Abstract—In the process of covering a generic robot withartificial skin, it is necessary to use design tools allowing designersto specify and validate tactile requirements for the scenarioat hand. In particular, given a set of well-defined functionalrequirements (e.g., minimum spatial sensitivity or minimumforce to detect), there are two needs to be fulfilled: (i) tocheck the artificial skin capability to meet these requirementsand criteria; (ii) to drive the customization process to find areasoned trade-off between different (and possibly conflicting)design parameters, such as dielectric thickness or taxel diameter.The main contribution of this article is the description of a robotskin design toolbox based on Finite Element Analysis, able toprovide the designer with insights in the behaviour of large scaletactile systems.
I. INTRODUCTION
During the past three decades, many examples of engi-neering solutions to tactile sensing have been presented inthe literature [1]. The majority of the discussed approachesis focused on the design and fabrication processes of thesingle transducer. Whilst this is important to investigate thesensing modality per se, robots required to perform tasksinvolving more complex forms of interaction are expected touse tactile feedback from large areas of their body. If we limitour attention to approaches specifically taking into accountlarge scale tactile sensing, the number of solutions drasticallyreduces to few examples [2], [3], [4], [5], [6], [7], [8].
The concept of robot skin entails a number of systemlevel problems that simply do not appear when focusingon small tactile sensor arrays. Robot skin must be modular,conformable, easy to manufacture and deploy [8]. Large scalesensing requires to consider such issues as infrastructure andnetworking [9] or calibration [10], just to name but few.
In the process of covering a generic robot with skin, it isnecessary to use tools capable of meeting functional and non-functional requirements in terms of, for instance, maximumdetectable pressure or spatial and frequency sensitivity. Theseimpact on the actual sensing capabilities, the infrastructureor the processing algorithms (and eventually on the tasks tocarry out, such as the capability of detecting gentle touches ormanipulating objects).
Approaches to simulate and analyse the behaviour of tactilesensors made with elastomeric materials borrow insights andmethods from classical contact mechanics [11]. As it has been
notably argued by Ellis and colleagues [12], [13], tactile sen-sors can be modelled more realistically using Finite ElementMethods (FEM), for they capture the geometry and boundaryconditions more faithfully than classical techniques. Indeed,much effort has been devoted to perform FEM analyses oftactile sensors based on different transduction principles [14],[15], [16], [17], [18], [19]. However, these are focused onsingle transducers. It has been only with the work described in[20], [21], [22] that a first direction towards a comprehensivestudy about the role played by different sensing parameters(e.g., thickness of the elastomer, spatial resolution of tactileelements or elastomer shore) has been carried out.
The main contribution of this article is the description ofa robot skin design toolbox (based on a specific robot skindesign [7], [8]) able to provide the designer with insights inthe behaviour of large-scale capacitive tactile systems1. Thetoolbox considers a number of issues normally faced whendesigning and tailoring robot skin for generic humanoid robots.These include the definition of geometrical parameters (e.g.,tactile elements size and pitch, thickness of the elastomer)and physical properties (e.g., elastic modulus, Poisson’s ratio,dielectric permittivity).
The article is organised as follows. Section II describes thereference tactile system. Section III introduces the relevanttoolbox components. Section IV discusses the toolbox valida-tion with respect to the reference tactile technology as well asan example of use. Conclusion follows.
II. THE REFERENCE TECHNOLOGY AND
MANUFACTURING PROCEDURES
The reference robot skin technology has been described in[7], [8]. The adopted sensing mode is based on the capacitivetransduction principle. A capacitive transducer (i.e., a tactileelement, or taxel) is organized in a layered structure: the lowerlayer consists of the positive electrode, which is mounted ona Flexible Printed Circuit Board (FPCB); the upper layer is aground electrode; the central layer is a soft elastomer actingas dielectric between the two electrodes.
As shown in Figure 1(a), the robot skin is made up ofa number of taxels geometrically organized in modules of
1From May 2012 the Toolbox will be available for free use. Please referto the official ROBOSKIN website at www.roboskin.eu.
153
2011 11th IEEE-RAS International Conference on Humanoid Robots Bled, Slovenia, October 26-28, 2011
Fig. 1. The reference skin system: (a) a skin patch; (b) robot skin conformedto a curved body parts.
triangular shape. Figure 1(b) shows how triangular modulesenforce coverage compliance with respect to robot body partswith varying curvatures. In the current prototype, each modulehosts 12 taxels as well as the readout electronics for convertingcapacitance values to digital signals. This is accomplished bya small Capacitance to Digital Converter (CDC) chip (namely,the AD7147 from Analog Devices). The CDC chip can mea-sure variations in capacitance values within the 4÷30𝑝𝐹 rangewith a sensitivity of 0.32𝑓𝐹 .
Normal forces exerted on the skin produce variations incapacitance values reflecting the varied pressure over the taxelpositions. In ideal conditions, the capacitance 𝐶 associatedwith each taxel linearly depends on the dielectric constant 𝜖0,the relative static permittivity of the dielectric material 𝜖𝑟, theoverlap area 𝐴 and the distance 𝑑 (i.e., the thickness of thedielectric) between the two electrodes:
𝐶 = 𝜖0 × 𝜖𝑟 × 𝐴
𝑑(1)
Considering the capacitance model for a single referencetaxel in (1), the output value of the CDC chip for each taxelcorresponds to
Δ𝐶 = 𝐶𝑝 − 𝐶𝑛 = 𝜖0 × 𝜖𝑟 ×𝐴× 𝑑𝑛 − 𝑑𝑝𝑑𝑝 × 𝑑𝑛
, (2)
where 𝐶𝑝 and 𝑑𝑝 are, respectively, the capacitance value andthe elastomer thickness in the contact case (i.e., when a pres-sure is exerted over the taxel), and 𝐶𝑛 and 𝑑𝑛 correspond to theno contact or nominal case. As it can be noticed, a pressureexerted on the taxel produces an increment in capacitance:consequently Δ𝐶 > 0 corresponds to an occurring contact.
All the parameters in (1) and (2) depend on geometricaland physical parameters of the skin design. The need arises tocustomize these parameters according to specific user require-ments and design needs. On the one hand, one may decidefor instance to locate a reduced number of taxels on a robotback torso (e.g., unnecessary tactile granularity or to limitthe computational load associated with tactile data processing)thereby requiring a higher pitch. On the other hand, manipu-lation tasks and social capabilities in human-robot interactionscenarios may require a major tactile sensitivity in the handsand the forearms, thereby requiring different materials (with
an increased relative permittivity) as dielectric medium, at theprice of an increased weight of the skin patch, because theenhancement of relative dielectric permittivity is achieved byfilling the elastomer with ceramic powders.
In order to select the proper combination of geometricaland physical parameters, the skin manufacturing and the skindesign process must be interleaved. As a matter of fact,a design toolbox can help in different manufacturing steps(outlined in [8]), specifically in the selection of the elastomermaterial and thickness, pitch and taxel size, as well as theanalysis and validation of the resulting design choices.
Fig. 2. The interface of the design toolbox.
III. THE SKIN DESIGN TOOLBOX
The design toolbox has been developed using COTS so-lutions. In particular, it is based on COMSOL Multiphysicsand MATLAB2. On the one hand, the COMSOL Multiphysicsenvironment allows skin designers to create the actual multi-physics model of the skin patch of interest and to simulatethe effects of contact phenomena using FEM analyses. Con-sidering the reference skin technology described in SectionII, the multi-physics analysis must take at least structuralmechanics and electrostatics into account, as induced me-chanical deformations directly influence electrical responsesand therefore sensory data. The structural mechanics moduleallows for the computation of deformations, deflections, andinternal forces or stresses within the modelled skin patch. Asa result of the pressure applied over the elastomer, strains anddisplacements are computed. A deformed mesh is producedwhich serves as the basis for the electrostatics module. As amatter of fact, an electric potential is applied between taxelsand the ground plane located above the elastomer, therebyinducing a set of capacitance sensory measurements directlyproportional to the charge on their plates. Since, at a given time
2For further information please refer to the official COMSOL and MATLABwebsites at www.comsol.com and www.mathworks.com.
154
instant, the deformed mesh is fixed, electric phenomena can beinvestigated under electrostatic conditions. On the other hand,as shown in Figure 2, MATLAB has been used to providethe toolbox with a friendly and intuitive user interface, whichcan be used to customize and automate part of the COMSOLsimulation, issue commands and further analyse the results.
The toolbox allows for FEM analyses at different levelsof detail. The simulated geometry can be as simple as thesingle triangular module or more complex, such as any user-defined, module-based patch. This is of utmost importancewhen investigating the property of large-scale tactile systems,because body parts can be covered with complex triangularmeshes.
For the purpose of skin design and customization, the avail-able models based on capacitive transducers are parametrizedwith respect to different classes of parameters, encompassinggeometric, elastomer and contact properties (Figure 2 on theleft).
Geometric Properties. The geometry of the triangular mod-ule is parametrized in many respects. The following parame-ters turn out to be important for the analysis of such functionalrequirements as, for instance, skin sensitivity and response:
∙ Taxel radius. This parameter affects the smallest areathat can be used to infer information about contactphenomena. Default: 2𝑚𝑚.
∙ Taxel pitch. The pitch is the distance between nearbytaxels: a higher pitch is associated with a lower spatialresolution. Default: 5𝑚𝑚.
∙ Module side. This parameter can be chosen by the skindesigner according to specifications on the skin spatialresolution, which depend on the body part the modulemust be fixed to. In principle, where lower spatial reso-lution is allowed, the size of the module can be increased.Default: 30𝑚𝑚.
∙ Elastomer thickness. As a matter of fact, this has greatimpact on capacitance values, since it corresponds to thedistance parameter 𝑑 of (1). Default: 2𝑚𝑚
Default values have been selected according to the currentprototype of the reference tactile system [7], [8].
Elastomer Properties. The electrical and mechanical charac-teristics of the elastomer play an important role in assessing theoverall sensor behaviour. In a design perspective, a completelydifferent behaviour can be obtained by choosing differentelastomer materials [20]. In order to provide a suitable libraryto select materials from, different types of elastomers havebeen characterized both from the mechanical and electricalpoints of view. In particular, bulk elastomers (e.g., PolytekPolyurethane, Smooth-on Ecoflex 00-30), foam silicone (e.g.,Smooth-on SomaFoama) and compounds made up of baserubbers (e.g., SomaFoama, Polytek, Ecoflex 00-30) with 50%wt. of different ceramic fillers (e.g., Dioxide Titanate, 𝑇 𝑖𝑂2,Strontium Titanate, 𝑆𝑟𝑇 𝑖𝑂3 and lead magnesium niobate-lead titanate, PNM-PT) have been considered. This featureis particularly desirable since material characterization proce-dures are usually expensive and time-consuming. Furthermore,
(a) Stress-Strain curve of SomaFoama (b) Young modulus of SomaFoama
Fig. 3. Material properties of SomaFoama.
it is possible to directly specify the most relevant materialparameters. Among them:
∙ Young modulus. This parameters is defined as 𝐸 = 𝜎𝜖 ,
where 𝜎 and 𝜖 are, respectively, the stress and strain [23].The Young modulus can be used to compute the strainof the elastomer subject to a specified exerted pressure.
∙ Poisson’s ratio. It is a measure of the Poisson effectoriginating from the expansion of the elastomer in thedirection normal to the pressure application direction.
∙ Dielectric permittivity. In any possible readout elec-tronics, a lower bound for the detectable capacitancevariation of the transducer is present. For instance, inthe considered technology, this lower bound is 0.32𝑓𝐹 .According to (1), the lower bound is a constraint relatingthe taxel area 𝐴, the elastomer thickness 𝑑 and thematerial dielectric permittivity 𝜖𝑟. In principle, in orderto obtain the same sensitivity performance, an increaseddielectric permittivity can allow for a decreased taxel area(thereby increasing skin sensitivity to small indentation)or for a increased elastomer thickness (thereby increasingsafety and allowing high pressures contact).
Fig. 4. Stress-Strain curves of two different samples of SomaFoama.
The procedure of selecting a specific material for simulationis particularly important. For instance, in the reference tactiletechnology, a foam silicone rubber is used (i.e., Smooth-onSomaFoama) as dielectric material. Figures 3(a) and 3(b)show, respectively, its stress-strain curve and the real andimaginary parts (respectively, red and blue curves in Figure3(b)) of the Young modulus with respect to the strain. This is
155
(a) Cellphone with touch screen
(b) Keyboard
(c) Mouse button
Fig. 5. Examples of contact areas for everyday tasks of four people.
the default material used in the toolbox and the basis for theexperiments reported in Section IV. The material is charac-terized by a number of advantages, namely high compliance,easily handling during skin manufacturing processes and lowweight. However, as shown in Figure 4, since the materialoriginates from a fast chemical reaction, the presence of airbubbles within the material determining a significant densityvariability causes a differing mechanical behaviour betweenany two samples, especially as long as the strain percentageincreases.
These considerations have major consequences over tool-box simulations when SomaFoama is used: (i) the toolboxbehaviour is based on a particular and specific SomaFoamasample, which has been subject to a peculiar characteriza-tion; (ii) there is no guaranteed agreement with respect tothe behaviour of skin patches based on other SomaFoamasamples. Analogous considerations hold for other materialsas well. The lesson learnt is that simulation, although basedon quantitative results, must always be considered critically,since no simulator exists able to represent all the cues intrinsicto manufacturing processes.
Contact Properties. In order to provide skin designers withthe possibility of analysing contact phenomena in terms ofpressure and contact areas for typical contact tasks, typicalindications for pressure and contact area must be provided.A literature survey shows that two different broad classes ofcontacts can be identified on the basis of pressure ranges. Inparticular, as reported in [24], the first class is referred to asgentle touch and is characterized by contact pressures in the0÷ 10𝐾𝑃𝑎 range, whereas the second class, namely manipu-lation like touch, involves pressures in the range 10÷100𝐾𝑃𝑎.
Simulation of large-scale and complex contact phenomenais an active research topic [25], [26]. Beside theoreticalconsiderations about the nature of contact models [11], majorproblems arise at the computational level. These issues areoutside the scope of the presented work. However, we tried tofind a reasoned trade-off between realistic contact scenariosand computationally tractable simulations. In particular, wefocused on contact areas as much akin to human fingertips
as possible, because contact modelling in grasping and ma-nipulation tasks represents a significant part of the researchactivities in the field [1]. To this aim, we performed simpleexperiments in order to obtain information about fingertipcontact areas. We asked 12 subjects to perform a number ofeveryday common touch tasks, such as operating the touchscreen of a cellphone, typing on a keyboard and clickingmouse buttons. Each experiment has been performed by eachsubject four times, in order to obtain a statistically significantnumber of trials. For each trial, the average shape of contactarea and exerted pressure are recorded and used in the toolbox.
TABLE ITYPICAL CONTACT AREAS
People Contact EventsAge Sex Cellphone [𝑚𝑚2] Keyboard [𝑚𝑚2] Mouse[𝑚𝑚2]23 F 132.22 197.37 192.9224 M 55.98 135.16 96.2525 M 96.27 88.30 207.4325 F 39.36 53.23 129.6526 F 55.65 90.52 197.3626 M 70.17 127.96 148.6327 M 78.59 103.62 200.6429 F 54.50 68.12 148.9833 M 32.12 81.42 133.28Average 69.91 108.80 163.61
Figure 5 shows the fingertip areas obtained in two typicaltouching scenarios, for four different subjects. These contactareas represent realistic indenter shapes that provide skin de-signers with a (possibly rough) idea about contact phenomenafor the considered tactile system. Table I shows numericalvalues for the different contact areas, in the three identifiedscenarios. These data are used as realistic test cases whenvalidating skin design choices related to its physical andgeometrical parameters.
IV. EXPERIMENTAL VALIDATION AND DISCUSSION
In order to assess the general properties of toolbox simula-tion results, we decided to compare real and simulated tactiledata. The evaluation has been performed on taxel capacitancevariations and spatial responses. On the basis of previous work,we replicated the same set of experiments discussed in [8].Experimental data have been obtained from a skin patch fixedon the palm of the iCub robot. A specific taxel of the patch hasbeen stimulated using an indenter located at the end-effectorof a Cartesian robot with a load cell mounted on it. A numberof cylindrical probes of varying diameter (respectively, 4, 5, 6and 10𝑚𝑚) have been used. A sequence of contact depths isimposed, with a resolution of 0.1𝑚𝑚. At each defined depth,probes are maintained in firm contact position for about 2𝑠,in order to minimize the effects of material hysteresis on themeasurements, and subsequently they are moved up. After 20𝑠,the subsequent contact depth is defined as the next set point.
Since the output from the CDC is a digital signal, it isnecessary to convert actual numerical values in capacitancevariations with respect to the nominal condition (i.e., no
156
(a) 4𝑚𝑚 probe (b) 5𝑚𝑚 probe
(c) 6𝑚𝑚 probe (d) Different Young modulus
Fig. 6. Comparison between experimental and simulation data for 4, 5 and6𝑚𝑚 probes. Blue lines are related to experimental data while red lines arerelated to simulation data.
contact), as follows:
Δ𝐶 = (𝑐𝑛 − 𝑐𝑑)× 2𝑏 × 𝐶𝑏, (3)
where 𝑐𝑛 is the digital value corresponding to nominal condi-tion, 𝑐𝑑 is the digital value under elastomer deformation, 𝑏 isthe number of shifted bits that are used for limiting the noiseeffect and 𝐶𝑏 is the capacitance value that corresponds to asingle bit (i.e., 0.32𝑓𝐹 ).
After this step, it is possible to compare real capacitancevariations with simulated ones, which are directly provided asthe result of the toolbox based simulation. In simulation, anarea of the same shape and size of the probe has been chosenas contact area, placed over the upper face of the elastomerand centred above a taxel; the same pressure detected from theload cell has been applied. The structural mechanics modulecomputes the elastomer deformation related to the pressure ap-plied and, according to its outcome, the electrostatics modulecomputes the capacitance values of each taxel.
Figure 6 shows the comparison between experimental andsimulated data when probes of different size are used. Ineach case, it can be noticed that experimental and simula-tion results agree until the value of exerted pressure reachesaround 20𝐾𝑃𝑎. When the exerted pressure exceeds this value,simulation results deviate from experimental ones as long asthe applied pressure increases.
As anticipated, this difference originates from the specificcharacterization of the particular sample used. As a matter offact, the SomaFoama sample whose characterization is storedin the toolbox library is different from the sample coveringthe iCub palm, as the consequence of the impossibility ofcontrolling all the aspects related to the SomaFoama manufac-turing process. Furthermore, since elastomers are visco-elasticpolymers, the response of the material to a stress distribution
(a) Simulation results (b) Experimental results
Fig. 7. Comparison between experimental and simulation data related to taxelspatial responses. The taxel in red is the one where a pressure is applied, theyellow ones are the taxels that result excited due to the elastomer deformation.
depends on the material history, in terms of previously exertedstimuli, as well as the velocity and duration of contact events.
In order to validate the simulation model against differentcharacterizations, we perturbed the curve of the characterizedelastic modulus of ±5% and ±10% and we repeated insimulation the tests for all the probes. Figure 6(d) showsa comparison between experimental results and simulationsobtained perturbing the elastic modulus of ±10%. Resultsobtained with a 10% increased elastic modulus are a goodapproximation of experimental results. From this insight, wecan infer that material characterization is very important toobtain realistic simulations.
Regarding taxel spatial activations the comparison betweensimulation and experimental results is shown in Figure 7.The Figure refers to the application of a 50𝐾𝑃𝑎 pressurewith a 6𝑚𝑚 probe. In the simulation results the four taxelsaround the mechanically stressed one are activated while in theexperimental results only two of them are. One possible reasonfor this behaviour is that the pressure area in the simulation isperfectly centred above the taxel, whereas in the experimentalresults a little displacement can have affected the spatial taxelactivation.
In order to test the design tool we selected two possiblescenarios. In the first an “object manipulation” pressure andan area of contact of the type “touching keyboard” havebeen selected. For a triangle module the default geometricconfiguration has been used, however the dielectric layer isSomaFoama filled with PMN-PT. In the second a “gentletouch” pressure has been chosen, whereas other parametersare the same of the previous case. Displacement results areshown in Figures 8(a) and 8(c) as displacement fields, whereasin Figures 8(b) and 8(d) the related taxel activations are shown,with a color gradient from red to yellow, where red representsthe activated taxels and yellow the unaffected ones.
V. CONCLUSION
This article describes a design and simulation tool for largescale capacitive tactile sensors, i.e., sensors possibly coveringlarge surfaces of robot bodies. FEM based analyses are used,exploiting the deformed mesh principle to obtain a multi-physics simulation. The reference sensing architecture is mod-elled in its most relevant features, adopting an incremental andparametrized approach allowing, in principle, to experimentwith different mechanical structures.
Models and simulation results are described, showing thatthis approach is a viable solution to investigate a numberof system level features otherwise difficult to assess withreal experiments, because of the necessity to have differenttechnological samples, in terms of cost and time. On-goingwork includes the assessment of the effects of such physicalquantities as friction and tangential forces, as well as simu-lations over curved surfaces and of more complex form ofcontact.
ACKNOWLEDGMENT
The research leading to these results has received fundingfrom the European Community’s Seventh Framework Pro-gramme (FP7/2007-2013) under Grant Agreement no. 231500(Project ROBOSKIN).
REFERENCES
[1] R. Dahiya, G. Metta, M. Valle, and G. Sandini, “Tactile sensing: fromhumans to humanoids,” IEEE Transactions on Robotics, vol. 26, no. 1,pp. 1–20, February 2010.
[2] Y. Ohmura, Y. Kunyoshi, and A. Nagakubo, “Conformable and scalabletactile sensor skin for curved surfaces,” in Proceedings of the 2006IEEE International Conference on Robotics and Automation (ICRA’06),Orlando, FL, USA, May 2006.
[3] T. Tajika, T. Miyashita, and H. Ishiguro, “Automatic categorization ofhaptic interactions - what are the typical haptic interactions between ahuman and a robot?” in Proceedings of the 2006 IEEE-RAS InternationalConference on Humanoid Robots (HUMANOIDS 2006), Genova, Italy,December 2006.
[4] I. Mizuuchi, T. Yoshikai, T. Nishino, and M. Inaba, “Developmentof the musculoskeletal humanoid kotaro,” in Proceedings of the 2007IEEE International Conference on Robotics and Automation (ICRA’06),Orlando, FL, USA, May 2006.
[5] T. Minato, Y. Yoshikawa, H. Ishiguro, and M. Asada, “CB2: A childrobot with biomimetic body for cognitive developmental robotics,”in Proceedings of the 2007 IEEE-RAS International Conference onHumanoid Robots (HUMANOIDS 2007), Pittsburg, PN, USA, December2007.
[6] T. Mukai, M. Onishi, S. Hirano, and Z. Luo, “Development of the tactilesensor system of a human-interactive robot ri-man,” IEEE Transactionson Robotics, vol. 24, no. 2, pp. 505–512, April 2008.
[7] G. Cannata, M. Maggiali, G. Metta, and G. Sandini, “An embed-ded artificial skin for humanoid robots,” in Proceedings of the 2008IEEE International Conference on Multi-sensor Fusion and Integration(MFI’08), Seoul, Korea, August 2008.
[8] A. Schmitz, P. Maiolino, M. Maggiali, L. Natale, G. Cannata, andG. Metta, “Methods and technologies for the implementation of large-scale robot tactile sensors,” IEEE Transactions on Robotics, vol. 27,no. 3, pp. 389–400, June 2011.
[9] E. Baglini, G. Cannata, and F. Mastrogiovanni, “Design of an embeddednetworking infrastructure for whole-body tactile sensing in humanoidrobots,” in Proceedings of the 2010 IEEE Conference on HumanoidRobotics (HUMANOIDS 2010), Nashville, TN, USA, December 2010.
[10] G. Cannata, S. Denei, and F. Mastrogiovanni, “Towards automated self-calibration of robot skin,” in Proceeding of the 2010 IEEE InternationalConference on Robotics and Automation (ICRA’10), Anchorage, Alaska,USA, May 2010.
[12] S. Ricker and R. Ellis, “2-d finite-element models of tactile sensors,”in Proceedings of the 1993 IEEE International Conference on Roboticsand Automation (ICRA’93), Atlanta, Georgia, USA, May 1993.
[13] R.Ellis and M. Qin, “Finite-element and singular-value analysis of tactilesensors,” in Proceedings of the 1994 IEEE International Conference onRobotics and Automation (ICRA’94), San Diego, CA, USA, May 1994.
[14] T. Maeno, T. Kawamura, and S. Cheng, “Friction estimation by pressingan elastic finger-shaped sensor against a surface,” IEEE Transactions onRobotics and Automation, vol. 20, no. 2, p. 222.228, April 2004.
[15] S. Yahud, S. Dokos, J. Morley, and N. Lovell, “Finite element analysisof a tactile sensor for a robotic hand,” in Proceedings of the 2008International Conference on Intelligent Sensors, Sensor Networks andInformation Processing (ISSNIP’10), Brisbane, Australia, December2010.
[16] M. Qasaimeh, S. Sokhanvar, J. Dargahi, and M. Kahrizi, “Pvdf-basedmicrofabricated tactile sensor for minimally invasive surgery,” Journal ofMicroelectromechanical Systems, vol. 18, no. 1, pp. 195–207, February2009.
[17] V. Ho, D. Dao, S. Sugiyama, and S. Hirai, “Analysis of sliding of asoft fingertip embedded with a novel micro force / moment sensor:Simulation, experiment, and application,” in Proceedings of the 2009IEEE International Conference on Robotics and Automation (ICRA’09),Kobe, Japan, May 2009.
[18] Y. Zhang, “Sensitivity enhancement of a micro-scale biomimetic tactilesensor with epidermal ridges,” Journal of Micromechanics and Micro-engineering, vol. 20, no. 8, August 2010.
[19] M. Tiwanaa, A. Shashankb, S. Redmondb, and N. Lovella, “Character-ization of a capacitive tactile shear sensor for application in robotic andupper limb prostheses,” Sensors and Actuators A: Physical, vol. 165,no. 2, pp. 164–172, February 2011.
[20] M. Shimojo, “Spatial filtering characteristic of elastic cover for tactilesensors,” in Proceedings of the 1994 IEEE International Conference onRobotics and Automation (ICRA’94), San Diego, CA, USA, May 1994.
[21] ——, “Mechanical filtering effect of elastic cover for tactile sensors,”IEEE Transactions on Robotics and Automation, vol. 13, no. 1, pp. 128–132, February 1997.
[22] C. Wen and W. Fang, “Tuning the sensing range and sensitivity of threeaxes tactile sensors using the polymer composite membrane,” Sensorsand Actuators A: Physical, vol. 145-146, pp. 14–22, February 2008.
[23] I. Ward and J. Sweneey, An Introduction To Mechanical Properties OfSolid Polymers. Wiley and Sons Ltd., 2004.
[24] S. Mannsfeld, B. Tee, R. Stoltenberg, C. Chen, S. Barman, B. Muir,A. Sokolov, C. Reese, and Z. Bao, “Highly sensitive flexible pressuresensors with microstructured rubber dielectric layers,” Nature Materials,vol. 9, pp. 859–864, September 2010.
[25] H. Alirezaei, A. Nagakubo, and Y. Kuniyoshi, “A highly stretchabletactile distribution sensor for smooth surfaced humanoids,” in HumanoidRobots, 2007 7th IEEE-RAS International Conference on, 29 2007-dec.1 2007, pp. 167 –173.
[26] Y. Fujimori, Y. Ohmura, T. Harada, and Y. Kuniyoshi, “Wearable motioncapture suit with full-body tactile sensors,” in Robotics and Automation,2009. ICRA ’09. IEEE International Conference on, may 2009, pp. 3186–3193.
