects on the sensing electrodes or their supporting electronic circuitry within the
`rigid core. A plastic fingernail was used to anchor the skin to the core on the dorsal
`side of the fingertip. This feature contributes to the sensitivity of the sensing array
`to tangential as well as normal forces [22], but it is not studied in this report. Salt
`water (described later) was introduced through the fill-tube at the proximal end of
`the fingertip to inflate the cured silicone skin away from the core. In the future, me-
`chanical fixation features can be incorporated into the mold to facilitate mounting
`of the fingertip to a mechatronic hand or mechanical test setup
`
`CARDWARE EXHIBIT 2028, Page 4 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`833
`
`Table 1.
`Effects of texturing of the internal surface of the skin on
`sensor behavior were studied for multiple combinations
`of silicone stiffness and sandpaper grit size
`
`Silicone durometer
`(Shore A hardness)
`
`10 (compliant)
`45
`60 (stiff)
`
`Sandpaper grit size
`
`60 (rough)
`
`(cid:1)
`
`100 (fine)
`(cid:1)
`(cid:1)
`(cid:1)
`
`Saturation of an electrode occurs when it becomes completely occluded, the
`volumetric flow path cannot become any more resistive and the impedance measure-
`ments do not increase. Texturing of the internal surface of the skin allows for small
`conductive pathways to exist even after the internal surface of the skin has been
`compressed against the electrode and increases the upper limit on the force-sensing
`range of the sensor. Thus, for preliminary characterization testing, we varied the
`stiffness and texture of the internal surface of the skin to investigate their effects
`on sensing behavior (Table 1). Strips of sandpaper of various grit sizes were used
`as the negative molds for test strips (9 × 20 mm) of textured silicone. These strips
`served as the internal surface of the skin while a softer, untextured Dragon Skin
`silicone elastomer served as the outer surface.
`The sensitivity of the device depends complexly on the conductivity of the fluid,
`the viscoelastic properties of the combined system of skin and pressurized fluid,
`volume and pressure of fluid, and the material and geometry of the electrode con-
`tacts. It is generally desirable for the fluid to have a low viscosity to minimize
`damping and hysteresis, and a high resistivity so that the measured impedance of
`the series circuit (electrodes plus fluid) is dominated by the fluid resistance rather
`than the capacitive reactance of the metal–electrolyte interfaces. In the tests de-
`scribed here, the fluid was water with a low concentration of NaCl (0.75 g/l; 1/12th
`the concentration of physiological saline).
`
`2.3. Signal Conditioning
`
`It is desirable to energize the electrode system in such a way that the voltages devel-
`oped across the metal–electrolyte interfaces are sufficiently low to avoid Faradaic
`current flow, which would corrode the metal contacts and cause electrolysis of the
`electrolytic fluid [24]. This can be accomplished by applying a small, alternating
`current and measuring the resulting voltage across the electrodes. The capaci-
`tance of one working electrode and ground electrode were found empirically to
`be 84.25 nF. This is consistent with capacitance per unit area values found in the
`literature [25, 26]. For a given electrode this capacitance has a reactance of 1.89 k
`
`CARDWARE EXHIBIT 2028, Page 5 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`834
`
`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`Figure 2. The signal-conditioning circuit used to collect characterization data from a single electrode.
`
`at 1 kHz, which would be a small component of the dynamic range of the total
`impedance of the sensor (5–1000 k ).
`The signal-conditioning circuit (Fig. 2) was driven with a 1-kHz 10-V AC
`sinewave in series with R1 to provide a constant current source. Based on the dy-
`namic range of sensor impedances to be measured, a 1-M resistor was chosen
`for R1. The voltage across the electrodes was buffered by a unity gain operational
`amplifier (National Semiconductor; LMC 6482) and frequencies higher than the
`1-kHz test signal were filtered by R2C2. Capacitor C1 blocked any DC bias from
`the sensor electrodes to prevent corrosion and electrolysis. Vout was digitized at
`50 kilosamples/s (National Instruments; SCB100). The double-layer capacitance
`between the electrolyte solution and the metal of the working electrode and ground
`electrode and fluid has been represented as a single capacitor in series with the vari-
`able resistor representing the deformable, volume-conductive fluid path between
`them. Peak-to-peak voltage was determined by a custom LabVIEW (National In-
`struments) program that translated this output into a calibrated impedance value for
`the graphs plotted herein.
`
`2.4. Static Characterization of a Single Electrode
`To characterize the static behavior of a single electrode for the prototype shown in
`Fig. 1C, we applied normal forces to the electrode of interest and the surrounding
`area. A linear drive (Nippon Pulse America; PFL35T-48Q4C(120) stepper motor
`and NPAD10BF chopper drive) was used to advance interchangeable probes: 2 mm
`diameter (1 mm radius of curvature) and 20 mm diameter (11 mm radius of curva-
`ture), and a large flat plate (2.6 × 3.3 cm). The probes were designed to have radii
`of curvature much smaller than, approximately equal to and much larger than the
`curvature of the sensor core, respectively Deflection was recorded from the point
`
`CARDWARE EXHIBIT 2028, Page 6 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`835
`
`Figure 3. Experimental set-up for static characterization of a single electrode.
`
`of the initial probe contact with the skin. Normal force was measured using a six-
`axis forceplate (Advanced Mechanical Technology; HE6X6-16) positioned below
`the vise holding the sensor (Fig. 3).
`To characterize the spatial resolution of an electrode, the sensor was systemat-
`ically probed in a 4 × 5 grid in 2 mm increments centered on the electrode. All
`impedance measurements were obtained at 4.5 mm indentation.
`
`3. Results
`
`3.1. Static Characterization of a Single Electrode
`
`3.1.1. Deflection Applied Directly Above the Electrode
`As the fingertip was compressed, there was a monotonic but nonlinear increase in
`electrode impedance over a range of 100–1000 times the starting value (Fig. 4). The
`slopes of the curves depended complexly on probe curvature, as discussed later. The
`reaction force of the fingertip also rose monotonically and nonlinearly as the fluid
`was displaced from under the skin (region A in Fig. 4) and the skin was compressed
`against the rigid core (regions B and C). Region C is broken down into two areas:
`C(1) is the region where the textured surface contacts the core; C(2) is where sensor
`begins to saturate, as maximal occlusion of the electrode occurs. The sigmoidal
`shape of these logarithmic curves is reminiscent of many biological transducers,
`which generally need to optimize local sensitivity over a wide dynamic range of
`possible inputs [28].
`Figure 5A–C shows how the sensor responded to variations in texture when
`indented with the 20-mm probe. For Fig. 5A, in the case of no texturing (10 duro-
`meter, no texture curve) the impedance rose very quickly to infinity once forces
`
`CARDWARE EXHIBIT 2028, Page 7 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`836
`
`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`Figure 4. (A) Impedance (log scale) as a function of force (log scale) applied normal to the electrode
`surface. Textured silicone: 60 durometer, 100 grit size. (B) Graphic correlating curve shapes to probe
`indentation (20-mm probe shown).
`
`exceeded 1 N. To increase this dynamic range and decrease the impedance/force
`slope, internal texturing of the device was necessary. The stiffer the silicone tex-
`ture used, the higher the forces required to maximally occlude a given electrode.
`Another factor affecting the contact point and the beginning of region C was the
`thickness of the textured strips. The 60 durometer strip contacted first and the
`10 durometer last. Their thicknesses were 2, 1.6 and 1.2 mm for the 60, 45 and
`10 durometer strips, respectively.
`Figure 5B shows the effect of texture depth on the operating range of the sen-
`sor. The deeper the texture (grit size), the more force was required to occlude the
`electrode. Figure 5C illustrates the effect of sensor volume (and subsequently hy-
`drostatic pressure) on the force–impedance curves. When the sensor was drained
`to 1 ml, the resting impedance increased because the finger was deflated, bringing
`the skin closer to the surface of the electrode. This resulted in (i) increased sensi-
`tivity and (ii) a flatter sigmoid — because the fluid was drained, it took less force to
`achieve a given impedance level. The two curves converged during contact of the
`textured silicone to the core.
`
`3.1.2. Temporal Responses
`All of the responses illustrated in Figs 4 and 5 were actually collected in both di-
`rections of loading and unloading, but only rising forces are illustrated. Generally,
`there was little or no hysteresis over most of the operating range. Figure 6 illus-
`trates the worst case hysteresis, which was seen only at the higher force levels. The
`time between data points was about 2 s, so the rates of loading and unloading were
`unphysiologically slow.
`
`CARDWARE EXHIBIT 2028, Page 8 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`837
`
`(A)
`
`(B)
`
`Figure 5. (A) Response of sensor to texture stiffness. (B) Response to texture depth (grit size). (C) Re-
`sponse to fluid volume. Panels (A)–(C) show response to the 20-mm probe.
`
`To estimate the true frequency response of the sensors, we applied a vertically
`oscillating flat probe to the fingertip while recording the vertical force and the en-
`velope of the sinusoidal Vout (Fig. 7). The temporal details of the mechanical input
`were well represented over the range of frequencies and loads tested informally
`(approximately 5–15 Hz at 3–5 N in Fig. 7A and 45 Hz at 10–15 N in Fig. 7B).
`3.1.3. Deflection Applied Above and Around the Electrode
`The sensor was systematically probed in a 4 × 5 grid; impedance measurements
`about the electrode of interest, located at (x, y) = (0, 0), demonstrate a spatial res-
`olution below 2 mm (Fig. 8). The spatial resolution is, of course, dependent upon
`contact location. If a deflection is made far away from any electrode, then resolution
`will diminish.
`
`CARDWARE EXHIBIT 2028, Page 9 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`838
`
`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`(C)
`
`Figure 5. (Continued.)
`
`Figure 6. Loading/unloading curve (20-mm probe).
`
`4. Discussion
`
`The response of a single electrode varied significantly with the radius of curva-
`ture and contact surface area of the probe, but generally consisted of three distinct
`phases of sensing behavior (Fig. 4). During the initial deflection of the skin, the
`fluid is displaced from between the skin and the core. As the skin is deformed
`above a given working electrode it constricts the flowpath for ions between the
`electrode and ground. The impedance rises non-linearly and is related to the overall
`constriction of the ionic flowpath In region A, there is relatively little deformation
`
`CARDWARE EXHIBIT 2028, Page 10 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`839
`
`Figure 7. (A) 5–15 Hz response of sensor. (B) 45 Hz response of sensor. The force waveform is unidi-
`rectional. The impedance waveform is bidirectional because it is capturing the envelope of a sinusoidal
`waveform.
`
`Figure 8. Impedance as a function of center of pressure with respect to the location of a working
`electrode at (x, y) = (0, 0) for 4.5 mm vertical deflections with a 2-mm diameter probe.
`
`In region B, the skin approaches the electrode and begins to contact it, cutting off
`the fluid access to the electrode and increasing the resistive component of the im-
`pedance. In region C, the textured skin has contacted the surface of the electrode
`and as force increases, the fluid channels through the grooves of the textured surface
`are gradually compressed (see Fig. 9).
`More force is required to compress the channels in the stiffer, high-durometer
`rubber. This increases the dynamic range of forces before the channels are com-
`
`CARDWARE EXHIBIT 2028, Page 11 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`840
`
`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`Figure 9. (A) The textured rubber is shown in an uncompressed state. (B) An applied force compresses
`the textured rubber and narrows the flowpath of the fluid to the electrode surface. The impedance
`measured in condition ‘(A)’ will be lower than that in condition ‘(B)’.
`
`pletely closed and the measured impedance saturates (Fig. 5A). As the grit size
`of the texturing is made rougher, this provides deeper channels that require more
`force to compress (Fig. 5B), also extending the force range. However, if the size
`of the irregular texturing is made too large in contrast to the size of the electrode,
`the sensor’s behavior will become dependent upon random interactions between the
`texture features and the electrode surface.
`The shape of the remaining conductive path also depends on the shape of the
`probe. With the 20-mm probe (similar in curvature to the sensor core), there was
`a gradual rise in impedance and force (region A20) for the first 0.2 N of force until
`the skin contacted the electrode. Then the impedance and force rose rapidly (re-
`gion B20) as the textured inner surface of the skin began to seal to the electrode.
`Once full contact was made (region C20(1)) the sensor exhibited a more linear re-
`sponse until response saturated as maximal occlusion was reached in region C20(2).
`With the 2-mm probe, the gradual rise in impedance took longer (A2) because the
`small probe deformed the inner contour of the skin, preventing even contact over the
`entire electrode surface until it reached a larger deflection (B2). For a small probe
`similar in size to the diameter of the electrode itself, even a slightly off-center po-
`sition would affect the transition from region A to B, which would account for the
`apparent rightward shift of curve region B2 versus B20 in Fig. 4A. With the flat
`probe, the impedance did not increase rapidly until force reached 7 N because the
`force was distributed over a large surface of skin (Af). The flat probe pushed the
`fluid downward as well as outward, causing a change in the shape of the volumetric
`flow path, but less change in the impedance than the 20-mm probe. Eventually, the
`skin contacted the core, producing a steeper increase in both impedance and force
`(Bf) and eventual saturation (Cf).
`It would appear that there is an ambiguity between impedance and deflection (or
`force) and the shape of the contacting object. If the shape of the object is not known
`a priori, how is one to determine deflection or force from impedance? This problem
`is solved through active exploration of the object to obtain other features such as
`object shape — which is exactly what the human haptic system does. The amount
`and timing of the deflection is caused by and known to the operator exploring the
`
`CARDWARE EXHIBIT 2028, Page 12 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`841
`
`object. Thus, the shape can be extracted from the time course of the impedance
`measured and comparison of the sensor response against the expected values based
`on a priori experience and expectation, as discussed later.
`Figure 5A–B describes how modulating the stiffness and depth (grit size) of
`the texturing affected the force-impedance relationship. Depending upon the force
`range desired, the internal texturing could be designed accordingly. Figure 5C
`demonstrates the effect of lower fluid volume and hydrostatic pressure. This re-
`sulted in increased sensitivity — the 1-ml curve had a larger impedance initially
`(as expected from the thinner fluid path), but a more gradual increase over the low
`range of applied forces. Other factors that would influence this include thickness
`and compliance of the skin and hydrostatic pressure of the fluid, which remain to
`be examined systematically.
`Figure 6 shows that above 3 N and for unphysiologically slowly changing forces
`(more than 2 between measurement points) there was noticeable hysteresis. We
`speculated that the divergence at high force levels may have been caused by static
`friction between the textured silicone and the lightly abraded surfaces of the elec-
`trode contacts and acrylic core, in which case it might be expected to be highly
`dependent on rate of loading and unloading. This was confirmed by the fast tempo-
`ral responses to a vibrating flat surface (Fig. 7). We have yet to undertake systematic
`dynamic testing, but gross hysteresis and drift are not expected in our sensor be-
`cause the silicone elastomer is a spring-like material that does not exhibit creep and
`the fluid has low viscosity.
`The spatial resolution displayed in Fig. 8 presumably depends on several me-
`chanical factors. These include the dimensions of the probe (2 mm diameter), the
`electrode (2.3 mm diameter) and the skin (around 4 mm thickness), whose inter-
`relationships remain to be determined systematically.
`
`4.1. Feature Extraction
`
`The positioning of the electrodes with respect to the contours of the core and over-
`lying fluid and skin causes distinct patterns of change in impedance as the sensor
`contacts objects and surfaces with various force vectors. For example, those elec-
`trodes located near the nailbed will detect large increases in impedance when shear
`forces are directed away from them, pulling the skin close to these electrodes. The
`following sections summarize phenomena whose characteristic features can be de-
`tected by the sensor, in principle.
`
`4.1.1. Contact Force Magnitude
`As the total force increases on the central area of the sensor, the thickness of the
`fluid layer overlying the electrodes in the compressed region decreases, causing the
`impedance measurements from the electrodes in that region to increase. The contri-
`butions of all such impedance increases are related to the total force of contact.
`
`CARDWARE EXHIBIT 2028, Page 13 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`842
`
`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`4.1.2. Contact Force Direction
`In most object manipulation tasks, the force vector imposed by the contacted object
`is not perfectly normal to the sensor surface. In biological skin, shear force compo-
`nents change the stress and strain distributions within the fingertip that are sensed
`by receptors located within dermal and subdermal tissues, but also by the distribu-
`tion of pressure around the perimeter of the finger pad, particularly where the skin
`is anchored by the nailbed [29]. This deviation of the force vector from normal is
`generally associated with a tendency of the grasped object to slip or rotate [29–34].
`In our tactile sensing fingertip, shear forces will cause the skin to slide and pull
`tight over electrodes located where the skin meets the fingernail. As a result, these
`electrodes will detect large increases in impedance when shear forces are directed
`away from them.
`
`4.1.3. Location of Force Centroid
`The impedance increases associated with the contact force measurement previously
`described can be related to electrode location to estimate the location of the center
`of force on the skin surface.
`
`4.1.4. Object Shape
`The relationship between impedance measured at a single electrode and the force
`applied to an object depends on the shape of the object (Fig. 4), and also on the loca-
`tion of the object with respect to that electrode (Fig. 8) and the radius of curvature of
`the rigid core on which the electrode is mounted. The impedances measured across
`an array of such electrodes can, in principle, be used to interpret object shape. For
`example, a small or narrow object will produce a local deformation of the skin that
`will cause large changes of impedance for only one or a few electrodes close to the
`point of contact. A sharp edge will cause an abrupt boundary between electrodes
`with high impedance and those with low impedance as a result of displacement of
`fluid and bulging of the skin. The optimal relationships among the curvature of the
`core, the number and spacing of electrodes, and the thickness and inflation of the
`skin remain to be determined. Features that are larger than the array will require
`additional haptic exploration to identify them.
`
`4.1.5. Object hardness/softness
`As previously stated, the sensor can detect mechanical transients associated with
`making contact with an object. If the sensor is affixed to a mechatronic finger mov-
`ing at a known velocity, the rate of deformation increase can be used to indicate
`the level of hardness or softness of a contacted object. A hard object will cause
`a rapid increase in deformation (and voltage response) for a given finger velocity
`when compared to a soft object.
`
`4.1.6. Contact transients and vibration
`The impedance of the electrodes in the biomimetic fingertip will undergo only very
`small changes when lightly loaded, but it may be possible to detect such changes
`by means of their synchronous phasing across the entire array of electrodes [32]
`
`CARDWARE EXHIBIT 2028, Page 14 of 21
`SAMSUNG V. CARDWARE PGR2023-00013
`
`

`

`N. Wettels et al. / Advanced Robotics 22 (2008) 829–849
`
`843
`
`Signal-averaging techniques can be applied to enhance the detection of the cor-
`related component of weak and noisy signals from an array of sensors. Direct
`measurement of hydrostatic pressure may be preferable, however (see Section 4.3).
`
`4.2. Further Feature Characterization
`
`We have demonstrated basic characterization of the sensor with respect to force,
`deflection and impedance. More thorough testing is required, including thermal
`response, destructive analysis, etc. Some contact features such as force centroids
`could be extracted analytically using a reasonable mathematical model. In our ar-
`ray of tactile sensors, force magnitude and point of application interact with each
`other. For example, a force vector applied close to the nailbed will create a different
`amount of net impedance change than if the same force vector were applied to the
`fingertip. Thus, impedance cannot be used as a measure of the applied force unless
`one accounts for the point of application.
`Our sensor array has properties similar to the biological fingertip, however, so
`it will likely require non-analytical signal processing methods similar to those
`employed by the biological nervous sys