review article tactile sensing for mechatronics*a …...5 m[h[ lee\ h[r[ nicholls:mechatronics 8...

$: Review Article Tactile sensing for mechatronics*a …...5 M[H[ Lee\ H[R[ Nicholls:Mechatronics 8 "0888# 020 application opening up in the future which will capitalise on those aspects$
\PERGAMON Mechatronics 8 "0888# 0Ð20

9846Ð3047:88:,*see front matter Þ 0887 Elsevier Science Ltd[ All rights reserved[PII] S 9 8 4 6 Ð 3 0 4 7 " 8 7 # 9 9 9 3 4 Ð 1

Review Article

Tactile sensing for mechatronics*a state of theart survey

M[H[ Leea\�\ H[R[ Nichollsb

a Department of Computer Science\ University of Wales\ Aberystwyth\ U[K[b Industrial Research Limited\ Christchurch\ New Zealand

Received 15 May 0887

Abstract

In this paper we examine the state of the art in tactile sensing for mechatronics[ We de_ne atactile sensor as a device or system that can measure a given property of an object or contactevent through physical contact between the sensor and the object[ We consider any propertythat can be measured through contact\ including the shape of an object\ texture\ temperature\hardness\ moisture content\ etc[

A comprehensive search of the literature revealed that there was a signi_cant increase inpublications on tactile sensing from 0880 onwards[ Considerable e}ort in the 0879s was spentinvestigating transduction techniques and developing new sensors\ whilst emphasis in morerecent research has focused on experiments using tactile sensors to perform a variety of tasks[

This paper reports on progress in tactile sensing in the following areas] cutaneous sensors\sensing _ngers\ soft materials\ industrial robot grippers\ multi_ngered hands\ probes andwhiskers\ analysis of sensing devices\ haptic perception\ processing sensory data and newapplication areas[

We conclude that the predominant choice of transduction method is piezoelectric\ witharrays using resistive or capacitive sensing[ We found that increased emphasis on understandingtactile sensing and perception issues has opened up potential for new application areas[ Thepredicted growth in applications in industrial automation has not eventuated[ New applicationsfor tactile sensing including surgery\ rehabilitation and service robotics\ and food processingautomation show considerable potential and are now receiving signi_cant levels of researchattention[ Þ 0887 Elsevier Science Ltd[ All rights reserved[

� Corresponding author[ Tel[] 9933 0869 511 319^ fax] 9933 0869 511 344^ e!mail] mhlÝaber[ac[uk

M[H[ Lee\ H[R[ Nicholls:Mechatronics 8 "0888# 0Ð201

0[ Introduction

Tactile sensing is the process of determining physical properties and events throughcontact with objects in the world[ We review the current state of the art in tactilesensing and examine the trends\ developments and application areas in this interestingand relatively new _eld[

Tactile sensors o}er exciting possibilities for use in mechatronic devices and mea!suring instruments in many areas of science and engineering[ Robotics and industrialautomation are the application areas that have generated the most interest becausecontact interactions are a fundamental feature of any physical manipulation system[However\ there are many other potential application areas including agriculture\food processing\ medicine\ dentistry\ entertainment\ and future domestic and serviceindustries[ This literature survey considers only papers with direct relevance forarti_cial or automated tactile sensing through contact between objects and sensors[Thus we will not consider force sensors or other such seemingly related devices exceptwhere they are particularly relevant to tactile problems[

Signi_cant previous surveys of tactile sensing were carried out by Harmon in 0873ð0Ł\ Nicholls and Lee in 0878 ð1Ł and Nicholls in 0881 ð2Ł[ In this paper\ we build onthose surveys by reviewing the literature of the 0889s[ Starting from the earlier surveyswe cover the newer material and thus provide entry points for researchers who wishto work in this _eld[ We also assess the likely developments and research motivationsfor the near future[

1[ The nature of tactile sensing

We de_ne a tactile sensor as a device or system that can measure a `iven property ofan object or contact event throu`h physical contact between the sensor and the object[Some previous de_nitions have assumed that tactile properties must involve force anda number of early tactile sensors were based on strain gauges or other force sensingarrangements for normal and:or shear stresses[ However\ we reject that narrowde_nition and accept any property that is measured through contact\ including theshape of an object\ texture\ temperature\ hardness\ moisture content\ etc[

Although most would agree that tactile sensing is essentially a spatially distributedprocess\ much of the literature describes devices and applications based on localisedsensing[ Many sensory devices have been developed in terms of tactile sensing points\pads or surfaces[ The objective of such work is the production of robust and reliablelocally discrete devices\ e[g[\ tactile sensors mounted on the gripping surface of arobot end!e}ector[ A more ambitious\ related objective is to develop a fully distributedtactile {skin|[

Another approach views tactile sensing as an integral part of manipulation[ In thisapproach\ the joint angles of a robot arm or _nger are often combined with surfacenormals detected via a _nger mounted tactile sensor to construct a 2D representationof a surface by following its contours[ Research in this endeavour concerns the roleof manipulators\ often multi_ngered robot grippers\ whose kinaesthetic and kinematic

M[H[ Lee\ H[R[ Nicholls:Mechatronics 8 "0888# 0Ð20 2

aspects are controlled and coordinated with the contact events taking placing duringmanipulation[ This very integrated approach draws inspiration from observations ofdextrous behaviour in the human system[

2[ The growth of tactile sensing

Tactile sensing began to develop in the 0869s but Nicholls and Lee ð1Ł report only_ve papers in that decade[ The surveys of Harmon ð3Ł and of Nicholls and Lee ð1Łboth show a wide diversity in the types of sensing device that were invented anddeveloped in the 0879s[ New transduction techniques were developed and a largenumber of experimental devices and prototypes were built and reported in the litera!ture[ However\ despite repeated suggestions in many papers regarding the potentialimportance of tactile sensing to industrial automation and other application areas\the literature reported very few applications beyond the experimental prototype stageand hardly any in serious regular use[ We will consider reasons for this slow com!mercial exploitation in industrial applications and assess the chances for future devel!opment[ First we examine the growth rates of the literature in this _eld[

Using bibliographic searching tools we have examined a large number of scienti_cdatabases and have collated numerical and qualitative results[ Figure 0 presents thesearch data in terms of numbers of journal articles published each year containingthe words {tactile sensor:sensing| in the title\ key words or abstract\ from 0870 to thepresent[

Two curves are shown*one is the tactile technology literature and the other ismedical or physiological material that relates human tactile sensitivity and otherproperties of the human tactile system to various medical conditions[ These medicalpapers were separated out as they are not relevant to the technology or use of tactilesensing[ Figure 0 shows that around 0880 a marked increase in the publishing rateoccurred "to more than double# in both scienti_c:engineering and medical areas[

In a separate examination of databases of conference papers we found a similarstory with an average rate of publications on tactile sensing of about 5 per year up to0889 and an average of over 01 per year after 0889[ The earlier in~ection point is inaccord with the tendency of journal articles to lag conference presentations[

It is noticeable that the absolute numbers of publications in tactile sensing are nothigh when compared with other sensing technologies[ For example\ by searching onkeywords {computer:machine vision| "which is known to capture just a subset of theliterature# over the same periods as above we _nd an average of about 019 journalarticles per year with an average of over 149 total articles per year over the last fouryears[ This order of magnitude increase over tactile sensing re~ects both the maturityof the computer vision _eld and the burgeoning applications of the technology[

3[ On sensing modalities

There are _ve main sensing modalities*sight\ sound\ smell\ taste and touch[ It isinstructive brie~y to compare them to gain insight into the development of tactilesensing[


Fig[ 0[ "a# Journal paper count per year^ "b# Cumulative count of tactile papers[

Computer vision systems are now commercially available and widely used forindustrial inspection\ recognition\ monitoring and many other application areas[Similarly\ for audio systems^ many kinds of audio processing systems are availableand there exists a long and substantial literature on the mathematics and technologies


of signal processing[ Human speech analysis and speech recognition are currentlyvery active research areas and have identi_ed a wide range of applications[ Even thesenses of smell and taste have their electronic analogies[ Devices known as electronicnoses are now readily available for the detection of a range of molecules\ and chemicaltests can be implemented to automatically analyse across particular spectrums in asimulation of taste[

Why then does the sense of touch seem to be so neglected< We do have a historyof research in tactile sensing\ but it is nowhere near as large in volume or as welldeveloped as in the other sensory modalities[ There seem to be several reasons forthis apparent neglect[ Drawing on the human analogy we see some of the di.culties]

, No localised sensory organ] unlike sight and hearing\ the sense of touch has nosingle sensory organ but operates throughout the skin as a distributed and di}useprocess[ The transduction of tactile signals is distributed over a much wider areathan in a single localised sensory organ\ such as eyes and ears[ The simulation ofthis through the creation of an arti_cial tactile skin is a much more di.cult taskthan the development of discrete sensing devices[

, Complex sensing] tactile sensing through the skin is not a simple transduction ofone physical property into an electronic signal[ Touch takes many forms andincludes the detection of shape\ texture\ friction\ force\ pain\ temperature and manyother related physical properties[ It is not yet very well understood how thesedi}erent aspects of the tactile phenomena are related and how they are processedby the nervous system[ Consequently\ it is not easy to _nd suitable technologicalanalogies in science or engineering[

, Di.cult to imitate] unlike sight and sound which are well de_ned physical quantitieswe do not know what are the best measures to adopt for a tactile sensor[ A wholerange of physical properties can be transduced and used as tactile signals\ but it isunclear as to which is the most appropriate for applications and which should bedeveloped further[ In comparison\ computer vision research is now mainly con!cerned with images and their understanding and interpretation\ not with developingnew cameras and optics[ Because the capture of visual data is a solved problemresearchers do not have to worry about such low level issues[ Tactile sensing hasnot yet reached this point] we are still concerned with the fundamentals of datacapture[

Considering these di.culties\ it is encouraging that so much has already been achievedand so many di}erent kinds of tactile sensor are available for experimentation andfurther development[ In fact\ if we consider tactile sensing in its most primitive form\i[e[\ simple contact sensing using a switch for presence or absence of an object\ thenwe see that it is not neglected at all and is used in all kinds of situations such as inmachinery\ vehicles and safety systems[ We refer to single binary devices as {spatialswitches| and\ as is usual\ we will exclude them from our survey and concentrate onmore advanced sensing\ involving scalar and array variables[

The lack of any widespread application of tactile sensing is partly due to thedi.culties mentioned above and also to the lack of availability of commercial sensorswith suitable con_gurations and characteristics[ Nevertheless\ we see new areas of


application opening up in the future which will capitalise on those aspects providedby tactile sensing that are missing in other sensory modalities[

4[ The state of the art in 0889

After a decade of experimentation and device development\ the surveys of tactilesensing published around 0889\ e[g[\ Nicholls and Lee ð1Ł\ report on a range of sensorsthat can detect object shape\ size\ presence\ position\ forces and temperature[ Veryfew examples of sensors that could detect surface texture\ hardness or consistencywere identi_ed[ Most of the devices were either of the scalar single!point contactvariety or were linear or rectangular arrays of sensing elements[ The main transductionmethods identi_ed were the following] Resistance and Conductance\ Capacitance\Piezoelectric and Pyroelectric\ Magnetic\ Magnetoelectric\ Mechanical\ Optical\Ultrasonic\ and Strain gauges[

The processing of tactile signals mainly centred on image processing techniques\derived from computer vision research and applied to static images from array sensors[Some statistical methods were reported for the extraction and recognition of features[Segmentation methods using edge extraction\ thresholding\ _ltering and boundarygrowing were all employed[ Geometric measures and the method of moments werealso popular[ However\ the importance of dynamic events was recognised and sensorswere developed for detecting stress changes\ slip and other temporal contact events[

The detection of certain material properties\ such as hardness and surface texture\were still problem areas for tactile sensing and had not been e}ectively solved[

Following the pattern in active vision research\ the ideas of active tactile sensingonly emerged after much work on static data had been completed[ Active control wasmore rapidly recognised as important for the tactile domain[ Haptic perceptionand the role of kinaesthetic information were being investigated and exploratoryprocedures for tactile property extraction were being developed[ Contributions fromphysiology\ psychophysics and neuroscience were giving valuable insight into thenature of human tactile perception[

Nearly all the cited papers dealt with experimental developments and very fewcommercial quality _nished products were available[ Despite industrial robotics oftenbeing cited as the major application area\ additional areas were being proposed forthe exploitation of the technology[ For example\ robotics intended for unstructured\real world applications such as in medicine\ exploration and hazardous environmentswere becoming increasingly represented in the literature[

In our review below\ we consider the progress made to early 0887[

5[ A taxonomy for the recent tactile sensing literature

It is useful to structure the various contributions from the literature into some formof classi_cation scheme[ We identi_ed a primary interest area for each paper and usedthese to group the material into natural clusters[ The primary interest areas are]


"0# Cutaneous sensors[ This category covers skin mounted sensors and we includesingle point devices\ imaging array sensors and arti_cial skins[

"1# Sensing _ngers[ Tactile sensors are often built into rigid _nger structures in orderto measure performance and test their capabilities in probing and exploratorytasks[

"2# Soft materials[ Recently softer materials have become of interest as potentialcomponents for tactile _ngers[

"3# Industrial robot grippers[ This is one of the original application areas envisagedfor tactile sensing[ However\ we _nd little signi_cant recent development[

"4# Multi_ngered hands[ Highly actuated arti_cial hands are necessary for dextrousmanipulation tasks[ This involves much consideration of integration issues andsheds light on the role of tactile sensors[

"5# Probes and whiskers[ Fine probes and whiskers can be used to detect points ofcontact and discern their spatial positions[ Recent work has investigated suchactive touch sensing devices[

"6# Analysis of sensing devices[ The analysis and evaluation of technological devel!opments is a vital part of good science[ This category covers material that shedslight on the design\ characteristics and performance of sensing devices[

"7# Haptic perception[ Active and haptic sensing combines cutaneous sensing withthe actions available to an agent such as a robot arm in an integrated andcoordinated control regime[ This _eld shows how tactile sensing plays an intimateand complementary role in highly integrated perceptual systems[

"8# Processing sensory data[ This category covers methods and algorithms for pro!cessing tactile sensory data[ It ranges from data interpretation using neuralnetworks\ fuzzy logic and model!based approaches\ to sensor data fusion wheretactile data must be integrated with sensory data from other modalities[

"09# New application areas[ Active areas identi_ed in this survey are medical andservice industries\ agriculture and food[

A selection of the material we surveyed is reported below under these interest areas[We have cited papers which either make a signi_cantly novel contribution or whichtypify papers reporting on a particular tactile sensing issue[

5[0[ Cutaneous sensors

The production of new designs and con_gurations of sensors continues apace[However\ most of the possible forms of physical transduction methods have nowbeen explored and there seems little scope for new fundamental transducers\ seeNicholls ð2Ł and Russell ð4Ł for details of the basic transduction methods[ The currentresearch o}erings are mainly concerned with novel packagings\ better designs\improved engineering and more complete analysis[

There is an important distinction in tactile sensing between extrinsic and intrinsicsensing[ Extrinsic sensors are devices that are mounted at or near the contact interfaceand deal with localised regions\ while intrinsic sensing refers to the derivation ofcontact data from force sensing within the mechanical structure of the system[ Force


sensing produces information such as contact vectors and point locations[ See Ciccettiet al[ ð5Ł for an example of an intrinsic sensor] this is a miniaturised six axis forcesensor built inside a _nger[ This survey is focused on extrinsic tactile sensing\ and sowe do not cover the details of such intrinsic designs "which are usually based on straingauges# or any other forms of force:torque sensors[

A wide range of tactile sensors are reported on in the literature[ When designing orassessing a sensor\ the following points should be considered for selecting designcriteria]

, Parameter"s# to be measured*what parameter will the sensor measure "e[g[\ force\torque\ presence:absence of contact\ thermal conductivity\ slip\ etc#< What is therange of detectable measurements<

, Spatial resolution*what are the dimensions of the sensing area\ and what is thespatial distribution of the tactile elements "{tactels|#<

, Response pro_le*what is the sensor|s response to increasing:decreasing stimuli "thisshould be a monotonic response that is stable\ repeatable and without hysteresis#<

, Time resolution of the sensor*how many samples per second<

5[0[0[ Sensin` devicesThere are two main forms of extrinsic tactile sensor] single value sensors and arraysof integral sensing elements[ In both cases integrated circuit technology has been usedto fabricate chip based devices[ We illustrate these ideas below with selected workfrom the literature[

Beebe et al[ ð6Ł describe a silicon based piezoresistive force sensor that addressesthe problems of robust packaging\ small size and overload tolerance[ The sensormeasures the force "rather than pressure# applied to a 3 mm raised dome on the devicesurface[ The device exhibits a linear response\ good repeatability and low hysteresis\and has a ~exible and durable packaging[ Arrays could be fabricated with an estimatedspatial resolution of 1Ð3 mm[ Trials are described using the device as a _nger mountedsensor for measuring pinch force[

Wol}enbuttel et al[ ð7Ł have researched extensively into silicon fabricated sensorsand see this approach as a way of avoiding some of the problems associated withelastic membranes such as hysteresis[ Their work on piezoresistive and capacitivemicromachined sensors has produced designs for arrays of force sensing elementsusing diaphragms or cantilevers as the sensing principle[

A combined three!axis force and slip sensor has been described by Yamada andCutkosky ð8Ł[ A domed tactile head transmits force to three nibs that each rest onpolyimide resistive sensor pads[ The applied force is resolved into three axes and slipis detected by a piece of piezoelectric PVDF _lm moulded into the head[ The signalrates reported were 014 Hz for force sensing and 0 kHz for the stress rate sensor[

Kolesar and Dyson ð09Ł describe a sensor based on an 7×7 array of sensing siteson a CMOS device covered with a PVDF polymer _lm[ Piezoelectric materials arenow widely used in tactile sensing\ due to their sensitivity and availability in variousplastic forms[ The Kolesar and Dyson sensor includes MOSFET ampli_ers on thechip which covers an area less than 099 mm1 and is compatible with _nger tip


applications[ It features a pre!charge bias technique which initialises the sensors beforeeach cycle in order to reduce the problems of stability and reproducibility that havelong been associated with piezoelectric e}ect devices[

Gray and Fearing ð00Ł report on an 7×7 capacitive fabricated array that is 0 mm1

in area[ This gives a spatial resolution at least 09 times better than the human limitof 0 mm and is intended for medical applications involving small manipulators andendoscopic surgery[ It could be mass produced and therefore disposable*a fairlynovel idea in tactile sensing[ Severe hysteresis was the main drawback[

Omata and Terunuma ð01Ł point out that most sensors are incapable of sensingmany of the range of physical properties that materials exhibit[ Most sensors measurepressure or force and are unable to sense many e}ects that are experienced by humans\e[g[\ friction\ stickiness\ texture\ hardness\ and elasticity[ With a view to palpationapplications in medical examination\ they argue that hardness and softness detectionrequire di}erent approaches\ especially for sensing variations in soft tissue\ anddescribe a sensor that approximates to humans in this respect[ The sensor has apiezoelectric resonator\ driven at 50 kHz\ and works on the principle that contactwith an object will cause a change in resonating frequency[ The device is packagedinto an acrylic tube\ 04 mm diameter by 54 mm long\ and in simulated cancer tests itdetected 2 mm diameter glass balls 19 mm below the surface of a silicone breastmodel[ The main problems were the need to maintain a constant contact pressure "19grams# and a slow time response[

Mixed function sensing is an interesting possibility[ Li and Shida ð02Ł describe amultifunction sensor consisting of two interleaved planar spiral coils 24 mm in diam!eter[ The coils can be used in three ways] as a capacitive sensor where the dielectricconstant of the object a}ects the capacitance between the coils^ as an inductor wherethe frequency transfer function between the coils can distinguish a magnetic\ non!magnetic or nonconducting material^ and as a thermal sensor where one coil acts asa heater and the other a temperature sensing resistor[ Problems include the thermaltime response of several seconds\ the need to insulate metallic objects "cellophanetape was used#\ and an implicit assumption of constant applied contact pressure[

Monkman and Taylor ð03Ł also report on a thermal sensor^ in this case based on apyrometer design that gives an order of magnitude improvement in response timeover other thermal methods[ As thermal sensors are fundamentally limited owing totheir sensing of a single material parameter\ Monkman and Taylor also argue formultifunction devices and describe a combined thermal\ electrical and mechanicalsensor[

Another\ and perhaps neglected\ aspect of tactile sensing concerns extensive surfacesensing over a bounding envelope or skin[ There are obvious applications for a sensingskin that can cover robots or other moving equipment in order to detect contact withobjects or humans in a local environment[ An experiment with a full!body sensingsuit for a small robot is described by Inaba et al[ ð04Ł[ The company Tekscan ð05Łproduces a range of sensing systems including pressure mats\ in!shoe foot pressurearrays\ and other gait and stance measuring systems[ The sensing principle is basedon arrays of resistive elements embedded in a large area ~exible _lm only 9[07 mmthick[ The F!Scan system allows the ~exible sheet to be cut to shape and inserted into


shoes for real!time pressure measurements[ This application has 859 sensing tactelseach covering around 4 mm1[ See Donaghue and Veves ð06Ł for a review of suchsystems\ including optical pedobarographs that use the same optical internal re~ectionprinciple as used in some tactile sensors[ Clearly\ if _tted to a robot|s surface area\this sensing _lm could be an ideal skin for service robots that work in cooperationwith humans and must detect any collisions or contacts[ Work by Yamada et al[ ð07Łhas dealt with the related problem of protecting the joints and elbows of robotmanipulators[ In this case a _xed skin is unsuitable as moving joints have constantlychanging surface envelopes[ Yamada used helical springs with 37 tactile sections^contact on a section causes an LC circuit to be activated and by swept frequencyscanning all sensors can be read through a single pair of wires[

Another notable commercial development is the FSR "Force Sensing Resistor#[These are resistive polymer _lm elements manufactured by Interlink ð08Ł and arewidely used in pointing and position sensing devices such as joysticks[ FSRs\ beinginexpensive and readily available\ are found in many experimental tactile systems[

A di}erent but promising looking technique is acoustic ultrasonic sensing[ Micro!phones are known to be useful for detecting surface noise that occurs at the onset ofmotion and during slip[ Ando and Shinoda ð19Ł describe a device that senses contactevents from their ultrasonic emission at the contact point[ A PVDF polymer is usedin a 1×1 array of receivers to localise the contact point on a silicone rubber sensingdome[ They reported that this sensor is very e}ective in detecting slip and surfaceroughness during movement[

In a variation of the design\ Shinoda and Ando ð10Ł\ used ultrasonic transmittersand receivers to detect changes in wavefronts due to distortion and could detectdisplacements as small as ten micrometers[ Hutchings\ Grahn and Petersen ð11Ł alsodescribe an ultrasonic sensor but this time using a cross _eld transmission in whichthe compression of an elastomer pad changes the transmission properties[ Anothersensor by Shinoda and colleagues ð12Ł uses microphones to detect changes in airpressure within cavities inside a raised spherical silicone dome[ A novel method ofdiscriminating surface composition was used by a mobile robot that tapped a boom!mounted microphone on the ~oor like a blind person|s cane ð13Ł[ The acousticsignatures from the impact allowed six ~oor types to be classi_ed with an accuracy of87)[

5[0[1[ The tactile inversion problemMany array sensors employ a protective covering made from an elastic material[ Thisgives mechanical compliance\ which assists the grasping process and increases therobustness of the device[ However\ this con_guration raises a serious di.culty\ knownas the inverse tactile transduction problem[ If an object is pressed into the surface ofan elastic layer then\ from analysis of the bulk material properties and the surfaceshape\ we can calculate the stresses that will be generated down at the sensing points*this is forward analysis[ In inverse analysis we wish to compute the changes onthe surface from the sensed data gathered remotely through the elastic medium[Unfortunately\ this does not give rise to a unique solution as there is no one!to!onecorrespondence between the stresses deep within an elastic material and those that


are applied normal to the surface[ In other words\ a given pattern of sensory valuesmay be caused by many di}erent patterns on the surface[ This is known as an ill!posed problem and can not be solved by direct analysis[ This particularly applies to_ne form discrimination and it can be seen that elastic materials act as a low pass_lter\ only transmitting large scale spatial patterns and attenuating any _ne detail[

A useful illustration of the inversion problem and its ill!posed nature is given inNowlin ð14Ł[

The tactile inversion problem has often been tackled by using a shaped membranefeaturing ribs\ tabs or other forms of raised surface nodules[ Yeung et al[ ð15Ł describethe use of this technique in enhancing the sharpness and resolution of tactile imagesas compared with uniform membrane surfaces[ They use a 05×05 sensing array\employing FSR resistive components[ By keeping the nodules in the compliant overlayentirely separate\ the compression of a nodule is made independent of the state ofother nodules and local crosstalk is reduced[ Another design using an elastic skin withraised tabs\ containing a force array and a dynamic sensor\ is described by Jockusch\Walter and Ritter ð16Ł[

An optical technique based on the birefringent generation of a phase!lead in cir!cularly polarised light in a photoelastic layer is proposed by Saad et al[ ð17Ł[ Thisrelies on the linear relationship between applied force and phase!lead and uses anoptimisation function to address the inverse tactile problem[ The method needs furthertesting and its computational complexity may prove to be a problem[

Another proposal involves an array of tensor cells sparsely embedded inside thevolume of the compliant sensing cover ð18Ł[ This aims to resolve all six axis stresscomponents in the sensing membrane and thus provide a di}erent solution to theproblem of tactile inversion[ However\ the results from a later design using ultrasonicresonant cavities ð29Ł suggest that the particular stress tensor used may give ambiguousoutput cases[

Many papers on array sensors with compliant membranes now recognise the tactileinversion problem and include some analysis or treatment[ Section 5[6[ gives furtherdetail on this topic[

5[1[ Sensin` _n`ers

One way of investigating the performance of tactile sensors is to construct anarti_cial sensate _nger than can be used to probe and explore the local environment[Researchers have usually designed such _ngers as cylindrical devices with the emphasison sensing on the hemispherical tip or over a curved surface region near the tip[ Darioð20Ł gives an example of a three!sensor _nger being used to probe and identifyunknown materials[

Problems with a sti} piezoresistive sensor for measuring grasped surface pro_leswere overcome by Russell and Parkinson ð21Ł who built a soft\ water!_lled _ngercovered in a neoprene skin[ Sensing was achieved by resistive measurement of thevolume of liquid between the intersections of an 7×09 grid of row and column wires[Another design using a sti} core\ but with a sponge body and rubber covering overFSR sensors was described by Borovac et al[ ð22Ł[ In a two _ngered experiment


the extra compliance assisted self accommodation during a successful peg!in!holeassembly task while any sensed errors were compensated by the control system[

In a comprehensive paper Howe and Cutkosky ð23Ł examine the problem of detect!ing _ne surface features and textures[ They use a stress rate sensor for intermediatefrequencies[

Studies of human _nger properties ð24Ł have shown a wide range of frictioncoe.cients[ Unlike rigid robot _ngers these coe.cients decrease with increasingnormal force[ Also friction during backwards movement of the _nger is larger thanfor forward movement\ probably due to sti}ness caused by the nail[

Shimojo and Ishikawa ð25Ł show how surface roughness can be measured by slidinga pressure sensing array over a surface and analysing the spatial frequency usingvariable _ltering techniques[

A general consensus is now discernible in that a general purpose sensing _nger willrequire at least two types of tactile sensor*one for contact point localisation and _neform spatial discrimination\ and one for detecting more spatially di}use dynamicevents such as contact slip[ This is in accord with our knowledge of human tactilereceptors\ where at least four di}erent types of receptor are tuned for di}erent spatialand temporal resolutions ð26Ł[ There seems to be an inverse relationship between thesei[e[ the better the frequency response of a receptor the worse its spatial resolution islikely to be\ and vice versa[ We see "in section 5[7[# that work on haptic perceptionshows that intrinsic force sensing is another complementary component of an e}ectivetactile system[ Hence intrinsic force and torque sensing is also likely to be includedfor kinaesthetic sensing of grasp or contact con_guration data[

See the _ngers of Taddeucci ð27Ł\ Howe ð23Ł and Dario ð28Ł for examples of suchcombined designs[ These papers also give the current state of the art[

5[2[ Soft materials for tactile sensin`

In the past\ most devices have relied on fairly rigid\ solid materials for theirconstruction\ including the all important contact surface[ Perhaps this was the naturalplace to start as rigid systems have less complexity and there are less variables tocontrol[ Following studies of human tactile performance and the physical nature ofthe tissues and skin\ it now seems that softer materials may have much to o}er[ Elasticoverlays and compliant contact surfaces are often advocated for their frictional andother properties\ although their low pass _ltering behaviour can be a disadvantage[But now even less rigid materials\ such as ~uids and powders\ are being examined[

Shimoga and Goldenberg ð39Ł have examined a range of materials with di}erentconsistencies and found that soft surfaces have more desirable characteristics forcontact surfaces than hard materials and that\ of the soft materials\ gels are betterthan plastics\ rubber\ sponge\ or paste\ with powders being the second best[ Thefactors considered included impact and strain energy dissipation\ conformability tosurfaces and hysteresis e}ects[

Sawahata\ Gong and Osada ð30Ł have described an interesting piezoelectric e}ect inpolymer gels[ A weak polyelectrolyte gel was shown to change pH when mechanicallycompressed[ The reverse e}ect also occurs*an applied potential causes the gel to


swell visibly[ Using polyacrylamide\ the authors constructed a simple tactile cell whichcaptured the electrical change and demonstrated a few millivolts being generated onloading[ The fact that human tissue is also composed of electrolytic materials withvery similar mechanical properties suggests intriguing possibilities for new designs ofsensing _ngers[

A di}erent use of gels involves electrorheological e}ects\ for example the applicationof a strong electric _eld across a suitable gel can change it from a ~uid to a plasticsolid[ Voyles\ Fedder and Khosla ð31Ł have designed a tactile actuator on this principletogether with a matching sensor[ The actuator:sensor pair have male:female symmetryfor the purpose of remote monitoring of touch sensing[ The _ngertip!shaped sensordetects contact events on its external surface using a gel layer as a dielectric incapacitive sensing\ while the similarly shaped actuator "or tactor# recreates theremotely sensed tactile events on its internal surface by changing the solidity of areasof the gel in contact with the human operator[ Taylor et al[ ð32Ł present experimentson a 4×4 tactor array using a rheological ~uid "silicone oil with 13) volume ofparticulate content#[ Such devices appear useful for tactile feedback applications buta serious drawback is the high voltages required "e[g[\ 1 kV#[

5[3[ Industrial robotics and `rippers

The emphasis on applications in tactile sensing has shifted away from the industrialarena[ Very few industrial or commercial devices are available and even less are inregular application in industry[

This category was intended for tactile sensing in robotic devices that had very strongexternal engineering requirements[ For example\ parallel action grippers\ poweredassembly tools and industrial machines of all kinds were included\ but dextrous robothands were not[ Several papers were found but these are not presented as they werevery specialised[ Of course\ there is a large literature on industrial robotics\ e[g[\grasping mechanics and kinematics\ but it contains little tactile material[

The criteria for sensors to be acceptable in robotic applications have been listedas robustness\ reliability\ accuracy and high resolution[ It seems likely that someapplications would have tolerated lower levels of accuracy and resolution\ and so weconclude that the more important requirements of robustness and reliability couldnot be satis_ed or demonstrated at suitable levels and at the right economic price forindustrial acceptance[

It may be surprising to _nd so little work on tactile sensing in industrial roboticsand automation*considering this had been promoted as a major future applicationarea[ The reason for the shift away from the industrial arena\ and towards other moreunstructured domains\ may be more to do with the strength of application pull andthe failure of technology push[ Many papers have justi_ed their designs and resultsby referring to various suggested criteria and speci_cations for desirable devices\ e[g[\that of Harmon ð3Ł[ However\ although widely quoted\ these speci_cations wereessentially speculative[ They only have meaning if the task to be automated is bestdone with the technology proposed in the way envisaged[ It is important to assess therequirements speci_cation of any sensing task in general terms without promoting


any particular technology[ Thus industrial tasks with apparent tactile sensing needsmay turn out to be better served by other methods[ Technical superiority is only onedimension^ managers might decide that vision or {spatial switches| provide solutionsthat are better in terms of cost\ minimising disruption\ recon_guration\ retraining\etc[

5[4[ Multi_n`ered hands for dextrous manipulation

Robotics has always had a strong interest in the design of arti_cial hands but theseare complex and expensive to develop[ Consequently only a few di}erent designs havebeen reported ð33Ł\ most consisting of three of more multi!jointed _ngers in anarrangement similar to the human hand[ Despite the control challenge posed by themany degrees of freedom in a fully actuated version\ the appeal of anthropomorphicdesigns is in the promise of reproducing manipulation dexterity approaching that ofhumans[

Tactile sensors have been less prominent until recently in this area but Johnston etal[ ð34Ł report on a comprehensive suite of sensors designed to _t all the _nger segmentsand palm of an established multi_ngered robot hand[ The sensing method is capacitiveand there are 02 separate pads\ containing arrays on 1[66 mm centres\ giving a totalof 633 sensing sites[ The scanning rate is 06 k tactels per s\ and the authors claim thatthe low sensor mass and high damping ratio allow the system to be used in transientforce control[ This system is notable as a commercially available product[

A strong advocate of multi_ngered robot grippers for dexterous manipulation is thegroup of Howe and his co!workers[ Howe ð35Ł argues that e}ective\ high performancedexterous manipulation must involve fast real!time processing of information relevantto the di}erent phases of grasping tasks as situations change from position to forcecontrol\ from rolling to slipping or sliding\ and power grasping[

Unfortunately\ we do not yet know enough about the most suitable or even necess!ary information for any given manipulation task[ For example\ to detect contactwe may have available a range of di}erent sensing techniques all with di}erentcharacteristics but selecting between these for a given task is a di.cult problem[ Howestresses that static arrays have limited applicability and that the essential aspect formanipulating objects in real!time is smooth and accurate _ne control[

The need for real!time control and dynamic sensing for rapid detection of contactevents raises various issues\ as described by Howe ð35Ł[ Some of these issues are dealtwith by Son and Howe ð36Ł in which contact events\ the contact shape and incipientslip are all handled using a stress rate sensor that can detect transients[ This sensoruses piezoelectric polymers and a ribbed structure for localising contact events[

Son and Howe ð36Ł examine the problem of uncertainty in orientation of graspedobjects*for an object held between two cylindrical _ngertips\ the error in poise angledue to friction will typically be around 04>[ Using an 7×7 capacitive array\ tactileinformation was used to correct alignment errors in a pig!in!hole insertion task[ Sonand Howe show how tactile sensing can reduce kinematic errors in sti}ness control


by locating precise contact points and tracking their changes[ However\ the methodrelies on assumptions about the shape of the grasped object[

Another group that have extensively explored multi_ngered systems is that ofMaekawa\ Tanie\ Komoriya and colleagues[ These researchers have exploited theoptical transduction method that they and the authors simultaneously discovered in0873[ In ð37\ 38Ł they describe the novel innovation of reshaping the internal re~ectionoptical method into a hemispherical design "an optical waveguide#[ This has theadvantage of giving the exact location of a contact point on a _ngertip "in polarcoordinates# and also\ by using a preformed elastic cover\ solves the problem ofsticking that occurs between the rubber membrane and the optical surface[ Theexperiments report a fast response time of 9[967 s[ Such sensing ability allows contactpoints to be tracked during rolling manipulations\ as described in ð49\ 40Ł[ Tactilefeedback is used dynamically to keep the grasp forces within the friction cone thusgiving stable real!time grasp control[ These results are very similar to those from theSon and Howe experiments[ In ð41Ł the data derived from tracking contact duringrolling is used to estimate the curvature of the object[

5[5[ Probes and whiskers

In the industrial manufacturing arena\ touch probes have been developed to a veryhigh degree of sophistication[ Coordinate Measuring Machines are extremely accuratesystems for recording the pro_les and spatial geometry of object surfaces[ Theseexpensive systems use sophisticated binary touch sensors combined with high precisionspatial actuators and are quite di}erent in purpose and design from the tactile sensingissues addressed in this survey[ Nevertheless\ the idea of sensing objects by pointcontact with a touch probe rather than by measuring areas of contact o}ers alternativepossibilities for tactile sensing[ Instead of {spatial switches|\ however\ we can use aprobe as an active wand thus gaining much more spatial information[ Tsujimura andYabuta ð42Ł have used knowledge of the de~ection characteristics of a ~exible probeand the force:torque values experienced at its base to compute the position of thecontact point[ In order to disambiguate false contact cases\ Russell ð43Ł added switchesto a similar sensor to detect contact at the tip and bending at the root[

Such whisker probes have the potential to be fast\ accurate and cheap[ They areessentially single point sensors and therefore are local and have low bandwidth\ butthey may o}er interesting possibilities such as being disposable[ In a series of papers\Kaneko and colleagues have argued for the utility of active whiskers by analogy withinsect antenna and feline whiskers[ In the experiments described ð44Ł a _ne whisker ismoved by a specially designed actuator drive that incorporates torque and positionsensing[ With a simple sweep action the contact location is estimated from knowledgeof the compliance of the whisker and the assumption that all objects to be detectedare rigid[ For simple bending the contact point is determined by equations with twounknowns and the values of torque and de~ection at the root allow these to be solved[Work on dynamic analysis ð45Ł shows the e}ects of changes in natural frequency aftercontact and a 2D version ð46Ł exposes the need to deal with lateral slip[ In order toremove the assumption of a sti} environment Kaneko et al[ ð47Ł have added a vision


sensor to discriminate bending in the whisker from compliance in the objects[ Thereseems to be little work in this area as all the Kaneko papers reproduce the relatedwork section from Russell ð43Ł[

5[6[ Analytical Pro`ress

While the 0879s can be seen as a period of exploration that proposed a wide rangeof novel sensing methods\ more recent work shows a concern for analysis and scienti_cunderstanding of the role and process of tactile sensing[

As mentioned above\ for unknown objects in unstructured manipulation tasks real!time sensing of contact location is necessary[ Son\ Cutkosky and Howe ð48Ł havecompared the use of tactile arrays and force!torque sensing for locating contact pointsduring gripper:object manipulation tasks[ Their experimental study showed that bothintrinsic and extrinsic methods could be e}ective with less than 0 mm error in contactlocation[

Ellis\ Ganeshan and Lederman ð59Ł have taken a rigorous approach towards design!ing a sensor for a given task and analysed the requirements for sensing the inertialproperties of a grasped object[ They rejected sensors with compliant surfaces "becausethey lose too much detailed strain information# and through careful analysis produceda strain gauge based design that is rugged and has high accuracy[

A number of investigators have used _nite!element "FE# models for the analysis ofmaterial deformation and stress:strain relations in tactile devices\ e[g[\ for a detailedFE analysis of a sphere in contact with a linear elastic plane see ð50Ł[ Ohka et al[ ð51Łdescribe a design based on the optical waveguide principle using large column shapedtabs on a silicon rubber sheet[ The tabs can be compressed or displaced along threedirections and smaller cones on the underside of the sheet are distorted in a way thatre~ects the contact forces[ A detailed FE mesh model was used to analyse how thematerials would deform[

The inverse tactile transduction problem was examined by Ricker and Ellis ð52Łwho used FE analysis as an alternative to the classical solid mechanics approach withlinear elastic half!spaces[ They showed how di}erent contact shapes can producesimilar sub!surface strain pro_les and how shear strain can complement the normalstrain _eld[ They argue that classical analysis may work for planar sensors but isinappropriate for sensors with _nger shaped geometries[ Ellis and Qin ð53Ł continuethe critique of previous analyses and are {not optimistic| about sensing _ne formgeometry from strain data within a rubber!like sensor based on traditional solidmechanics methods[ The radius of the contact area is the only feature they feel canbe detected with con_dence[

Nicholson and Fearing ð54Ł have shown that methods for deriving object curvatureare sensitive to both calibration techniques and contact models and suggest sensorconstruction aspects determine the relevance of the analysis and processing method[Another study of the low!pass _ltering e}ect of compliant tactile surfaces\ again usingFE methods\ is described by Shimojo ð55Ł[ Results show the reduction in gain withincreasing thickness and changes in material properties of an elastic cover[ An alter!native approach to the inversion problem is to develop a forward model in which


changes in tactile images are related to changes in position[ Chen\ Zhang and Rinkð56Ł have produced a tactile Jacobian that controls a tactile!servo for surface following[The Jacobian has to be constructed piecemeal beforehand and stored in a look!uptable[

Slip detection is an important tactile sensing topic that has also been analysed indetail[ Holweg et al[ ð57Ł describe slip as having four stages] pre!slip tensioning\ start\post!slip movement\ and stop[ They report on a joint investigation into two methodsusing a rubber!covered 05×05 resistive array[ The two methods are complementary\one dealing with the onset of slip and the other detecting movement[ Their _rstmethod used a FFT to analyse the shifting of the centre of pressure distribution thatoccurs just before slip starts[ The low frequency components give a clear indicationof incipient slip[ In the second method\ a power spectrum density was taken to detectthe 54 Hz signal generated during slip by the {catch and snap| behaviour of the rubberat the contact interface[

Incipient slip signals are detected by humans when localised slipping occurs justbefore gross sliding at the contact surfaces[ To sense this e}ect\ Tremblay and Cut!kosky ð58Ł have used skin acceleration sensors embedded in a foam _nger[ A siliconeskin covers the foam and is covered in small protrusions that produce small vibrationswhen they slip[ The system was able to detect incipient slip\ compute the frictioncoe.cient and adjust the applied grasp force to compensate[

Son et al[ ð69Ł comment that acceleration sensors do not give much spatial local!isation and have developed a stress!rate sensor using PVF1 piezoelectric _lm[ Thiswas able to detect local skin curvature and some _ne surface detail[ They suggest thatarray sensing is best used for spatial frequencies in the range 9Ð09 Hz\ stress!ratesensors are best for 09Ð49 Hz and skin acceleration sensors are e}ective in the range49 HzÐ0 kHz[ The importance of incipient slip tactile data for humans has beencon_rmed by telemanipulation experiments with a master:slave system ð60Ł[ Two_ngered pinch grasps were performed by human subjects and dynamic adaptationand recovery from slip was clearly demonstrated over a remote link[

5[7[ Haptic perception\ telepresence and virtual reality

Most early work dealt with static tactile images or contact events[ It is usefulto remember the di}erent roles of non!static sensing in order to distinguish theirrequirements and purpose[ We de_ne Active Sensin` as a dynamic form of tactilesensing where relative movement of the sensor is involved[ For example\ rollingcontact can be used to gain shape information or squeezing might be used to estimateelasticity properties[ The term Haptic Perception refers to the integration of cutaneoussurface sensing with kinaesthetic data derived from the position and movementvariables of the manipulator system[ Teleoperation and telepresence are concernedwith the remote human operation of a handling system and the authentic feedbackof a range of sensory modalities[ Finally\ the idea of Virtual Reality "VR# deals withthe creation of synthetic sensory experiences for realistic simulation scenarios[

Active sensing is now well represented in the literature and is featured in many ofthe papers discussed in this survey[ Haptic Perception emerged in the 0889s and is now


accepted as a major approach in tactile sensing work[ Teleoperation and telepresencesystems must directly confront the problems of sensing\ encoding and reproducingtactile sensations and so will be of interest here[ VR does not involve any remoteactuation or sensing other than the delivery of relevant stimuli to the participant[However\ some of the features of VR technology are closely related to telepresenceand we will include some relevant work[

Examples of active sensing schemes are seen in studies of tactile tracking\ where arobot must trace out the shape of an unknown surface\ ideally in real!time\ bymaintaining contact between the surface and a tactile sensor[ Chen\ Rink and Zhangð61Ł provide a mathematical surface notation that links contact constraints and kin!ematics\ and then derive a tactile driven control scheme that performs surface tracking[Another approach to curvature tracking is to use one or more _ngers of a multi_ngeredhand^ Charlebois et al[ ð62Ł use surface normals from surface tracing experiments torecognise object shape using the idea of Lederman|s Exploratory Procedures[ Otherswho have experimented with tactile procedures include Buttazzo\ Bicchi and Darioð63Ł\ who used three separate routines to deduce hardness\ texture density and frictioncoe.cients[

Haptic Perception is concerned with relating and reconciling local tactile infor!mation to the global spatial structure of the objects in the sensed environment[ Thismeans that tactile data is necessarily sparse and has to be integrated via some formof kinaesthetic spatial framework[ Caselli et al[ ð64Ł explore the use of two kinds ofpolyhedral volumetric shells for building up models of unknown objects duringexploratory action[ This work assumes that the objects are convex and are larger thanthe sensor contact areas[ An example of the converse case where the object surfacesare smaller than the sensor surfaces is seen in Kinoshita ð65Ł\ where a lattice of objectfeatures in terms of points\ edges and planes can be constructed[ In both cases thespatial structure of the objects provide the linkage between di}erent local tactile data^a linkage that can be seen as characteristic of the encountered objects or as a map ofkinaesthetic exploration[ "We notice that concave surfaces can be explored usingtactile probes ð66Ł[# Beccari et al[ ð67Ł continue to explore sparse tactile data and showhow the pose of objects can be recovered using principal axes analysis with 24 contactsbeing su.cient for pose!independent recognition of any one of 19 objects[ They alsodiscuss the di}erences between tactile and visual data and argue that feature basedmethods are not appropriate[ For further background on the haptic aspects of tactilesensing see ð2Ł\ and for the current state of the art of haptic exploration with dextrousrobot hands see ð68Ł[

A telepresence system is normally arranged to transmit some desired operatorforces and movements to a slave system which will then re~ect back to the operatorthe reactions and sensed events[ This means that both tactile sensing and display arerequired and\ ideally\ the system mechanisms must be structurally transparent\ i[e[\the only forces and sensory data experienced by the operator should be those generatedsolely from the application and not from the system itself[

Kontarinis et al[ ð79Ł report on an experiment with identical two!_ngered masterand slave manipulators _tted with force and tactile feedback[ The tactile system usedthe Fearing design ð70Ł of an 7×7 capacitive sensing array on 1 mm spacing and a


tactile display consisting of an array "5×3# of pin actuators "shape memory alloy#arranged under the operators _ngertip[ In a simulated medical trial subjects wereasked to locate a hard rubber cylinder inside a block of foam rubber[ The error wasnot more than 0 mm for 49) of the trials but without tactile feedback the mean errorwas over 02 mm[ These results con_rm the conjecture that force re~ection alone isnot su.cient for _ne form sensing[

Another experiment\ Cohn\ Lam and Fearing ð71Ł used a tactile sensor array anddisplay device to evaluate the transmission of tactile force\ pattern and displacement[Displacements as low as 9[0 mm were detected by human subjects[ Fearing\ Moy andTan ð72Ł discuss the problems of spatial detail and aliasing in tactile displays andshow why an elastic "low pass# layer is necessary between human _nger and thetactors[ A novel _nger tracking device is described by Yoshikawa and Nagura ð73Ł inwhich contact is correlated with events in a virtual environment[ As the user touchesvirtual objects so the tracker makes contact and applies resistance[

Caldwell et al[ ð74Ł describe experiments with a multi!sensor glove for feedback ofa range of tactile signals[ Taking inspiration from the human sensory nervous systemthey designed a glove that could generate sensations of contact force\ _ne form surfacedetail and thermal conduction[ The actuators used were piezoelectric pulse generator"high frequencies for textures#\ piezoelectric bi!morph "medium frequencies for edges#\air bladders "low frequencies for contact forces#\ and Peltier e}ect heat pump forthermal e}ects[ Shimojo et al[ ð75Ł also use a glove with a comprehensive set of padsthat sense the loading on _nger sections and palm using conductive rubber methods[The experiments used a range of instruments to measure grasping posture\ wristposition\ applied forces and to capture video images during task performance[

An interesting way of controlling the sensed roughness of a surface is described byWatanabe and Fukui ð76Ł[ The application of ultrasonic vibration to a solid surfacereduces the perception of high frequency spatial features\ thus subjects reported arough rusty surface as {smooth| when vibrating at 64 kHz[ The e}ect begins atabout 09 kHz and saturates at around 19 kHz[ Conversely\ an increase in perceivedroughness was reported by subjects when short bursts of vibration "duration 09 ms#were applied[ The amplitude modulation of the burst seems to determine the perceivedsurface structure and even {virtual protrusions| were reported[ The mechanism respon!sible may be simulating the {catch and snap| phenomenon of sliding friction withelastic surfaces[ The idea of a controllable friction:texture surface seems to o}erexciting possibilities for tactile feedback[

Many authors in the _eld of tactile sensing for Haptic Perception\ Telepresenceand Virtual Reality refer to the psychophysical and physiological literature to gaininsight\ understanding and support for new developments[ For example\ visual infor!mation may need to be coordinated with haptic information in order to resolveredundant or complimentary data[ In this case\ results suggest that vision is usuallythe dominant sense when con~icts arise[

Wu and colleagues ð77Ł have studied human perception of length and curvature ofobjects using vision and tactile sensing[ The results showed how humans overestimateand underestimate parameters during sensory fusion situations and suggest how thesemight bene_t the design of virtual sensing systems[


Another consideration is the role of dynamic contact sensing\ that is\ the detectionof grasped object properties such as size\ mass and moments of inertia by sensingtheir dynamic in~uence on the system[ With so much still unknown\ much theorisingand debate is raised in this area but nevertheless valuable ideas are gained by studyingthe human tactile system[ An example is seen in the discussions about the perceptionof the length of a statically grasped rod^ Lederman\ Ganeshan and Ellis ð78Ł argue indetail that teleoperation systems "handling long objects such as rods# will need tomodify the force and torque feedback to take account of human bias in weight andlength perception[ This work also derived the design of a tactile sensor speci_cally fordynamic sensing\ described in ð59Ł[

5[8[ Tactile data processin`

As any sensing technology matures\ we should expect a section of the literature tofocus on the processing of the data generated[ This involves not just signal processingbut interpretation and correlation of the signals in terms of the semantics of theapplication[ As we have seen\ areas where considerable processing could be requiredinclude the tactile inversion problem and the extraction of shape and curvature[ Thetechniques that are currently popular for such tasks in similar _elds are] neuralnetworks\ fuzzy logic methods\ rule!based systems and model!based systems[ Thesecan all be found in the tactile sensing literature[

A fuzzy logic method for discriminating material types from tactile data is describedin Li and Shida ð89Ł[ Tactile sensors of four types "thermal\ inductive\ capacitive\optical re~ectance# were used to collect data on seven di}erent materials "wood\rubber\ styrene and four metals#[ Although high discrimination rates were achievedthis work is at an early stage and the results were not conclusive[

Other fuzzy logic methods have been tried\ as well as a few rule!based systems\ butthe most popular processing technique appears to be neural networks[ Part of theattraction is the speed of parallel nets that will be valuable for real!time sensing[Neural processing methods have been reported for learning parameters for unknownsurface shapes ð80Ł and for detecting discontinuities and other defects in orientedtextured surfaces ð81Ł[ Pati et al[ ð82Ł describe an approach to the tactile inversionproblem which uses an analogue neural network for deconvolution[ Further devel!opment\ to consider stress tensors rather than just normal forces\ is given by Caiti\Canepa and De Rossi ð83Ł who use regularization theory and experiments with twotypes of network "back!propagation and radial basis functions#[ Another way ofdealing with the tactile inversion problem is to incorporate additional information[Motor actions were used by Rucci and Dario ð84Ł in a neural network to correlatetactile changes with small motor movements while maintaining tactile contact[ In afurther development Rucci and Bajcsy ð85Ł show how visual data can be coordinatedwith tactile data as a step towards a general framework for integrating multimodalattention stimuli[

A major concern in sensor data processing is the problem of merging or fusing datafrom more than one sensor or even from sensors of markedly di}erent characteristics[


This is di.cult because the data from physical sensors are rarely easy to reconcile\ andnoise and distortion must be separated from genuine di}erences in sensed variables[

Yamada\ Ishiguro and Uchikawa ð86Ł use a model!based method for fusing sensordata from vision and tactile sources but the tactile data is treated as a second stagere_nement of a coarse parametric model[

Joshi and Sanderson ð87Ł contrast homoèneous sensor data\ where all data comesfrom the same type of sensor and the fusion can take place at the si`nal level\ andheteroèneous sensor data\ where di}erent sensing modalities may be employed andthen fusion must take place at the feature level[ Extending their earlier work on singlesensor model matching ð88Ł\ Joshi and Sanderson have produced a general model!based fusion method based on information!theoretic criteria[ A worked example onfusing tactile and visual data is given[

This approach automatically weighs the priority of sensor data in terms of sensoraccuracy\ precision and resolution and has good rejection of outliers and appears tobe general and very robust[ See ð099Ł for improvements and further analysis[

5[09[ New application areas

During the 0879s\ most authors predicted that the major application area for tactilesensing would be industrial automation tasks such as robotic assembly[ Severalcommercial tactile devices were marketed\ clearly targeted at this area[ However\ thedemand for such sensors proved to be low\ as did the numbers sold[ Commercialforce sensing robot wrists are still much more numerous that tactile systems[ Thereare still very few fully developed applications of tactile sensing but this sectionconsiders current trends and estimates the potential role of this technology in thefuture[

We see three main _elds where tactile sensing is likely to play a key role[ Theseare] medical procedures\ especially surgery^ rehabilitation and service robotics^ andagriculture and food processing[ We consider these in turn[

5[09[0[ Surìcal applicationsSurgery is perhaps the most exciting and rapidly developing area where tactile sensingis actually of central importance[ Minimally Invasive Surgery "MIS# is only 09 yearsold and yet is now routinely used as the preferred choice for many operations[However\ despite its advantages\ MIS severely reduces the surgeon|s sensory per!ception during manipulation[ Surgery is essentially a visual and tactile experience andany limitations on the surgeon|s sensory abilities are most undesirable[

For example\ in laparoscopy long slender tools are inserted through small punctureopenings in the abdominal wall and the surgeon uses a range of tip mounted instru!ments guided by video feedback images[ As the instruments are rigid rods ande}ectively have _xed pivots at the entry points\ the available degrees of freedom arerestricted and therefore demand extra operator expertise[ Voges ð090Ł reports themain di.culties experienced are] restricted manipulation mobility\ lack of depth


from 1D vision and the almost complete lack of a sense of touch[ The relevance oftelepresence is clear and Voges predicts that future systems will have new designs of~exible instruments with greater mobility\ force re~ection\ 2D monitoring and dataenhancement[ A table of contrasting features for current and future laparoscopicsystems is also given[ It is clear that tactile sensing is greatly needed in this area andresearchers are responding to the opportunity[

The reason that tactile sensing is so important in surgery is that soft tissue can onlybe properly examined and identi_ed by assessing its softness\ viscosity and elasticityproperties[ The palpation of tissues and organs is an essential procedure that surgeonsvalue highly[ Indeed\ surgeons have been known to insert their _ngers through theaccess openings during MIS simply to perform direct tactile exploration ð71Ł[ Dario|sshort but far!sighted review ð28Ł cites medical applications in which the hardness ofsoft tissues is detected through palpation[ Howe et al[ ð091Ł discuss the topic of remotepalpation technology[

Bicchi et al[ ð092Ł gives a good description of the issues in MIS and describes anexperiment with a commercial instrument modi_ed to sense force "by strain gauge#and position "by LED and optical detector#[ By correlating force against deformationthe system was able to identify _ve objects of di}erent elastic properties[

An experiment with a sensor for laparoscopic attachment has been described byFischer et al[ ð093Ł[ A 53 point sensor of area 0 cm1 was connected to a _ngertipvibrotaction display[

The medical sensor described in section 5[0[ has been used in sixteen cases to localisepulmonary modules ð094Ł and for measuring heart wall sti}ness ð095Ł[ The ultrasonictactile method is entirely non!destructive in its use as a sensor and therefore is muchsafer than other methods using needles[ A force sensing catheter tip with a diameterof only 0[5 mm has been built\ using an IC pressure sensor\ by Tanimoto et al[ ð096Ł[The sensor can be used inside blood vessels and allows the surgeon to distinguishfrictional resistance from direct pressure on vessel walls at junctions and bends[ Amicrorobot designed for colonoscopy and using a pneumatic {inchworm| propulsionmethod is described by Dario et al[ ð097Ł[

The di.culties of adopting fully autonomous robotic systems in surgery are dis!cussed by Ho et al[ ð098Ł and an approach is developed where the surgeon maintainssupervision and control but is constrained from driving the cutting tools outside forcelimited regions[

A series of designs for endoscopic and laparoscopic tools are discussed by Cohn etal[ ð009Ł who intend incorporating their tactile telepresence apparatus ð71Ł[ An inter!esting idea raised here is the possibility of using the capacitive tactile sensor\ not tomeasure applied force\ but to detect the varying dielectric permittivity of di}erenttissue[ Cohn et al[ suggest that water\ fat\ blood vessels\ and cancerous tissue mightall be discriminated by this means[

An interesting sensor has been developed for orthodontics[ Umemori et al[ ð000Łstate that it proves very di.cult to measure pressure distributions between the lipsand cite over 19 measuring devices that have been assessed but have proved unsuitable[In order to assess changes after surgery\ they developed a sensor\ based on the opticalwaveguide principle\ that can measure lip sealing forces\ contact areas and pressure


distributions[ A striking feature of the design is that it uses disposable cartridges] byselecting inexpensive materials "OHP _lm\ silicone rubber sheet\ photographic _lm#the sensing head could be replaced for each patient[

Such attention to packaging could well be very signi_cant for the success of medicalsensors[

5[09[1[ Rehabilitation and service roboticsA major concern for the next century is the enormous numerical increase in the elderlypopulation that will generate great economic pressures\ especially in the developedcountries[ This demographic change is well accepted and many governments haveinitiated programmes of research in health care\ hospital services and social support[It is clear that there will be greatly increased demand on these services and researchersare looking for methods of support and assistance that do not involve central servicesbut can be distributed as aids within the home and community[ Robotics can play amajor role here\ but much of what we have learned may not be readily applicable[This is the opposite end of the robotics spectrum from the industrial arena and isconcerned with very unstructured and irregular environments and focuses on themajor issues of human compatibility\ safety and reliability[ Interestingly\ issues suchas accuracy and high resolution are of much less importance[

Dario et al[ ð001Ł discuss these issues and describe three current investigations intoautomated aids for the disabled in hospital and the home involving manipulators andmobility systems[ In order to be accepted and operated by the elderly or in_rm suchaids must extend the very limits of user friendliness^ as Dario states ð002Ł\ interactionswith the systems must be assessed in terms such as {satisfaction and pleasure| and {theuser|s feelings toward the system ðmust beŁ kept in great consideration|[ An exampletask for a home service robot is] take a dish of food from a refrigerator to a microwavecooker and\ after heating\ serve to bed!ridden user ð002Ł[ Another human!factorsaspect is the physical appearance^ it may be necessary for the system to have someanthropomorphic characteristics in order to gain acceptance[

Despite the contrast with industrial robotics\ there are some emerging interests inhuman!robot cooperation within manufacturing[ As the next stage in automation\Toyota envisage workers and robot machines coexisting in a {safe partner| relation!ship[ Tobita et al[ ð003Ł stresses the need to make automation more attractive topeople and describes ways of dealing with the safety issue[ The conventions andstandards on robot safety state _rmly that all active robots must be isolated fromhuman contact[ However\ if this restraint is to be lifted there are two ways of providingsafety] either very comprehensive collision avoidance and fail!safe features must bebuilt in\ or the equipment could be designed to be intrinsically safe[ Tobita et al[ takethe latter approach and design low power\ reduced weight devices with many safetyfeatures[ This is rather like producing smarter and more useful power tools for humanworkers[ The other approach is seen in Suita et al[ ð004Ł who assume that collisionsmight occur and design the robot systems with compliant coverings and collisiondetection and avoidance mechanisms so that any contact will be well below thethresholds of human pain or damage[ Ikeura and Inooka ð005Ł show the advantages


of variable impedance "damping# control schemes for robots that follow humantrajectories in cooperative lifting tasks[

As yet there is little in this area that is new speci_cally concerning tactile sensing[However\ as for medicine\ we can see many opportunities where a sense of touch willbe a real need[ An example is seen in the robot ~oor cleaner of Ulrich\ Mondada andNicoud ð006Ł which needs to sense the shapes of furniture and other objects in orderto clean round them[ This is a very new _eld and is likely to show signi_cant growthin the near future*already new journals\ such as Service Robot ð007Ł are beginningto emerge[ Funding from major sources include the TIDE "Technology Initiative forDisabled and Elderly people# programme of the European Commission and the RealWorld Computing Programme in Japan[ The RWC programme is aimed at developing~exible human assistants that have intuitive abilities\ i[e[\ {Interactive computers usingspeech\ facial expression\ and gesture^ systems that infer user|s intentions\ and robotsthat can work with humans by learning by example{ ð008Ł[ For a view of health carerobotics in the U[S[A[\ see ð019Ł[

5[09[2[ A`riculture and food processin`The _eld of agriculture and food production is now well automated but\ like servicerobotics\ also does not feature many new technological advances in tactile sensing[Uses of tactile sensing are often mundane "e[g[\ binary switches are su.cient andsatisfactory to sense crop height in an automated gain harvester ð010Ł#\ but theimportance of this area is its potential for imminent development[ Unlike manu!facturing automation\ the processing of natural produce usually involves high num!bers of human operators[ This is because the problems of handling soft\ delicate andhighly variable items by machine have not yet been solved at low enough costs[Various methods for inspection have been developed but the handling and assemblyof food produce remains as one of the last bottlenecks to remain largely un!automatedin the highly organised food and agriculture industries[ Recently\ there has beenincreased interest in the prospect of reducing human involvement in order to reducehygiene risks\ eliminate human errors and use e.cient but more hazardous environ!ments "e[g[\ low temperature factories#[

Examples in this _eld are very speci_c as a few cases illustrate[ Stone and Brettð011Ł describe a design of intrinsic sensor for monitoring gripping forces and slipduring the handling of dough!like materials[ Dough gives an example of the require!ments for gripping soft\ compact objects and relates to soft fruit and confectionaryitems[ Asparagus is another delicate product and a pneumatic gripper with _ne forcefeedback and a fuzzy controller is described by Mattiazzo et al[ ð012Ł[ Among manyother studies of robot grippers for soft objects are the investigations of Naghdy andEsmaili ð013Ł in Australia and Friedrich in New Zealand[

For example\ Friedrich and Lim ð014Ł combine research into novel sensory grippersand control modules[ The intended application areas are the handling of soft\ naturalproducts such as those commonly found in the food industry[ They have developeda gripper with hardness evaluation as an integral part of the gripping process and avariety of gripping strategies including active control of slip during object handling[


Despite a long history of gripper design it seems that the time is now right for theintegration of our knowledge of force\ tactile and spatial sensing with actuation\material design and sophisticated control\ to produce some smart soft!product hand!ling systems with abilities comparable with human operators[

6[ Conclusions

A number of observations can now be made]

, In tactile device design\ a shift of interest is discernible away from novel transductiontechnology and towards the engineering of sensors[ Thus\ there are now morecon_gurations that give better reliability\ robustness\ durability and overload tol!erance[ Packaging is important with considerations of ~exibility\ size and dis!posability all in evidence[

, The previously wide range of transduction methods seems to have settled down withthe predominant choices being piezoelectric methods for many devices\ with arraysusing resistive or capacitive sensing[ Strain gauges are still the preferred choice forforce sensing[ Spatial resolutions have increased considerably and many fabricatedIC devices have been designed[ New advances have also been made in sensing _neform texture and properties such as softness of materials[

, In our classi_cation scheme\ the areas which had grown the most in terms ofnumbers of publications were] analysis and evaluation\ methods of processing tactiledata\ and medical applications[ Little new work was found on tactile sensing inindustrial robotics and automation[

, On the analysis side\ much attention has been directed at the inversion problem andthe dynamics of slip[ Also\ in manipulation we see much progress] {the state of theart in tactile sensing and dextrous manipulation planning control are reaching thepoint at which autonomous haptic exploration becomes feasible| "Okamura\ Turner+ Cutkosky ð68Ł#[

, The _eld is maturing and we see a deeper understanding of human tactile sensing\the physical properties of skin and the recognition that several types of sensor mustbe involved[ Whilst not promoting an anthropomorphic approach\ the improvedunderstanding of human systems enables a more integrated approach to be adoptedfor tactile sensing in mechatronic systems[ In an arti_cial manipulation system\ wewould now expect to _nd at least three kinds of sensing being performed] intrinsicsensing of internal forces and torques^ shape and contact location sensing usingarrays^ and dynamic sensors for slip and transient event sensing[

In summary\ this survey has found increased emphasis and understanding of devicematerials\ tactile sensing processes and requirements^ the recognition of the valueof results from dextrous manipulation and telepresence systems^ and an emergingengineering approach to sensor packaging[ All these\ together with the applicationpull from medicine and service industries\ bode very well for the next stage in thedevelopment of this fascinating _eld[


References

ð0Ł Harmon LD[ Tactile sensing for robots[ In] Davidson HF\ Brady M\ Gerhardt LA\ editors[ Roboticsand Arti_cial Intelligence "NATO ASI series#[ Springer!Verlag\ New York\ 0873[ pp[ 098Ð47[

ð1Ł Nicholls HR\ Lee MH[ A survey of robot tactile sensing technology[ Int[ J[ Robotics Research0878^7"2#]2Ð29[

ð2Ł Nicholls HR\ editor[ Advanced Tactile Sensing for Robotics[ World Scienti_c\ Singapore\ 0881[ð3Ł Harmon LD[ Automated tactile sensing[ Int[ J[ Robotics Research 0871^0"1#]2Ð20[ð4Ł Russell RA[ Robot Tactile Sensing[ Prentice!Hall\ Englewood Cli}s\ NJ\ 0889[ð5Ł Cicchetti A\ Eusebi A\ Melchiorri C\ Vassura G[ Intrinsic tactile sensor for robotic manipulation[

In] 6th Int[ Conf[ on Advanced Robotics\ Catalonia\ Spain\ 19Ð11 Sept[ 0884[ pp[ 778Ð83[ð6Ł Beebe DJ\ Hsieh AS\ Denton DD\ Radwin RG[ A silicon force sensor for robotics and medicine[

Sensors and Actuators A*Phys[ 0884^49]44Ð54[ð7Ł Wol}enbuttel MR\ Regtien PPL[ Polysilicon bridges for the realization of tactile sensors[ Sensors

and Actuators AÐPhys[ 0880^15]146Ð53[ð8Ł Yamada Y\ Cutkosky MR[ Tactile sensor with three!axis force and vibration sensing functions and

its application to detect rotational slip[ In] IEEE Int[ Conf[ on Robotics and Automation\ San Diego\CA\ 7!02 May 0883[ IEEE\ Robot + Automat Soc\ 0883[ pp[ 2449Ð6[

ð09Ł Kolesar ES\ Dyson CS[ Object imaging with a piezoelectric robotic tactile sensor[ J[ of Mic!roelectromechanical Systems 0884^3]76Ð85[

ð00Ł Gray BL\ Fearing RS[ A surface micromachined microtactile sensor array[ In] IEEE Int[ Conf[ onRobotics and Automation\ Minneapolis\ Mn\ 11Ð17 April 0885[ IEEE\ Robot + Automat Soc\ 0885[pp[ 0Ð5[

ð01Ł Omata S\ Terunuma Y[ New tactile sensor like the human hand and its applications[ Sensors andActuators A*Phys[ 0881^24]8Ð04[

ð02Ł Li DS\ Shida K[ Monostructure touch sensor with multifunction for discrimination of materialproperties[ Electrical Engineering in Japan 0885^006]57Ð64[

ð03Ł Monkman GJ\ Taylor PM[ Thermal tactile sensing[ IEEE Trans[ on Robotics and Automation0882^8]202Ð7[

ð04Ł Inaba M\ Hoshino Y\ Nagasaka K\ Ninomiya T\ Kagami S\ Inoue H[ A full!body tactile sensor suitusing electrically conductive fabric and strings[ In] "IROS 85# Proceedings of the IEEE:RSJ Int[Conf[ on Inteligent Robots and Systems\ Osaka\ Japan\ 3Ð7 Nov 0885[ IEEE\ 0885[ pp[ 349Ð46[

ð05Ł F!SCAN System[ Tekscan Inc[\ MA\ U[S[A[\ 0886[ð06Ł Donaghue VM\ Veves A[ Foot pressure measurement[ Orthopaedic Physical Therapy Clinics of

North America 0886^5]0]0498Ð405[ð07Ł Yamada Y\ Shin K\ Tsuchida N\ Komai M[ A tactile sensor system for universal joint sections of

manipulators[ IEEE Trans[ on Robotics and Automation 0882^8]401Ð6[ð08Ł FSR Sensors[ Interlink Electronics Inc[\ CA\ U[S[A[\ 0886[ð19Ł Ando S\ Shinoda H[ Ultrasonic emission tactile sensing[ IEEE Control Systems Magazine 0884^04]50Ð

8[ð10Ł Shinoda H\ Ando S[ A tactile sensor with 4!d deformation sensing element[ In] IEEE Int[ Conf[ on

Robotics and Automation\ Minneapolis\ MN\ 11Ð17 Apr 0885[ IEEE\ Robot + Automat Soc\ 0885\pp[ 6Ð01[

ð11Ł Hutchings BL\ Grahn AR\ Petersen RJ[ Multiple!layer cross!_eld ultrasonic tactile sensor[ In] IEEE\Int[ Conf[ on Robotics and Automation\ San Diego\ CA\ 7Ð02 May[ IEEE\ 0883\ pp[ 1411Ð7[

ð12Ł Shinoda H\ Uehara M\ Ando S[ A tactile sensor using 2!dimensional structure[ In] IEEE Int[ Conf[on Robotics and Automation\ Atlanta\ GA\ 1Ð5 May[ IEEE\ 0882\ pp[ 324Ð30[

ð13Ł Roy N\ Dudek G\ Freedman P[ Surface sensing and classi_cation for e.cient mobile robot navi!gation[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\ 11Ð17 Apr 0885[ IEEE\Robot + Automat Soc\ 0885\ pp[ 0113Ð7[

ð14Ł Nowlin WC[ Experimental results on bayesian algorithms for interpreting compliant tactile sensingdata[ In] 0880 Int[ Conf[ on Robotics and Automation\ Sacramento\ CA\ 8Ð00 Apr 0880[ IEEE\0880\ pp[ 267Ð72[


ð15Ł Yeung SK\ Petriu EM\ McMath WS\ and Petriu DC[ High sampling resolution tactile sensor forobject recognition[ IEEE Trans[ on Instrumentation and Measurement\ 0883^32]166Ð71[

ð16Ł Jockusch J\ Walter J\ Ritter H[ A tactile sensor system for a three!_ngered robot manipulator[ In]IEEE Int[ Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ Apr 0886[ IEEE\ Robot+ Automat Soc\ 0886\ pp[ 2979Ð5[

ð17Ł Saad RE\ Bonen A\ Smith KC\ Benhabib B[ Distributed!force recovery for a planar photoelastictactile sensor[ IEEE Trans[ on Instrumentation and Measurement\ 0885^34]430Ð5[

ð18Ł Shinoda H\ Morimoto N\ Ando S[ Tactile sensing using tensor cell[ In] IEEE Int[ Conf[ on Roboticsand Automation\ Nagoya\ Japan\ 10Ð16 May[ IEEE\ 0884\ pp[ 714Ð29[

ð29Ł Shinoda H\ Matsumoto K\ Ando S[ Acoustic resonant tensor cell for tactile sensing[ In] IEEE Int[Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ April 0886[ IEEE\ Robot +Automat Soc\ 0886\ pp[ 2976Ð81[

ð20Ł Dario P\ Rucci M\ Guadagnini C\ Laschi C[ An investigation of a robot system for disassemblyautomation[ In] IEEE Int[ Conf[ on Robotics and Automation\ San Diego\ CA\ 7Ð02 May 0883[IEEE\ Robot + Automat Soc\ 0883\ pp[ 2404Ð10[

ð21Ł Russell RA\ Parkinson S[ Sensing surface shape by touch[ In] IEEE Int[ Conf[ on Robotics andAutomation\ Atlanta\ GA\ 1Ð5 May 0882[ IEEE\ Robot + Automat Soc\ 0882\ pp[ 2976Ð81[

ð22Ł Borovac B\ Seslija D\ Stankovski S[ Soft sensored grippers in assembly process[ In] IEEE Int[ Conf[on Robotics and Automation\ Nice\ France\ May 0881[ IEEE\ Robot + Automat Soc\ 0881\ pp[0172Ð7[

ð23Ł Howe RD\ Cutkosky MR[ Dynamic tactile sensing*perception of _ne surface!features with stressrate sensing[ IEEE Trans[ on Robotics and Automation\ 0882^8]039Ð40[

ð24Ł Han H\ Shimada A\ Kawamura S[ Analysis of friction on human _ngers and design of arti_cial_ngers[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\ 11Ð17 April 0885[IEEE\ Robot + Automat Soc\ 0885\ pp[ 2950Ð5[

ð25Ł Shimojo M\ Ishikawa M[ An active touch sensing method using a spatial!_ltering tactile sensor[ In]IEEE Int[ Conf[ on Robotics and Automation\ Atlanta\ GA\ 1Ð5 May[ IEEE\ 0882\ pp[ 837Ð43[

ð26Ł Johansson RS\ Westling G[ Tactile a}erent signals in the control of precision grip[ Attention andPerformance\ 0889\ 7 pp[ 566Ð602[

ð27Ł Taddeucci D\ Laschi C\ Lazzarini R\ Magni R\ Dario P\ Starita A[ An approach to integrated tactileperception[ In] IEEE Int[ Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ April0886[ IEEE\ Robot + Automat Soc\ 0886\ pp[ 2099Ð4[

ð28Ł Dario P[ Tactile sensingÐtechnology and applications[ Sensors and Actuators A*Phys[ 0880^15]140Ð5[

ð39Ł Shimoga KB\ Goldenberg AA[ Soft materials for robot _ngers[ In] IEEE Int[ Conf[ on Robotics andAutomation\ Nice France\ May 0881[ IEEE\ Robot + Automat Soc\ 0881\ pp[ 0299Ð4[

ð30Ł Sawahata K\ Gong JP\ Osada Y[ Soft and wet touch!sensing system made of hydrogel[ Mac!romolecular Rapid Comm[ 0884^05]602Ð5[

ð31Ł Voyles RM\ Fedder G\ Khosla PK[ Design of a modular tactile sensor and actuator based on anelectrorheological gel[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\ 11Ð17April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 02Ð6[

ð32Ł Taylor PM\ Hosseini!Sianaki A\ Varley CJ[ An electrorheological ~uid!based tactile array for virtualenvironments[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\ 11Ð17 April0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 07Ð12[

ð33Ł Venkataraman ST\ Iberall T[ Dextrous Robot Hands[ Springer!Verlag\ 0889[ð34Ł Johnston D\ Zhang\ P\ Hollerbach J\ Jacobsen S[ A full tactile sensing suite for dexterous robot hands

and use in contact force control[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\11Ð17 April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 2111Ð6[

ð35Ł Howe RD[ Tactile sensing and the control of robotic manipulation[ Advanced Robotics 0883^7]2]134Ð50[

ð36Ł Son JS\ Howe RD[ Tactile sensing and sti}ness control with multi_ngered hands[ In] IEEE Int[ Conf[on Robotics and Automation\ Minneapolis\ MN\ 11Ð17 April 0885[ IEEE\ Robot + Automat Soc\0885\ pp[ 2117Ð22[

ð37Ł Maekawa H\ Tanie K\ Komoriya K\ Kaneko M\ Horiguchi C\ Sugawara T[ Development of a _nger!


shaped tactile sensor and its evaluation by active touch[ In] IEEE Int[ Conf[ on Robotics andAutomation\ Nice\ France\ 01Ð03 May[ IEEE\ 0881\ pp[ 0216Ð23[

ð38Ł Maekawa H\ Tanie K\ Komoriya K[ A _nger!shaped tactile sensor using an optical wave!guide[ In]IEEE Int[ Conf[ on Systems\ Man and Cybernetics\ Le Touquet\ France\ 06Ð19 Oct[ IEEE\ 0882\pp[ 392Ð7[

ð49Ł Maekawa H[\ Tanie K\ Komoriya K[ Tactile sensor!based manipulation of an unknown object by amulti_ngered hand with rolling!contact[ In] IEEE Int[ Conf[ on Robotics and Automation\ Nagoya\Japan\ 10Ð16 May[ IEEE\ 0884\ pp[ 632Ð49[

ð40Ł Maekawa H\ Tanie K\ Komoriya K[ Dynamic grasping force control using tactile feedback for graspof multi_ngered hand[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\ 11Ð17April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 1351Ð8[

ð41Ł Zhang H\ Maekawa H\ Tanie K[ Sensitivity analysis and experiments of curvature estimation basedon rolling contact[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\ 11Ð17 April0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 2403Ð8[

ð42Ł Tsujimura T\ Yabuta T[ A tactile sensing method employing force torque information throughinsensitive probes[ In] IEEE Int[ Conf[ on Robotics and Automation\ Nice\ France\ 01Ð03 May[IEEE\ 0881\ 0204Ð19[

ð43Ł Russell RA[ Using tactile whiskers to measure surface contours[ In] IEEE Int[ Conf[ on Roboticsand Automation\ Nice\ France\ May 0881[ IEEE\ Robot + Automat Soc\ 0881\ pp[ 0184Ð8[

ð44Ł Kaneko M[ Active antenna[ In] IEEE Int[ Conf[ on Robotics and Automation\ San Diego\ CA\ 7Ð02 May 0883[ IEEE\ Robot + Automat Soc\ 0883\ pp[ 1554Ð60[

ð45Ł Ueno N\ Kaneko M[ On a new contact sensing strategy for dynamic active antenna[ In] IEEE Int[Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot + AutomatSoc\ 0884\ pp[ 0019Ð4[

ð46Ł Kaneko M\ Kanayama N\ Tsuji T[ 2!d active antenna for contact sensing[ In] IEEE Int[ Conf[ onRobotics and Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot + Automat Soc\ 0884\pp[ 0002Ð8[

ð47Ł Kaneko M\ Kanayama N\ Tsuji T[ Vision based active antenna[ In] IEEE Int[ Conf[ on Roboticsand Automation\ Minneapolis\ MN\ 11Ð17 April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[1444Ð59[

ð48Ł Son JS\ Cutkosky MR\ Howe RD[ Comparison of contact sensor localization abilities duringmanipulation[ Robotics and Autonomous Systems 0885^06]106Ð22[

ð59Ł Ellis RE\ Ganeshan SR\ Lederman SJ[ A tactile sensor!based on thin!plate deformation[ Robotica0883^01]232Ð40[

ð50Ł Speeter TH[ 2!dimensional _nite!element analysis of elastic continua for tactile sensing[ Int[ J[ ofRobotics Research 0881^00]0Ð08[

ð51Ł Ohka M\ Mitsuya Y\ Takeuchi S\ Kamaekawa O\ Ishihara H[ A 2!axis optical tactile sensor*"FEMcontact analysis and sensing experiments using a large!sized tactile sensor#[ In] IEEE Int[ Conf[ onRobotics and Automation\ Nagoya\ Japan\ 10Ð16 May[ IEEE\ 0884\ pp[ 706Ð13[

ð52Ł Ricker SL\ Ellis RE[ 1!d _nite!element models of tactile sensors[ In] IEEE Int[ Conf[ on Roboticsand Automation\ Atlanta\ GA\ 1Ð5 May[ IEEE\ 0882\ pp[ 830!6[

ð53Ł Ellis RE\ Qin M[ Singular!value and _nite!element analysis of tactile shape recognition[ In] IEEEInt[ Conf[ on Robotics and Automation\ San Diego\ CA\ 7Ð02 May 0883[ IEEE\ Robot + AutomatSoc\ 0883\ pp[ 1418Ð24[

ð54Ł Nicholson EJ\ Fearing RS[ The reliability of curvature estimates from linear elastic tactile sensors[In] IEEE Int[ Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May[ IEEE\ 0884\ pp[0015Ð22[

ð55Ł Shimojo M[ Mechanical _ltering e}ect of elastic cover for tactile sensor[ IEEE Trans[ on Roboticsand Automation 0886^02]017Ð21[

ð56Ł Chen NN\ Zhang H\ Rink RE[ Touch!driven robot control using a tactile Jacobian[ In] IEEE Int[Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ April 0886[ IEEE\ Robot +Automat Soc\ 0886\ pp[ 0626Ð31[

ð57Ł Holweg EGM\ Hoeve H\ Jongkind W\ Marconi L\ Melchiorri C\ Bonivento C[ Slip detection by tactile


sensors*algorithms and experimental results[ In] IEEE Int[ Conf[ on Robotics and Automation\Minneapolis\ MN\ 11Ð17 April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 2123Ð8[

ð58Ł Tremblay MR\ Cutkosky MR[ Estimating friction using incipient slip sensing during a manipulationtask[ In] IEEE Int[ Conf[ on Robotics and Automation\ Atlanta\ GA\ 1Ð5 May 0882[ IEEE\ Robot+ Automat Soc\ 0882\ pp[ 318Ð23[

ð69Ł Son JS\ Monteverde EA\ Howe RD[ A tactile sensor for localizing transient events in manipulation[In] IEEE Int[ Conf[ on Robotics and Automation\ San Diego\ CA\ 7Ð02 May[ IEEE\ Robot +Automat Soc\ 0883\ pp[ 360Ð5[

ð60Ł Howe RD[ A force!re~ecting teleoperated hand system for the study of tactile sensing in precisionmanipulation[ In] IEEE Int[ Conf[ on Robotics and Automation\ Nice\ France\ 01Ð03 May[ IEEE\0881\ pp[ 0210Ð5[

ð61Ł Chen N\ Rink R\ Zhang H[ Local object shape from tactile sensing[ In] IEEE Int[ Conf[ on Roboticsand Automation\ Minneapolis\ MN\ 11Ð17 April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[2385Ð490[

ð62Ł Charlebois M\ Gupta K\ Payandeh S[ Shape description of general\ curved surfaces using tactilesensing and surface normal information[ In] IEEE Int[ Conf[ on Robotics and Automation\ Albu!querque\ New Mexico\ April 0886[ IEEE\ Robot + Automat Soc\ 0886\ pp[ 1708Ð13[

ð63Ł Buttazzo G\ Bicchi A\ Dario P[ Robot tactile perception[ In] C[S[G[ Lee\ editor[ Sensor!BasedRobots] Algorithms and Architectures "NATO ASI series#[ Springer!Verlag\ New York\ 0880\ pp[14Ð28[

ð64Ł Caselli S\ Magnanini C\ Zanichelli F\ Cara. E[ E.cient exploration and recognition of convexobjects based on haptic perception[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\MN\ 11Ð17 April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 2497Ð02[

ð65Ł Kinoshita G\ Mutoh E\ Tanie K[ Haptic aspect graph representation of 2!d object shapes[ In] IEEEInt[ Conf[ on Robotics and Automation\ Nice France\ May 0881[ IEEE\ Robot + Automat Soc\0881\ pp[ 0537Ð42[

ð66Ł Kaneko M\ Higashimori M\ Tsuji T[ Pulling motion based tactile sensing for concave surface[ In]IEEE Int[ Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ April 0886[ IEEE\ Robot+ Automat Soc\ 0886\ pp[ 1366Ð73[

ð67Ł Beccari G\ Caselli S\ Zanichelli F[ Pose!independent recognition of convex objects from sparse tactiledata[ In] IEEE Int[ Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ April 0886[IEEE\ Robot + Automat Soc\ 0886\ pp[ 2286Ð391[

ð68Ł Okamura AM\ Turner ML\ Cutkosky MR[ Haptic exploration of objects with rolling and sliding[In] IEEE Int[ Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ April 0886[ IEEE\Robot + Automat Soc\ 0886\ pp[ 1374Ð89[

ð79Ł Kontarinis DA\ Son JS\ Peine W\ Howe RD[ A tactile shape sensing and display system forteleoperated minipulation[ In] IEEE Int[ Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot + Automat Soc\ 0884\ pp[ 530Ð5[

ð70Ł Fearing RS[ Tactile sensing for shape interpretation[ In] Workshop at the 0877 IEEE Conf[ onRobotics and Automation ] Dextrous Robot Hands\ Philadelphia\ PA\ 0887\ in ð33Ł\ pp[ 198Ð27[

ð71Ł Cohn MB\ Lam N\ Fearing RS[ Tactile feedback for teleoperation[ In] Telemanipulator Technology\SPIE Proc[\ Boston\ November 0881[ 0881\ pp[ 139Ð43[

ð72Ł Fearing RS\ Moy G\ Tan E[ Some basic issues in teletaction[ In] IEEE Int[ Conf[ on Robotics andAutomation\ Albuquerque\ New Mexico\ April 0886[ IEEE\ Robot + Automat Soc\ 0886\ pp[ 2982Ð8[

ð73Ł Yoshikawa P\ Nagura A[ A touch and force display system for haptic interface[ In] IEEE Int[ Conf[on Robotics and Automation\ Albuquerque\ New Mexico\ April 0886[ IEEE\ Robot + AutomatSoc\ 0886\ pp[ 2907Ð13[

ð74Ł Caldwell DG\ Tsagarakis N\ Wardle A[ Mechano!thermo and proprioceptor feedback for integratedhaptic feedback[ In] IEEE Int[ Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\April 0886[ IEEE\ Robot + Automat Soc\ 0886\ pp[ 1380Ð5[

ð75Ł Shimojo M\ Sato S\ Seki Y\ Takahasi A[ A system for simultaneous measuring grasping posture andpressure distribution[ In] IEEE Int[ Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May0884[ IEEE\ Robot + Automat Soc\ 0884\ pp[ 720Ð5[


ð76Ł Watanabe T\ Fukui S[ A method for controlling tactile sensation of surface roughness using ultrasonicvibration[ In] IEEE Int[ Conf[ on Robotics and Automation\ Nogoya\ Japan\ 10Ð16 May 0884[IEEE\ Robot + Automat Soc\ 0884\ pp[ 0023Ð8[

ð77Ł Wu JL\ Morita S\ Kawamura S[ Human sensory fusion on visual and tactile sensing for virtual!reality[ In] IEEE Int[ Conf[ on Robotics and Automation\ Minneapolis\ MN\ 11Ð17 April 0885[IEEE\ Robot + Automat Soc\ 0885\ pp[ 1254Ð69[

ð78Ł Lederman SJ\ Ganeshan SR\ Ellis RE[ E}ortful touch with minimum movement*revisited[ J[ ofExperimental Psychology!Human Perception and Performance 0885^11]740Ð57[

ð89Ł Li DS\ Shida K[ Fuzzy algorithm to discriminate material properties using touch sensors[ ElectricalEngineering in Japan 0886^007]50Ð69[

ð80Ł Canepa G\ Morabito M\ DeRossi D\ Caiti A\ Parisini T[ Shape from touch by a neural net[ In] IEEEInt[ Conf[ on Robotics and Automation\ Nice\ France\ May 0881[ IEEE\ Robot + Automat Soc\0881\ pp[ 1964Ð79[

ð81Ł Branca A\ Delaney W\ Lovergine FP\ Distante A[ Surface defect detection by texture analysis witha neural network[ In] IEEE Int[ Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May0884[ IEEE\ Robot + Automat Soc\ 0884\ pp[ 0386Ð491[

ð82Ł Pati YC\ Krishnaprasad PS\ Peckerar MC[ An analog neural network solution to the inverse problemof early taction[ IEEE Trans[ on Robotics and Automation 0881^7]085Ð101[

ð83Ł Caiti A\ Canepa G\ DeRossi D\ Germagnoli F\ Magenes G\ Parisini T[ Towards the realization ofan arti_cial tactile system*_ne!form discrimination by a tensorial tactile sensor array and neuralinversion algorithms[ IEEE Trans[ on Systems Man and Cybernetics 0884^14]822Ð35[

ð84Ł Rucci M\ Dario P[ Autonomous learning of tactile!motor coordination in robotics[ In] IEEE Int[Conf[ on Robotics and Automation\ San Diego\ CA\ 7Ð02 May 0883[ IEEE\ Robot + Automat Soc\0883\ pp[ 2129Ð5[

ð85Ł Rucci M\ Bajcsy R[ Learning visio!tactile coordination in robotic systems[ In] IEEE Int[ Conf[ onRobotics and Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot + Automat Soc\ 0884\pp[ 1567Ð72[

ð86Ł Yamada Y\ Ishiguro A\ Uchikawa Y[ A method of 2!d object reconstruction by fusing vision withtouch using internal models with global and local deformations[ In] IEEE Int[ Conf[ on Roboticsand Automation\ Atlanta\ GA\ 1Ð5 May 0882[ IEEE\ Robot + Automat Soc\ 0882\ pp[ 671Ð6[

ð87Ł Joshi R\ Sanderson AC[ Model!based multisensor data fusion] A minimal representation approach[In] IEEE Int[ Conf[ on Robotics and Automation\ San Diego\ CA\ 7Ð02 May 0883[ IEEE\ Robot +Automat Soc\ 0883\ pp[ 366Ð73[

ð88Ł Joshi R\ Sanderson AC[ Shape matching from grasp using a minimal representation size criterion[In] IEEE Int[ Conf[ on Robotics and Automation\ Atlanta\ GA\ 1Ð5 May 0882[ IEEE\ Robot +Automat Soc\ 0882\ pp[ 331Ð8[

ð099Ł Joshi R\ Sanderson AC[ Multisensor fusion and unknown statistics[ In] IEEE Int[ Conf[ on Roboticsand Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot + Automat Soc\ 0884\ pp[ 1569Ð6[

ð090Ł Voges U[ Technology in laparoscopy*what to expect in the future[ Urologe Ausgabe "a#0885^24]197Ð03[

ð091Ł Howe RD\ Peine W\ Kontarinis DA\ Son JS[ Remote palpation technology[ IEEE Engineering inMedicine and Biology\ May:June 0884\ pp[ 207Ð12[

ð092Ł Bicchi A\ Canepa G\ DeRossi D\ Iacconi P\ Scilingo EP[ A sensorized minimally invasive surgerytool for detecting tissutal elastic properties[ In] IEEE Int[ Conf[ on Robotics and Automation\Minneapolis\ MN\ 11Ð17 April 0885[ IEEE\ Robot + Automat Soc\ 0885\ pp[ 773Ð7[

ð093Ł Fischer H\ Heilig R\ Trapp R\ Brhel K[ Tactile optical sensor for use in minimally invasive surgery[Langenbecks Archiv Fur Chirurgie 0885^0189[

ð094Ł Ohtsuka T\ Furuse A\ Kohno T\ Nakajima J\ Yagyu K\ Omata S[ New tactile sensor techniques forlocalization of pulmonary nodules[ Int[ Surgery 0886^71]01Ð3[

ð095Ł Miyaji K\ Furuse A\ Kaneko Y\ Ohtsuka T\ Sugiura S\ Omata S[ Regional myocardial sti}nessmeasured by a new tactile sensor system[ Japanese Heart 0886^27]4]698Ð04[

ð096Ł Tanimoto M\ Arai F\ Fukuda T\ Iwata H\ Itoigawa K\ Gotoh Y\ Hashimoto M\ Negoro M[ Micro


force sensor for intravascular neurosurgery[ In] IEEE Int[ Conf[ on Robotics and Automation\Albuquerque\ New Mexico\ April 0886[ IEEE\ Robot + Automat Soc\ 0886\ pp[ 0450Ð5[

ð097Ł Dario P\ Carrozza MC[ Lencioni L\ Maganani B\ D|Attanasio S[ A micro robotic system forcolonoscopy[ In] IEEE Int[ Conf[ on Robotics and Automation\ Albuquerque\ New Mexico\ April0886[ IEEE\ Robot + Automat Soc\ 0886\ pp[ 0456Ð61[

ð098Ł Ho SC\ Hibberd RD\ Cobb J\ Davies BL[ Force control for robot surgery[ In 6th Int[ Conf[ onAdvanced Robotics\ Catalonia\ Spain\ 19Ð11 Sept[ 0884\ pp[ 10Ð20[

ð009Ł Cohn MB\ Crawford LS\ Wendlandt JM\ Sastry SS[ Surgical application of milli!robots[ RoboticSystems 0884^01]5]390Ð05[

ð000Ł Umemori M\ Sugawara J\ Kawauchi M\ Mitani H[ A pressure!distribution sensor "PDS#\ for evalu!ation of lip functions[ Amer[ J[ of Orthodontics and Dentofacial Orthopedics 0885^098]362Ð79[

ð001Ł Dario P\ Guglielmelli E\ Genovese V\ Toro M[ Robot assistants] Applications and evolution[Robotics and Autonomous Systems 0885^07]114Ð23[

ð002Ł Dario P\ Guglielmelli E\ Laschi E\ Guadagnini C\ Pasquarelli G\ Morana G[ Movaid] A newEuropean joint project in the _eld of rehabilitation robotics[ In] 6th Int[ Conf[ on Advanced Robotics\Catalonia\ Spain\ 19Ð11 Sept[ 0884[ pp[ 40Ð8[

ð003Ł Tobita H\ Kawamura T\ Sugimoto Y\ Nakamura H[ The development of {safe partner equipment|[In] IEEE Int[ Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot+ Automat Soc\ 0884\ pp[ 1319Ð5[

ð004Ł Suita K\ Yamada Y\ Tsuchida N\ Imai K\ Ikeda H\ Sugimoto N[ A failure!to!safety {kyozon| systemwith simple contact detection and stop capabilities for safe human!autonomous robot coexistence[In] IEEE Int[ Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot+ Automat Soc\ 0884\ pp[ 2978Ð85[

ð005Ł Ikeura R\ Inooka H[ Variable impedance control of a robot for cooperation with a human[ In] IEEEInt[ Conf[ on Robotics and Automation\ Nagoya\ Japan\ 10Ð16 May 0884[ IEEE\ Robot + AutomatSoc\ 0884\ pp[ 2986Ð091[

ð006Ł Ulrich I\ Mondada F\ Nicoud J[!D[ Autonomous vacuum cleaner[ Robotics and AutonomousSystems 0886^08^2]122Ð134[

ð007Ł Service!Robot[ Bradford] MCB University Press\ 0884![ð008Ł Real World Computing Program[ Real World Computing Partnership[ 1!4!01\ Higashi Kanda\

Chiyoda!ku\ Tokyo 090\ Japan\ 0884[ð019Ł Fiorini P\ Ali K\ Seraji H[ Health care robotics] A progress report[ In] IEEE Int[ Conf[ on Robotics

and Automation\ Albuquerque\ New Mexico\ April 0886[ IEEE\ Robot + Automat Soc\ 0886\ pp[0160Ð0165[

ð010Ł Gale GE[ Automatic height control of a stripper harvester using a tactile sensor to detect the crop[J[ of Agricultural Engineering Research 0884^50]106Ð15[

ð011Ł Stone RSW\ Brett PN[ A sensing technique for the measurement of tactile forces in the gripping ofdough!like materials[ Proceedings of the Institution of Mechanical Engineers Part B*J[ of Engin!eering Manufacture 0885^109]150Ð8[

ð012Ł Mattiazzo G\ Mauro S\ Raparelli T\ Velardocchia M[ A fuzzy controlled pneumatic gripper forasparagus harvesting[ Control Engineering Practice 0884^2]0452Ð69[

ð013Ł Naghdy F\ Esmaili M[ Soft fruit grading using a robotics gripper[ Robotics and Automation0885^00]2]82Ð090[

ð014Ł Friedrich WE\ Lim PK[ Smart end e}ector sensing for variable object handling[ In] Field and ServiceRobotics Conf[ "FSR |86#\ Canberra\ Australia\ 0885\ pp[ 352Ð5[

review article tactile sensing for mechatronics*a …...5 m[h[ lee\ h[r[ nicholls:mechatronics 8...

Documents