Development of a Tactile Sensor Based on Biologically Inspired

Edge Encoding
Craig Chorley, Chris Melhuish and Tony Pipe Jonathan Rossiter
Bristol Robotics Laboratory Department of Engineering Mathematics
University of the West of England University of Bristol
Bristol, UK Bristol, UK
[email protected]

Abstract— Here we present a novel biologically inspired Epidermis

tactile sensor based on the structure of the human fingertip.
Unlike previous biologically inspired sensors, it is based on Meissner’s
new theories of the functional morphology of the fingertip Corpuscle
skin features and the Meissner’s Corpuscles mechanoreceptors
pervading them, particularly in the encoding of tactile edge
Dermal
information. Through mimicking the layered macro-structure Papillae
of fingertip skin the sensor is highly conformable and very
sensitive, as well as a strong and practical gripping tool. The Merkle Cell
tactile sensor is composed of a thin flexible rubber skin with Complex

structural details emulating those of the glabrous epidermis.

This encases a clear, highly compliant polymer melt blend with
similar mechanical properties to the dermis and subcutaneous
fat. A camera is used to track markers on the internal structural Ruffini Ending
Pacinian Corpuscle
details of the rubber skin enabling remote, detailed, direct
and sensitive detection of surface deflections. Initial results
presented here show the design to be a very capable, highly
sensitive sensor as well as a very practical, affordable and
scalable robotic fingertip.
Fig. 1. Cross Section Illustraion of Glabrous Skin
I. INTRODUCTION
Investigations into the functional morphology of the hu-
man fingertip have produced some significant results for
robotics research in tactile sensor design.
forces deforming the skin, which is focused like a lens
A. Biologically Inspired Sensors
down through the papillae structure to these mechanoreceptor
Tactile sensing in robotics has traditionally taken matching nerve endings. This functional morphology has inspired a
the human example as its ultimate goal[1]. It is only natural few tactile sensor designs based around this lensing effect
then to look to biology for inspiration. Previous biologically as a way of sensing applied surface force distribution, see
inspired sensors have tried to gain advantages from mimick- Mukai[4] for an example.
ing the physiology of the skin, or the function of the nerves,
to greater or lesser extents. It is widely accepted that the Meissner’s Corpuscles are a
It is taken that, of the four tactile mechanoreceptor nerve rapidly adapting sensory nerve ending and have a reported
endings pervading the glabrous (hairless) skin of the palms sensitivity to vibrations between 10 - 60Hz. This, together
of our hands and soles of our feet, the Pacinian Corpuscle, with their proximity to the skin surface, has led to the general
the deepest and largest tactile nerve ending, is the most perception that they are purely a receptor of light touch
attributed to the sensation of vibration. This has inspired vibration. However, a recent hypothesis of the workings of
many tactile sensor designs exploiting the vibrations created the Meissner’s Corpuscle by Iwanaga[5] has shed light on the
from explored textures, frictional properties of surfaces and possible directional sensitivity of these Mechanoreceptors.
in the detection of incipient slip of grasped objects[2]. Kuroki et al. built on Iwanaga’s work and produced theories
Merkel Cells, however, can be seen to be sited at very that the structure of the Meissner’s Corpuscles, together
specific locations on the tips of the Dermal Papillae, see with their location between the papillae, encode normal
Fig.1, on the epidermal/dermal boundary. Theories of the strain forces by detecting normal horizontal strain[6]. The
functional reasons for these Dermal Papillae in relation to the work presented here attempts to illustrate that this functional
workings of the Merkel Cells have been derived from Finite morphology of the Meissner’s Corpuscles, together with the
Element studies of simulated fingertips[3]. It is believed Dermal Papillae, results in the encoding of edge information
that Merkel Cells respond to strain energy density due to of a tactilely sensed scene.

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on October 21,2024 at 08:40:26 UTC from IEEE Xplore. Restrictions apply.
B. Optical Based Tactile Sensors when the sensing elements themselves have a detrimental
There have been a number of vision based tactile sensor effect on the desirable mechanical properties of the arti-
designs. Some of these make use of a ridged transparent ficial skin. Copper wires, flexible plastic PCBs or other
internal layer acting as an optical waveguide. An external such electronics introduced into the contact skin area all
rubber cover skin is then deformed, with tactile pressure, contribute to affecting the mechanical properties and the
against the waveguide causing Frustrated Total Internal Re- sensor’s achievable conformability. This attenuates the level
flection, detectable by a camera. Early examples, such as of sensed detail possible and places these vital components
Jiar et al.[7] and Begel[8], were so large the they could not where they are most likely to be damaged over time and use,
realistically be used as a practical end-effector. This design often to result in a costly repair.
concept, however, was further developed by Ohka et al. [9]
D. Edge Sensitivity
to be of a more realistic size and shape. These designs
can be seen to relate, to a greater or lesser extent, to the Our sense of touch is particularly sensitive to edges. By
working structure of the Merkel Cell nerve endings. Some edges we mean the sharp gradient changes in the surfaces that
even make use of Dermal Papillae-like features on the rubber we touch. These could be local raised points, as exploited
skin to cause detectable deformations of their tips against the in Braille, local features and textures or the actual edges
waveguide, creating a detectable pressure distribution map of and geometry of an object. This keen tactile sensitivity we
points. have to edges highlights the importance edges have in all
The other main type of vision based tactile sensor use a our fine control and understanding of what we touch; in
camera to scan a pattern on an elastic surface that is then exploring, manipulating, learning and recognising objects.
deformed with applied pressure. One of the most notable of Imagine trying to blindly place a square peg into the square
this configuration is the GelForce sensor by the Tachi Lab, hole without being able to the feel the shape of the peg,
Tokyo[10]. Their design consists of two layers of dots, one the shape of the hole or the ordination of the peg you are
red and one blue, within a clear silicon rubber. The relative holding. Without edge information, it is hard even to imagine
movement of these layered dots, as the rubber is deformed, is being able to find the locations of holes themselves. Rarely
captured by the camera and force magnitude as well as force have tactile sensor designs considered sensitivity to edges as
vector information is calculated through elastic theory. They an important factor, though it is so prominent for us.
have also recently managed to scale this design down to a II. SENSOR OPERATION
fingertip size[11]. These configurations, however, suffer from
what is known as the Inverse Tactile Transduction problem. As highlighted by Kuroki et al.[6], the Meissner’s Corpus-
This arises due to the limited correspondence between the cles, positioned within the troughs of the Derman Papillae,
stresses deforming and displacing the markings deep within Fig.2, have been recently theorised to be sensitive to lateral
an elastic material, to those that are applied to the surface. strain. Kuroki et al. wrote of this lateral strain (horizontal
The patterns of sensory data obtained by the camera may normal strain) as a result of a normal force causing move-
be a result of a combination of patterns on the surface. This ment of the papillae. However, when looking at the fingertip
is particularly relevant in fine detail discrimination as the skin as a whole, this papillae movement occurs only where
elastic material acts as a low pass filter with only relatively there is a change in gradient of the deflection of the skin
large scale spatial patterns and forces transmitted through the surface. This movement and therefore the lateral strain the
elastic layer, attenuating any fine detail. Meissner’s Corpuscles detect, is actually proportional to the
change in gradient of the deflection, not just simply a normal
C. Skin Layers and Conformability force. This results in an output directly proportional to the
The structure of our fingertips plays important roles in our edges in a sensed scene.
sense of touch and our ability to grip, manipulate and explore Consider a finger pressing on a raised point on a flat
objects. Our specialised fingertip skin is both strong and surface, as depicted in Fig.3. The sharp gradient changes of
highly compliant. This high compliance of the skin enables the raise dot produces horizontal spreading and narrowing
us to sense fine detail and light touch [12], while the layered of the gaps between the papillae of the locally deformed
structure of the thin, strong, inelastic but flexible epidermis, skin. Where as the normal force applied by the flat surface
surrounding this very soft core, gives the glabrous skin its is does not. This spreading and narrowing encodes the
strength to grip, lift, explore and manipulate objects and aids relative gradient changes that the raised dot produces. The
the direct transmission of the tactile information to our nerve flexibility and inelastic nature of the epidermis ensures the
endings. direct and sensitive transmission of this surface detail to the
The importance of conformability and strength of the skin deflection of the skin and the movement of the papillae.
structure is often overlooked in sensor design. There is a If it were an elastic deformation, the detail would be lost
preference to design for elastic deformation rather than in- in the stretching of the skin, creating the Inverse Tactile
elastic conformability, or the use of a homogeneous semi-soft Transduction problem. The papillae also act as a stimulus
rubber as a compromise between compliance and strength, amplifier as the local skin deflection is exaggerated by the
that neither has high strength nor high conformability[13]. papillae movement. A very similar affect has been exploited
Another of the problems found in tactile sensor design occurs in the ’Tactile Contact Lens’, where the movement of an

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on October 21,2024 at 08:40:26 UTC from IEEE Xplore. Restrictions apply.
 CCD Camera

Camera
Spacer
Sensor
Skin Acrylic
IR LED Window
Illumination

Clear Gelatinous Nodule Markers

Polymer Blend
Fig. 2. Histology of Meissner’s Corpuscles within Glabrous Skin

Lateral Strain
Gradient Change
Fig. 4. Sensor Design

and robustness. Finally, an enclosed centre core, mimicking

the very high compliance of the dermis and subcutaneous
fat, would act to give the sensor its overall conformability,
Normal Force by forcing the flexible skin to match the surface detail when
pressed against it. This results in a very capable and reliable
sensor design with potential for economical manufacture.
III. SENSOR DESIGN
The sensor design consists of a 40mm diameter hemi-
sphere of cast 0.3mm thick, black, Shore Hardness A 50
urethane skin. The thickness and hardness of the rubber
Fig. 3. Demostration of Papillae Edge Encoding ensures a very flexible but strong, relatively inelastic skin.
The inside layer of the skin has an array of artificial papillae
features that are 1.5mm high, 1mm wide and spaced 1.5mm
array of pins placed against a person’s fingertip act to encode apart, with white markers on the tips.
and amplify surface gradients, see Kikuuwe et al.[14]. The The hemisphere is filled with a water clear, very highly
enclosed, highly compliant core of the finger pad ensures compliant, melt polymer blend and is sealed with a clear
that the epidermis, pressed against the surface, conforms to acrylic disk window. An arrangement of four infrared LED,
the whole shape of the feature, enabling the maximum detail around the inside circumference of the disk, is used to
to be sensed. illuminate the markers. The skin is extended over a 35mm
A sensor could, therefore, be designed to mimic these long plastic tube that acts as a focal length spacer and a
papillae features on a strong, thin, inelastic, flexible rub- mount for the camera. A 640 x 480 30fps CCD camera is
ber skin, and be used to sense surface deflections, with mounted to capture the whole marker array. This set up gives,
a particular sensitivity to edges, by detecting the relative for this particular camera and lens, a focal distance of around
movement of these artificial papillae. One method might be 50mm for a 45mm wide field of view. For testing, the sensor
to integrate sensors into the skin that measure the relative was mounted vertically in a retort stand and lowered onto test
lateral strain or movement of the artificial papillae. This indenters placed on a balance. This enabled measurement of
would be very complex, very costly and adversely affect the applied force compared to the resultant skin deflections,
the skin’s mechanical properties, also reducing the sensitivity observed through the marker movement by the camera.
and detail of the skin defections, as well as raising issues of
reliability over time. With the use of markers on the ends of IV. R ESULTS
the papillae nodules, a camera can be used to track the entire A range of indenter shapes and sizes were used to observe
skin remotely. This would mean the sensing surface and the reaction of the sensor.
contact area of the skin is completely removed of wires or Fig.5 illustrates an example of the reaction of the sensor
electronics, maintaining the desirable mechanical properties to a single point indenter. On the left is the sensor at rest and

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on October 21,2024 at 08:40:26 UTC from IEEE Xplore. Restrictions apply.
 Fig. 5. Example of a 0.5N 5mm Diameter Point Force

Fig. 8. Annulus Indenter

the right as a result of a point, in this case, 5mm diameter
and at force of 0.5N.

Fig. 9. Flat with Square Edge

Fig.9 is from an object that had a flat top and a sharp

square edge. This example clearly shows the edge encoding
in action, contrasting the reaction to a sharp edge to that of
a flat surface.
Fig. 6. Velocity Vector Field - 5mm Diameter Point, 0.25N Force

The point images were analysed using a velocity vector

field to measure the relative pixel movement across the
marker array. Fig.6 shows an example of the velocity vector
field for a 5mm diameter point at 0.25N of force.

Fig. 10. Two Point Discrimination - 5mm Spacing, 0.5mm Diameter Point

Fig.10 shows an example of a two point discrimination

test. Here the 0.5mm diameter indenters were spaced 5mm
apart, producing a result clearly visible in the image. On the
human fingertip, we can discriminate between two points
at a minimum distance of 2-3mm. Considering the current
Fig. 7. Straight Edge relative size of the sensor, about four times that of a fingertip,
this is quite an impressive result. Smaller two point widths
Fig.7 is an example of the reaction of the sensor to a are harder to see in a photographed image, however the future
straight edge indenting the sensor. work on processing the image data will make a much more
Fig.8 illustrates the reaction of the sensor to a annulus accurate two point discrimination possible.
shape. The edge of the annulus is clearly visible, demon- Tests were then conducted to illustrate how sensitive the
strating the movement of the markers due to the changes in sensor could prove to be. A 0.5mm point indenter was
gradient across the shape. However, the circular shape is seen applied at a force of 0.05N and processed by tracking the
here as more of a pentagon pattern, due to the pentagonal centroids of the markers, show in Fig.11, and with a velocity
layout arrangement of the nodules on the hemisphere. vector field Fig.12. The resulting detected movement was

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on October 21,2024 at 08:40:26 UTC from IEEE Xplore. Restrictions apply.
 detect other mechanoreceptor related modes. Merkel Disc
related information could be sought by looking at the size
and shape changes of the markers, relating to pressure dis-
tribution. Overall movement would relate to shearing forces,
most attributed to Ruffini’s Endings.
Though the current results here are mostly images, they
show the sensor to be very sensitive, demonstrating a de-
tectable point force of 0.05N and showing promise of greater
sensitivity with future processing. The two point discrimina-
tion tests illustrated that a 5mm gap between points can be
seen in a static image, meaning a narrower point gap may
be detectable in the future. Such a sensor system could be
Fig. 11. Sensitivity Test - 0.5mm Point, 0.05N Force - Centroid Movement designed able to detect and determine one Braille character
from another.
The nature of the sensor output is inheritly dynamic. The
camera is there to detect the papillae movement and how the
papillae move in reaction to a stimulus, gives up more in-
formation about the sensed object. As mentioned previously,
Messiner’s Corpuscles are rapidly adapting mechanorecep-
tors, there purpose is to be sensitive to changes in mechanical
stimulus. As your finger is pressed or moved across an edge
feature, more of the fine detail is revealed. Through suitable
processing techniques focused on the dynamic nature of the
sensor data, it is believed that greater detail, sensitivity and
two point discrimination could be achieved.

VI. CONCLUSION AND FUTURE WORK

Fig. 12. Sensitivity Test - 0.5mm Point, 0.05N Force - Velocity Vector
Field
Here we have presented a novel tactile sensor for robotics.
Through this sensor design it has also been possible to give
clear in both cases. Further processing will increase this a practical demonstration of new theories for the functional
sensitivity as even a 0.03N applied force produces visually morphology of the structures within the glabrous skin of our
detectable movement of the markers. fingertips and show how these might be exploited to create
In shear, the array of marker moves strongly relative to tactile sensors that can encode edge information. Sensitivity
its direction, making shearing forces easily detectable also. to edges is a vital part of our ability to finely manipulate
As a whole, the biologically inspired layered design means objects as well as our ability to explore, investigate, learn
the sensor also acts very successfully as an end-effector and recognise them. A tactile sensor with this sensitivity to
and gripping surface. The thin, strong outer skin encasing a edges and the strength and conformability afforded to it by a
highly compliant core enables a large, conformable contact layered skin should open new directions for tactile research.
area for a given applied force, a strong grip and the resistance The results shown here demonstrate a clear ability to deliver
to shearing forces needed in lifting and manipulating objects. edge information of various shapes and sizes. The current
sensor design has shown a two point discrimination per-
V. DISCUSSION formance down to 5mm separation and sensitivity to forces
The results presented here clearly demonstrate the concept of 0.05N or lower. The physical design is relatively simple,
of edge encoding by the biologically inspired structures. This and could therefore be cheap, easy to produce and maintain,
is particularly evident in the case of the flat rectangular since there is no wires or electronics within the contact skin.
object with sharp square edges, Fig.9. Here it can be seen The design relies on a moulding process that can be easily
that the forces normal to the sensor, due to the flat top configurable and scalable to different sizes of sensor and
surface, are not weak, contrasted against the strong reaction papillae densities. This new sensor offers future robotics a
to its edges. If the functional morphology of the Meissner’s very capable, affordable, scalable and practical tactile sensor
Corpuscles is that they are sensitive to lateral strain; then this design. The next stage of work is to investigate various
illustrates that, together with the Dermal Papillae, they may different appropriate methods for processing the image data.
be working in this way to encode tactile edge information. The sensor response can then be characterised and calibrated.
This raises points about the source of our edge sensitivity and Further investigation into the dynamic characteristics will
shows that the sensor produces an interesting new sensing then follow, looking into the possible other sensor modes,
modality, that of edge information. Further investigation into of pressure distribition and shear, that may be possible to
the camera output may show that is it is also possible to detect.

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on October 21,2024 at 08:40:26 UTC from IEEE Xplore. Restrictions apply.
 R EFERENCES [8] Begej, S., ”Planar and finger-shaped optical tactile sensors for robotics
manipulation”, IEEE Transaction on Robotics and Automation, vol. 4,
[1] Harmon, L. ”Touch-sensing technology: A review Tech” rep. MSR80- no. 5, pp. 472-484, 1988.
03 . Dearborn, MI: SME Technical Report, Society of Manufacturing [9] Ohka, M., Mitsuya, Y., Takeuchi, S., Ishihara, H., Kamekawa, O., ”A
Engineers, 1980. three-axis optical tactile sensor”, IEEE International Conference on
[2] Robert D. Howe and Mark R. Cutkosky, ”Sensing Skin Acceleration Robotics and Automation. pp. 817-824, vol.1, 1995.
for Slip and Texture Perception”, IEEE International Conference on [10] Kamiyama, K. Kajimoto, H. Kawakami, N. Tachi, S. , ”Evaluation
Robotics and Automation, pp.145-150 vol.1, 1989. of a vision-based tactile sensor”, Proceedings of IEEE International
[3] Gregory J. Gerling and Geb W. Thomas, ”The Effect of Fingertip Conference on Robotics and Automation, Vol 2, pp. 1542-1547, 2004.
Microstructures on Tactile Edge Perception”, The First Joint Euro- [11] Sato, K. Kamiyama, K. Nii, H. Kawakami, N. Tachi, S., ”Measurement
haptics Conference and Symposium on Haptic Interfaces for Virtual of force vector field of robotic finger using vision-based haptic sensor”,
Environment and Teleoperator Systems, 2005. International Conference on Intelligent Robots and Systems, 2008.
[4] Mukai, T., ”Soft areal tactile sensors with embedded semiconductor [12] Lynette A. Jones and Susan J. Lederman, ”Human Hand Function”,
pressure sensors in a structured elastic body”, Proceedings of IEEE Oxford University Press, 2006.
Sensors, pp.1518-1521 vol.3, 2004. [13] Craig Chorley, Chris Melhuish, Tony Pipe, Jonathan Rossiter and
[5] H. Iwanaga and H. Shimoda, ”The three-dimensional microanatomy Graham Whiteley, ”A Biologically Inspired Fingertip Design for Com-
of Meissner corpuscles in monkey palmar skin”, Journal of Neuro- pliance and Strength”, Proceedings of Towards Autonomous Robotic
physiology, Vol. 32, No. 4, pp.363-371, 2003. Systems, (TAROS), 2008.
[6] Kuroki, S., Kajimoto, H., Nii, H., Kawakami, N., and Tachi, S. [14] Kikuuwe, R., Sano, A., Mochiyama, H., Takesue, N., Tsunekawa, K.,
”Proposal of the Stretch Detection Hypothesis of the Meissner Corpus- Suzuki, S., Fujimoto, H., ”The tactile contact lens”, Proceedings of
cle”, In Proceedings of the 6th International Conference on Haptics: IEEE Sensors, pp. 535-538 vol.2, 2004.
Perception, Devices and Scenarios, 2008. [15] Nagata, K. Ooki, M. Kakikur, M., ”Feature detection with an image
[7] Jiar, Y., Lee, K., Shi, G., ”High Resolution and High Compliance based compliant tactile sensor” International Conference on Intelligent
Tactile Sensing System for Robotic Manipulations”, Proceedings of Robots and Systems, 1999.
the 1993 IEWRSJ International Conference on Intelligent Robots and
Systems Yokohama, Japan, July 26-30, 1993.

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on October 21,2024 at 08:40:26 UTC from IEEE Xplore. Restrictions apply.