28 PERVASIVEcomputing Published by the IEEE Computer Society ■ 1536-1268/07/$25.00 © 2007 IEEE

Pervasive Computingand Autism: AssistingCaregivers of Childrenwith Special Needs

Caring for children and older indi-viduals with autism is often a com-plex, lifelong challenge. Familiesand care networks for children withautism often face numerous choices

regarding types of therapies, doctors, and drugand diet regimens. Each child’s unique naturemakes predicting which treatments will work dif-ficult, and timeliness is often critical. For these

reasons, parents might chooseto try several treatments simul-taneously, making it evenharder to determine whichtreatments work and whichdon’t. Further complicationsresult when caregivers don’t see

the effects of treatment immediately, perhapsbecause the changes are minor or imperceptible.

Often, many caregivers help determine if achild is receiving the best treatments possible. Thesheer amount of data they must collect can oftencause difficulty in making the best choices for thatchild. Also, individuals with autism often can’tcommunicate their internal states, so caregiversmust rely on externally perceivable characteris-tics to determine the person’s needs.

Technology can make caring for children withautism more efficient or hide some of its complex-

ities. Data capture and analysis is an important partof making decisions about whether a treatment isworking. Pervasive technology can help increasethe amount of data collected, make it easier to col-lect, and help caregivers quickly scan through datato make better decisions. Technology can also helpthe caregivers on a child’s team communicate moreeffectively and efficiently.

What is autism?Pervasive development disorder, also known as

autism spectrum disorder (ASD), is a cognitiveimpairment characterized by deficiencies in com-munication, social interaction, and creative orimaginative play.1 This spectrum includes autisticdisorder (autism), Asperger’s syndrome, pervasivedevelopment disorder not otherwise specified, andRett’s disorder. Individuals on this spectrum oftenexhibit stereotypical, self-stimulatory behaviorssuch as rocking, hand flapping, or vocalizations.Autism is the most common on the spectrum,affecting an estimated 1.5 million Americans todayand growing at the astonishing rate of 10 to 17percent annually. Autism is typically diagnosedbetween the ages of two and six, although varia-tions of ASD can sometimes be diagnosed earlieror later. The disorder has recently been a populartopic of discussion in the mainstream media, prob-

Pervasive computing technologies can support children with autism and their caregivers. Work continues on systems that aid record collection and analysis, decision making,communication, and assessment of children’s internal states.

H E A L T H C A R E

Julie A. Kientz, Gillian R. Hayes,Tracy L. Westeyn, Thad Starner,and Gregory D. AbowdGeorgia Institute of Technology

ably because it’s being diagnosed at amuch higher rate than ever before. Cur-rent estimates from the Autism Society ofAmerica are that 1 in 166 children will bediagnosed with some form of autism atsome point in their life.

Although autism typically presentsitself in individuals in some known ways,it manifests differently in each personand sometimes differently in the sameperson over time. This extreme variationamong children has led practitioners andfamily members alike to make the anec-dotal comment, “If you’ve seen one childwith autism, you’ve seen one child withautism.”

Evidence from the behavioral, educa-tional, and social sciences indicates thatearly diagnosis and intervention could bekey to achieving greater independence, anultimate goal of these care activities.2 Aswith other language acquisition disorderssuch as deafness, children who don’tachieve functional language by school agecan be severely limited in their future abil-ities to interact with others.3 Thus, care-givers are often in a race against time to

find treatments that will work for theirchildren. Treatments include pharmaco-logical interventions, special diets, holis-tic approaches such as occupational ther-apy and sensory integration, behavioraltherapies such as applied behavior analy-sis or functional behavior analysis, andsymptom-specific treatments such asspeech or language therapy (see the side-bar “Additional Computing TechnologiesSupporting Individuals with Autism”).Moreover, caregivers often try differentinterventions simultaneously.

Three technologies to support caregivers

Our work at Georgia Tech for the pastthree years has focused on supportingindividuals with autism and their care-givers with computing technology. We’veexplored several dimensions of the carecycle and network; here we present threeof those projects.

Abaris: Supporting collaborativedecision making

Discrete Trial Training (DTT) therapy

is a current best-practice intervention forchildren with autism in which therapistteams teach basic skills intensively andone-on-one. Therapists assign a grade forthe level of independence at which thechild was able to complete the skill. Aftereach therapy session, the therapists cal-culate the percentage correct for each skilland then plot the data points on hand-drawn graphs. Every one or two weeks,all the therapists meet to discuss the child’sprogress and how they might change theirtherapy practice to make the child moresuccessful. During these meetings, com-mon discussion points are the skills thechild can accomplish and clarifications onthe grading of particular skills. To beeffective, DTT involves rigorous data col-lection and analysis. Unfortunately, it’salso subject to inconsistencies and inac-curacies because all therapists have dif-ferent skills and different interpretationsof therapy’s progress.

We developed the Abaris system to sup-port teams of DTT therapists.4 Abarissupports this practice through phoneme-spotting voice recognition, which pro-

JANUARY–MARCH 2007 PERVASIVEcomputing 29

O ther researchers have explored technology for use by the indi-

viduals themselves rather than by their caregivers. These de-

vices include Simone Says, a system using voice recognition technol-

ogy to teach and analyze language skills.1 The Discrete Trial Trainer

(www.dttrainer.com) is commercial software that attempts to replace

the therapist by administering similar therapy and education through

the computer. Simone Says and DTT can ease some of the burden on

caregivers by letting a computer administer therapy.

Other technologies have focused on how to teach individuals

with autism social skills and aid their communication. Andrea Tar-

taro has explored storytelling with virtual peers to teach social skills

to children with autism in a risk-free setting,2 while researchers at

MIT have looked at how to emphasize people’s images to help indi-

viduals with autism understand subtle emotional cues.3 Recent work

by Anne Marie Piper and her colleagues explored using a multi-

player tabletop game called SIDES to encourage children with autism

to learn social interactions and turn-taking.4 These types of socially

based applications, used in conjunction with other treatments, can

provide a richer education in social skills and leave the caregivers

with more time for other, more pressing issues.

REFERENCES

1. J.F. Lehman, “Toward the Use of Speech and Natural Language Tech-nology in Intervention for a Language-Disordered Population,” Proc.3rd Int’l ACM Conf. Assistive Technologies, ACM Press, 1998, pp. 19�26.

2. A. Tartaro, “Storytelling with a Virtual Peer as an Intervention for Chil-dren with Autism: Assets Doctoral Consortium,” Proc. 7th Int’l ACMSIGACCESS Conf. Computers and Accessibility, ACM Press, 2005, pp. 42�44.

3. R. Kaliouby and P. Robinson, “The Emotional Hearing Aid: An AssistiveTool for Children with Asperger Syndrome,” Universal Access in the Infor-mation Soc., vol. 4, no. 2, 2005, pp. 121�134.

4. A.M. Piper et al., “SIDES: A Cooperative Tabletop Computer Game forSocial Skills Development,” Proc. 20th Anniversary Conf. Computer Sup-ported Cooperative Work (CSCW 06), ACM Press, 2006, pp. 1�10.

Additional Computing TechnologiesSupporting Individuals with Autism

vides indices into videos of therapy ses-sions (see figure 1a). Therapists use anAnoto digital pen to capture data andprovide additional indices to the videos.Therapists can access the videos duringtheir team meetings via an interface (seefigure 1b). Because of the easy indexing,a therapist can use video evidence toreview progress, find inaccuracies in grad-ing, and show problem areas to othercaregivers. Abaris also lets therapists auto-matically generate graphs and associatethem with the data sheets they used dur-ing therapy (see figure 1c).

We deployed Abaris for four monthswith one particular home-based DTTteam and found that the system enabledtherapists to use objective evidence morefrequently in the decision-making pro-cess.5 Therapists used videos, graphs, anddata sheets more frequently than they didwithout Abaris. Prior to Abaris, theywould rely on their own recollections ofwhat happened during therapy and onlyconsult written data sheets if necessary.We also determined that using Abaris dur-ing team meetings increased collaborationamong the therapists.

CareLog: Capturing andanalyzing behavioral data

Children with autism and other devel-opmental delays exhibit behaviors thatare often inappropriate and might be dis-ruptive or dangerous to themselves or oth-ers. To address these behaviors, teachers,specialists, and parents often try to under-stand their cause. One best practice fordetermining cause is known as functionalbehavior assessment. When done in thenatural environment, FBA includes directobservation. Caregivers, seeing a behav-ioral incident, note the incident’s contextand what happened directly before (an-tecedent) and after it (consequence). Oncethe number of incidents recorded by the

30 PERVASIVEcomputing www.computer.org/pervasive

H E A L T H C A R E

Figure 1. Screen shots from Abaris datacapture and analysis technology: (a) capture interface, (b) video previewscreen, and (c) data analysis graphs of a child’s progress.

caregiver reaches a point of data satura-tion (usually 20 to 30 incidents over a fewweeks), a behavioral specialist analyzes thedata to determine the behavior’s func-tion—for example, the child might bescreaming to gain the teacher’s attention.In classrooms and homes, observing thesebehaviors rigorously enough to have con-fidence in an assessment can be difficult.Incidents might occur at unexpected timesand places, and caregivers might not beable to write down what happened beforeor after while simultaneously managingthe behavior itself.

CareLog is a prototype system designedto help caregivers document and analyzespecific, unplanned incidents of interest aspart of an FBA. CareLog uses audio andvideo buffering6 to allow selective archiv-ing of these media after events occur. Care-givers use a wireless button to triggerarchiving. They then use a standard desk-top computer to watch the videos and tagthem with metadata (see figure 2a). Fi-nally, CareLog provides graphs and otheranalytic tools for functional assessment(see figure 2b).

We deployed CareLog in four class-rooms at a special-education school in theAtlanta area as part of a semicontrolledstudy of its effectiveness.7 Each teacherconducted an FBA for two children, one

using the traditional pen-and-papermethod and one using our technology-augmented process. The ordering of thetwo methods was counterbalanced andgroups randomly assigned. Overall, weobserved increases in both the accuracyand the efficiency of the process. Fur-thermore, the teachers reported enjoyingthe process more while feeling less bur-dened and more confident in the resultsof their assessments using CareLog. Oneteacher will continue using CareLog withmultiple students during the entire 2006�2007 school year.

Wearable sensors: Giving nonverbal children a voice

A major difficulty in caring for somechildren with autism is their inability tocommunicate when something is upset-ting them or making them uncomfortable.Pervasive computing might be able to helpcaregivers sense when a child is upset. Adefining characteristic of children withautism is self-stimulatory behavior, orstimming, which can include rocking,hand waving, clenching of the fists, ornonword vocalizations. Some researchersbelieve that the amount of stimming mightindicate a child’s affective state.8 Moni-toring stimming behavior might give care-givers a better idea of treatments’ effects,

when to try calming or soothing the child,or when the child might be most receptiveto learning.

We’ve researched technological meansfor detecting self-stimulatory behaviorsusing wireless body-worn sensors. Wedeveloped a proof-of-concept systemcapable of collecting data from a childwith autism and providing automaticindices into that data to aid researchers’analysis of autism.9 Our pilot studyinvolved several technical challengesregarding the selection and placement ofsensors and recognition of collected data.Using wireless Bluetooth accelerometers(see figure 3), we specifically targeted self-stimulatory behaviors of repetitive move-ments because these are often the mostsocially disruptive.

Accelerometer positioning was a ma-jor consideration for this work. Forexample, if a child were exhibiting repet-itive hand flapping, an accelerometer po-sitioned at the wrist would be more ap-propriate than on the ankle. Our goalwas to minimize the number of sensorsneeded to collect data for as many stim-ming events as possible. In the pilot study,an adult mimicked several trials of seventypical self-stimulatory behaviors inter-spersed with random daily activities. Weaffixed three accelerometers to the sub-

JANUARY–MARCH 2007 PERVASIVEcomputing 31

Figure 2. Screen shots from the CareLog functional behavior assessment system: (a) video viewing and annotation screen with fourcamera angles and (b) automatically generated graphs showing when and how often a particular behavior occurs.

(a) (b)

ject’s thigh, waist, and wrist using Velcrostraps and used the Georgia Tech Ges-ture Toolkit for recognition experiments.Results showed that automatic indexinginto data is possible (see figure 4). How-ever, although we could accurately detectthat a stimming event had occurred, wecouldn’t always label the type of behav-ior correctly. We plan to test a child-safeversion of the sensor system on threechildren with autism.

Design considerationsWe found four issues that needed to be

addressed as we designed Abaris, Care-Log, and the wearable sensors.

Understand the domainIn the year before developing any tech-

nology and throughout our design cycles,we spent time conducting in-depth con-textual inquiries of the autism domain.10

These explorations included interviewswith parents, caregivers, therapists, andteachers who interact with these childrenon a daily basis. We were also trained tobecome therapists and participated in ther-apy sessions and team meetings with ther-apists. We spent extensive time observingclassrooms, home settings, therapy ses-sions of varied types, support groups, andother domain-specific activities. Our de-sign team also included individuals withfirsthand and expert knowledge of ouruser population, including parents of chil-dren with autism, autism researchers, andbehavioral specialists. Having firsthandknowledge of children with autism helpedus attune to how well our target userswould adopt these technologies.

Make changes invisibleGetting groups of users to change their

practices to adopt any new software isalways difficult, but trying to change rit-uals with children with autism can benearly disastrous. Many children rely onstrict regimens and have difficulty adjust-ing to deviations. Pervasive computingcan seamlessly blend into the environ-ment and thus has a distinct advantageover other technological solutions in thisdomain. In the case of Abaris, we believewe succeeded in data capture because wecontinued to use the paper form factor,so the child was unlikely to notice achange.

When we did make changes, such ashaving the therapist use a headset micro-phone for voice recognition, they dis-tracted the children—they often wantedto touch it. For CareLog, we placed cam-eras in the corners of the classrooms nearthe ceiling to reduce distraction. Despitethese efforts, in one classroom, somehigher-functioning students noticed thecameras. Also, in all classrooms, stu-dents clearly recognized changes whenwe were present for data collection (inaddition to the installed cameras). Thisevidence shows that even minor changesto existing practices could interfere withthe interventions.

The easier, the betterTeams of caregivers and educators are

often diverse, coming from different back-grounds and having different levels ofcomputing expertise. Additionally, they’rerarely collocated and don’t see each otherregularly, if at all. In our studies, childrenoften saw a behavior therapist, an occu-pational therapist, and a speech therapistall on the same day, where none of theexperts could access the others’ recordsdespite the fact that doing so would be

beneficial. Secondary caregivers such astherapists and behavioral specialists mighthave limited interaction with a child, seemultiple children, and look at significantamounts of data every week. Even inschools, speech therapists were outsidecontractors, and the behavioral therapistsmoved between rooms and schoolsthroughout the day. Low-tech solutionshave included passing paper recordswith the child or parent; however, theserecords can be a burden and oftendon’t capture the data that caregiversneed.

Technologies that can automaticallycollect and share data can help supportthese types of interactions. Keeping tech-nology simple and straightforward canalso help track all the data from all thecaregivers involved.

Customizability is criticalChildren with autism each have a uni-

que set of behaviors, needs, and treat-ments. What works for one child mightnot work at all for another child with sim-ilar abilities. As a result, therapies andtreatments must often be highly tailoredto a specific child’s needs. Any technol-ogy developed to support therapies orcare must be able to reflect this unique-ness through customization. In the caseof Abaris, the skills tested changed almostdaily, so we had to design our system tosupport these changes.

With CareLog and behavior manage-ment, each behavior and its context canbe incredibly different, even for the samechild. For the wearable sensors, we hadto consider many different types of stim-ming behaviors; in the end, we built asystem that would be useful only in asmall percentage of cases. In all thesecases, the basic underlying process is rel-atively standardized, but the technologymust support variable instantiation ofthat process for various children andconditions.

32 PERVASIVEcomputing www.computer.org/pervasive

H E A L T H C A R E

Figure 3. Relative size and placement of a wireless accelerometer and sensor on an adult.

Design challengesThe studies we conducted also re-

vealed three challenges that needed to beaddressed.

Difficulties in relying on inputfrom children

Autism inherently creates communi-cation difficulties, and even those indivi-duals who can communicate might knowlittle about the processes caregivers use.So, relying on their input when design-ing technology is often difficult. Mostchildren in our studies were unable toconsent to the research process, let aloneprovide us with insights into what wouldwork and what wouldn’t. Instead, wegenerally relied on parents and caregiversto provide methods for working withthese children. Although most caregivershave the child’s best interests at heart,even experienced caregivers sometimesfind it hard to know a child’s needs.

In the case of wearable sensors, we hadto design ones that would both providegood data and be comfortable to wear.This is a difficult problem because mostactivity recognition relies on a coopera-tive subject who can perform an activityon demand. Collecting training data forchildren with autism would be hard be-cause they’re unlikely to perform thedesired activities on demand. Further-more, these children are often sensitive towearing restrictive clothing or new de-vices. Therefore, in developing our sen-sor system, we involved only neurotypicaladults to help refine it before involvingchildren.

Hardware for on-body sensingSeveral trade-offs exist in the devel-

opment of on-body sensing systems, in-cluding sensor type, power consumed,and form factor. The types and locationof sensors influence the quality of datawe can record. Also, sensors that requiremore power (such as cameras) will re-quire large batteries, making the sensorheavy and heat-dissipating. Although wecan use smaller, lighter batteries, they’llrequire more frequent charging, thus in-creasing system downtime and possiblyburdening caregivers. Regarding theform factor, the sensor should be easy tocharge, have maximum protection fromdaily wear, be unobtrusive, and remainin position during use. Keeping the sen-sor in the proper position is difficult forunwilling subjects. For example, a childmight remove or damage bracelets andremove sensors in clothing or shoes.

With this vulnerable population, bal-ancing these trade-offs is often difficult.Ideally, a sensor would require little main-tenance and go unnoticed by the childwearing it. We might also be able to lookfor ways to use environmental sensors inaddition to or instead of wearable ones.

Ethical and privacy considerationsBecause children with autism often

can’t express their needs, it’s crucial toensure we meet their best interests. Atthe same time, caregivers, teachers, andother professionals have direct concernsregarding data collection that mightaffect their jobs or have legal repercus-sions, especially if videos are taken out ofcontext. For example, restraining a childmight provide comfort, but to unin-formed outside viewers, this might looklike abuse. With all the data-recordingneeds, some technologies might violatethe child’s or caregivers’ privacy if notcarefully considered.

In addition to these challenges, coordi-nation between individuals includes issuesregarding privacy, security, and legislativecontrol. For example, in the US, patientsare protected under the Health InsurancePortability and Accountability Act, andstudents are protected under the FamilyEducational Rights and Privacy Act.These regulations ensure that medical andeducational data, respectively, are keptprivate and protected from malevolence,and software designed to share and dis-cuss this data must ensure that data isn’tmisused. The design of applications thatcan follow these laws will also helpaddress any concerns raised by the indi-viduals receiving the care or caregiversclose to that individual. Our CareLog sys-

JANUARY–MARCH 2007 PERVASIVEcomputing 33

Figure 4. Three xyz-axis data streams (wrist, ankle, and waist) with corresponding hand labels (the vertical columns) and automatic-recognition results (the short horizontal bars directly under eachcolumn). The bottommost horizontal bar shows a zoomed-in view of one of the automatic-recognition results.Zooming in shows the classification of stimming events and minor boundary errors.

34 PERVASIVEcomputing www.computer.org/pervasive

H E A L T H C A R E

tem used selective archiving of camera re-cordings in the environment to balancethe desire to catch relevant behavioralepisodes retroactively against the privacyand information overload concerns ofcontinuous recording throughout theschool day.

What lies ahead?Our work in pervasive technology for

supporting the care of children withautism has given us and our collaboratorsthe confidence and motivation to con-tinue working in this domain, becausemany challenging problems remain to besolved. Our group has begun several newprojects with other collaborators.

Early detection of autismEarly detection of autism and other

developmental delays can be integral toproviding appropriate individualizedcare for children with these delays. How-ever, manually tracking children’s manymilestones is a huge challenge. We’reworking with the US Centers for DiseaseControl to design a computing systemthat will enable easier tracking and doc-

umentation of a child’s developmentalmilestones. The system will proactivelyencourage caregivers to look for devel-opmental progress and alert them whenfurther investigation by a specialist mightbe advisable. The design and develop-ment of this system will examine how touse sensors and smart toys to detectautomatically when children achievemilestones. To motivate parents to usethe system, it will automatically createkeepsakes for their child, such as scrap-

books or DVDs. We’ll also determinehow parents share information withother caregivers or medical profession-als and find out how much informationis needed to help medical professionalsmake better diagnoses.

Collaboration in schoolsThe Abaris effort explored how a

home therapy team working with anindividual child can collaborate anduse more reliable evidence in decisionmaking. Our explorations with Care-Log showed the need to aid data col-lection in schools because teachersmust often engage in activities thatmight make them unable to capturedata. In a project in collaboration withthe University of Washington, we’reexploring extending the technologiesof Abaris and applying it to school set-tings. In particular, we’re looking athow automatic data analysis andgraphing can encourage teachers tovisit Abaris more regularly. Additionalcomputing problems in this domainwill explore how multiple caregiverscan collect data simultaneously for

multiple children and how to combinerelated data from different therapytreatments for overall analysis.

Support for autism-related researchMany researchers specializing in autism

in the education, neurology, or medicalfields are interested in studying the effectsof new drugs or treatments on childrenwith autism. Many of these studies mightbenefit from having technology assist indata collection and analysis. For example,

we’re looking at how to use technology toobserve the sleep behaviors of childrenwith autism over time to see if a certaindrug might help. Other work is looking athow wearable sensors can determinewhich factors might stress or calm chil-dren with autism.

Sound analysis for vocalstimming behavior

The difficulty of putting wearablesensors on children with autism hasencouraged us to explore other waysto use sensing technologies to detectstimming behaviors. One way we’reexploring this problem is by usingwearable or environmental micro-phones to detect vocal stimming be-haviors. This project will explore howsound-recognition technologies cancharacterize stimming and recognize itin real environments to get an overallimpression of how frequently stimmingoccurs in children.

Adapting FBA technologiesfor home-based care

Although parents rarely conduct FBAand other formalized behavioral inter-ventions at home, they often adapt schoolprocesses to support their needs. For ex-ample, at home, a goal might be to engagea child at the dinner table and keep himfrom getting up and running around orbehaving in other inappropriate ways.Furthermore, parents also often need toshare examples of both these and positiveappropriate behaviors with other care-givers outside the home. Thus, we’re de-veloping a simplified version of theCareLog interface that takes advantageof selective archiving in the home. Thisprototype system, known as BICapture(behavioral-imaging capture), lets care-givers capture and share short snippets ofaudio and video through a streamlinedinterface similar to CareLog but withoutthe more complex graphing and visual-izations required for FBA.

Early detection of autism and other

developmental delays can be integral

to providing appropriate individualized care

for children with these delays.

The lessons we learned fromthese projects might benefitother researchers in the au-tism domain as well as devel-

opers of technologies to support long-term healthcare. Many of the designissues we’ve encountered extend beyondthe autism domain and into the man-agement of care for other chronic con-ditions. Caring for individuals with au-tism has many similarities to caring forthe elderly or individuals with chronicconditions, such as diabetes or cancer.Additionally, many caregiver coordina-tion issues share a common theme withthat of computer-supported cooperativecare.11 We believe the promise of perva-sive computing can help begin to addressthese common challenges.

REFERENCES1. Diagnostic and Statistical Manual of Men-

tal Disorders (DSM-IV), 4th ed., Am. Psy-chiatric Assoc., 1994

2. R. Shore, Rethinking the Brain: New In-sights into Early Development, Families andWork Inst., 1997.

3. P. Howlin et al., “Adult Outcome for Chil-dren with Autism,” J. Child Psychology andPsychiatry, vol. 45, no. 2, 2004, pp. 212�229.

4. J.A. Kientz et al., “Abaris: Evaluating Auto-mated Capture Applied to StructuredAutism Interventions,” Proc. 7th Int’l Conf.Ubiquitous Computing (Ubicomp 05),Springer, 2005, pp. 323�339.

5. J.A. Kientz et al., “From the War Room tothe Living Room: Decision Support forHome-Based Therapy Teams,” Proc. 20thAnniversary Conf. Computer SupportedCooperative Work (CSCW 06), ACMPress, 2006, pp. 209�218.

6. G.R. Hayes et al., “Experience Buffers: ASocially Appropriate, Selective ArchivingTool for Evidence-Based Care,” CHI 05

Extended Abstracts on Human Factors inComputing Systems, ACM Press, 2005, pp.1435�1438.

7. G.R. Hayes et al., CareLog: A Selective Ar-chiving Tool for Behavior Management inSchools, tech. report GIT-GVU-06-20,Georgia Inst. of Technology, 2006, pp. 1�10.

8. P. Howlin, Children with Autism and As-perger Syndrome: A Guide for Practitionersand Careers, John Wiley & Sons, 1998.

9. T. Westeyn, “Recognizing Mimicked Au-tistic Self-Stimulatory Behaviors UsingHMMs,” Proc. 9th IEEE Int’l Symp. Wear-able Computers (ISWC 05), IEEE CS Press,2005, pp. 164�167.

10. G.R. Hayes et al., “Designing CaptureApplications to Support the Education ofChildren with Autism,” Proc. 6th Int’lConf. Ubiquitous Computing (Ubicomp04), Springer, 2004, pp. 161�178.

11. S. Consolvo et al., “Technology for CareNetworks of Elders,” IEEE Pervasive Com-puting, vol. 3, no. 2, 2004, pp. 22�29.

For more information on this or any other comput-ing topic, please visit our Digital Library at www.computer.org/publications/dlib.

JANUARY–MARCH 2007 PERVASIVEcomputing 35

the AUTHORSJulie A. Kientz is a doctoral candidate in the College of Computing and the Graph-ics, Visualization, and Usability Center at the Georgia Institute of Technology. Herresearch focuses on embedded capture and access for data-based decision making.She received her BS in computer science and engineering from the University of To-ledo. She’s a member of the IEEE Computer Society and the ACM. Contact her atthe GVU Center, Georgia Inst. of Technology, Technology Square Research Bldg., 855th St. NW, Atlanta, GA 30332-0760; julie@cc.gatech.edu; www.juliekientz.com.

Gillian R. Hayes is a doctoral candidate in the College of Computing and the GVUCenter at the Georgia Institute of Technology. Her research focuses on enabling cap-ture of rich data in informal and unstructured settings and allowing access to thatdata. She received her BS in computer science and mathematics from Vanderbilt Uni-versity. She’s a member of the IEEE Computer Society and the ACM. Contact her atthe GVU Center, Georgia Inst. of Technology, Technology Square Research Bldg., 855th St. NW, Atlanta, GA 30332-0760; gillian@cc.gatech.edu; www.gillianhayes.com.

Tracy L. Westeyn is a doctoral candidate in the College of Computing and the GVUCenter at the Georgia Institute of Technology. Her primary research areas are on-body sensing, wearable computing, and activity recognition. She received her MS incomputer science from the Georgia Institute of Technology. Contact her at the GVUCenter, Georgia Inst. of Technology, Technology Square Research Bldg., 85 5th St.NW, Atlanta, GA 30332-0760; tracy.westeyn@cc.gatech.edu; www.cc.gatech.edu/ccg/people/tracy.

Thad Starner is an associate professor in the College of Computing and the GVUCenter at the Georgia Institute of Technology. His research focuses on mobilehuman-computer interaction, intelligent agents, computer vision, and pattern re-cognition, and his work focuses on computational assistants for everyday-use wear-able computers. He received his PhD from the MIT Media Laboratory. He’s a mem-ber of the IEEE. Contact him at the GVU Center, Georgia Inst. of Technology,Technology Square Research Bldg., 85 5th St. NW, Atlanta, GA 30332-0760;thad@cc.gatech.edu; www.cc.gatech.edu/~thad.

Gregory D. Abowd is an associate professor in the College of Computing and theGVU Center at the Georgia Institute of Technology and director of the Aware HomeResearch Initiative. His research focuses on an application-driven approach to ubiq-uitous computing concerning both human-computer interaction and software engi-neering. He received his DPhil in computation from the University of Oxford. He’s amember of the IEEE Computer Society and the ACM. Contact him at the GVU Cen-ter, Georgia Inst. of Technology, Technology Square Research Bldg., 85 5th St. NW,Atlanta, GA 30332-0760; abowd@cc.gatech.edu; www.gregoryabowd.com.