MIThril: context-aware computing for daily life

Richard W. DeVaul, Steven J. Schwartz, Alex “Sandy” PentlandThe Media Laboratory

Massachusetts Institute of Technology{rich,schwartz,sandy}@media.mit.edu

May 16, 2001

Abstract

MIThril is a functional, operational body-worncomputing architecture for context-aware human-computer interaction research and general-purposewearable computing applications. It is ergonomic,flexible, and designed to be integrated into ordinarystreet clothing. The MIThril architecture combinesa multi-protocol body bus and body network, inte-grating a range of sensors, interfaces, and comput-ing cores. At the time of this writing, it has been indevelopment for over a year and continually refinedthrough daily use for the last seven months. MIThrilis an open-source hardware/software platform ini-tiative designed to foster collaboration among re-searchers and focus attention on the last frontier ofwearable computing: daily life.

1 Introduction

Wearable computing research has traditionally beenassociated with bulky gear, geeky researchers, andmilitary or industrial applications. Such associa-tions are not unwarranted; prototype hardware andspecialized applications far removed from ordinaryexperience are the norm. However, the last fiveyears has seen an explosion in popularity of portabledigital applications for every-day use, such as cellphones and PDAs. For our work to be relevant, wemust turn our attention to the applications and prob-lems of daily life.

1.1 Applications for Daily Life

The growing popularity of portable consumer elec-tronic devices for communications (cell phones andpagers), logistics and memory (PDAs), shopping(smart cards), athletic performance monitoring,etc.demonstrates the demand for technological solutionsfor the problems of daily life. These applications areusually packaged as separate devices, with little con-

cern for integration, interoperability, or competingdemands for the user’s attention.

We believe that the promise of wearable com-puting is an always on, always available informa-tion resource; an integrated multi-application frame-work that consolidates processing power, mass stor-age, wireless networking, and user interaction, whilebalancing demands on the user’s time and atten-tion. The wearable should be an invisible, reliableassistant oraid-de-campthat understands the user’stask, goals, and preferences, and uses this knowledgeto provide assistance when necessary and otherwisestay out of the way.

In order to realize this vision, the wearable mustbe highly accessible yet unobtrusive, able to inter-rupt the user but careful to do so only under appro-priate circumstances. The interface must maximizethe value of the information provided while mini-mizing the physical and cognitive burdens imposedby accessing it.

1.2 The Need for Context

Implicit in this description is the need for context. Itis impossible to provide situation-appropriate infor-mation without knowledge of the situation. Requir-ing the user to explicitly provide this contextual in-formation (by typing,etc.) places too much of the in-teraction burden on the user. Fortunately, computerperception techniques can be used to determine theuser’s context, trading the burden of explicit interac-tion for a sensing and modeling problem.[10, 3, 6]This tradeoff is not without cost; computer percep-tion requires more bandwidth and processing powerthan a traditional user interface, and presents a num-ber of design problems. In particular, the followingthree questions must be answered for any context-aware application:

1. What type of context (location, activity,etc.) isimportant to the application?

2. How do we sense, classify, and model these

1

features within bandwidth and computing con-straints?

3. How do we best use this information to mini-mize the cognitive and physical burdens on theuser?

Answering these questions requires careful designand empirical testing. Since the appropriate com-bination of sensors, modeling techniques, and inter-action modalities for a particular application cannotbe known in advance, context-aware application de-velopment requires a flexible platform that providesthe widest variety of sensing, computing, and inter-action options.

2 Architecture and Application:MIThril and Memory Glasses

In order to focus our research, we chose a singleclass of application that combines the important fea-tures of context awareness and situation-appropriateinformation delivery: the Memory Glasses. TheMemory Glasses is a proactive, context-aware re-minder delivery system. The user specifies re-minders to be delivered conditioned on context,e.g.“The next time I’m near a grocery store, remind meto get milk,” or “The next time I talk to my advi-sor, remind me to ask for a raise.” The MemoryGlasses represents a broad class of contextual infor-mation delivery applications, ranging from medicalapplications for amnesia and prosopagnosia (face-blindness) to general organization and support fordaily living.

2.1 Memory Glasses Platform Require-ments

Having chosen the Memory Glasses, we turned ourattention to selecting an appropriate platform forits development. The key to making the MemoryGlasses work is the ability to sense and classify auseful range of contexts and the ability to deliver in-formation to the user in a way that does not distractfrom the user’s primary task.

Ergonomics and wearability are as important asthe flexibility, computing power, and bandwidthof the development platform. Bulky or unsightlyequipment that physically or socially encumbers thewearer limits the use and effectiveness of the appli-cation and biases user behavior. Based on our anal-ysis, we drew up a list of requirements for the Mem-ory Glasses research platform:

1. Wearability: As much as possible, the hardwareshould disappear into the user’s ordinary cloth-

ing, leaving only a minimal user interface. Re-liability and ergonomics must be high, weightmust be low, and uptime must be long. Ide-ally, the user puts the system on in the morning,takes it off in the evening, and completely for-gets about it in between except when activelyusing the functionality.

2. Flexibility: The widest possible range of physi-cal and functional configurations should be ac-commodated by the design so that the widestrange of users and behaviors may be studied.Reconfiguring the system should be a simplematter of reconnecting components.

3. Sensing and Bandwidth: A wide range of sen-sors and protocols should be supported, rangingfrom cheap off-the-shelf hardware to custommicrocontroller-based devices. The on-bodysensing bus should provide sufficient band-width and flexibility for the simultaneous useof heterogeneous sensors.

4. Interaction: The system should support a rangeof unobtrusive peripherals, including a light-weight head-mounted display, audio input andoutput, a chording keyboard for text entry,clothing-integrated peripherals such as embroi-dered keypads,etc.These peripherals should beeasily reconfigurable based on application andtask, and should form the only visible or notice-able components of the system, “the interface isthe computer.”

5. Computation and Networking: On-body com-puting power should be low-power, distributedand scalable, supported by a wired peer-to-peerbody network. Since useful resources are likelyto be found off the body as well as on it, the sys-tem should support low- and high-bandwidthwireless networking options,e.g. CDPD cellmodem and 802.11.

6. Open Design: All components of the platform,hardware and software, should be fully publish-able and unencumbered by restrictive licensingthat might discourage collaboration or fosterwasted effort through re-invention.

Evaluated against these requirements, existing re-search platforms are inadequate for developing ourresearch applications. Industrial wearable comput-ers such as the Xybernaut[15] and PC-104 researchwearables like the Lizzy[9] provide sufficient com-puting power and bandwidth, but impose significantsocial and ergonomic burdens on their users. Cur-rent research projects, such as the PLEB[14] andLart[2] are promising hardware development efforts,

2

but are not focused on the specific problems of wear-ability and context-aware computing. Light-weightconsumer portable devices lack flexibility in config-uration and sensing and present limited interactionoptions. None of the options we investigated weresufficient for our needs, hence our decision to de-velop a new platform for body-worn context awarecomputing: MIThril.

3 MIThril Architecture and Im-plementation

The MIThril architecture grew out of our expe-rience with previous wearable platforms, notablythe Lizzy[9] and the Smart Vest[7], as well as in-dustrial wearables like the Xybernaut[15] and con-sumer electronic devices like the Compaq iPAQH3600. Each of these platforms has its strengthsand weaknesses, but none have proven adequate forour present research agenda. MIThril was developedto combine the best features of existing wearablesand meet all of the previously stated requirementsfor a suitable Memory Glasses application develop-ment platform.

3.1 Architectural Overview

The MIThril architecture can be broken down intofive major categories: packaging, body bus and sens-ing, body network and computing, user interface,and software. The clothing-integrated package con-strains the overall form, weight, and mechanicalproperties of the system. The body-bus connectssensors to computing cores, and the body networkconnects computing cores to each other and to off-body networks (user interface and software are self-explanatory). Diagrams of the logical and physicalorganization can be found in Figure 2 and Figure 1,and more implementation detail is available throughthe MIThril project web site[4].

The current “kitchen-sink” MIThril prototype in-cludes one pound (six to eight hours) of lithium-ion batteries, two computing cores (BSEV andCerfBoard with 1GB IBM MicroDrive), an 802.11bridge, two I2C-based sensors (Figure 3), sevenbody-bus cables and two junctions, a body-networkhub, a MicroOptical QVGA clip-on HMD and aTwiddler chording keyboard. The total power con-sumption is at or below four and a half watts whenidle and peaks as high as six or eight during pe-riods of high disk-access and 802.11 use. Addinga light cotton/poly shirt “chassis” brings the totalsystem weight to three and three-quarters pounds,lighter than a typical leather jacket. The next ver-sion of MIThril, expected to be operational by the

Figure 1: One possible MIThril configuration.

Figure 2: MIThril functional schematic.

3

Figure 3: An I2C based three-axis accelerome-ter for MIThril.

Figure 4: The author and his MIThril.

date of this publication, will incorporate the ETHWearARM core[1] and substantially reduce sys-tem weight while improving performance and er-gonomics.

3.2 Packaging

Packaging is an extremely important contributor tooverall wearability, flexibility, and reconfigurability.The first MIThril package was a light-weight meshvest suitable for wearing under a shirt or jacket. Thispackage, though not truly clothing-integrated, pro-vided excellent comfort, access, and configurabil-ity. We have been developing variants of clothing-integrated vest packaging (from Polar fleece outer-wear to dress formal) ever since.

The culmination of our packaging research is azip-in vest liner not unlike our original mesh vest(see Figure 4). This liner is made of a light-weight,cool mesh fabric and structural belting, and is com-patible with a range of outer wear options, from cot-ton shirts to Armani suits. Off-the rack clothing iseasily modified to accommodate the liner with theaddition of a single zipper. Distributing the systemacross the torso maximizes the useful sensor andcomponent real-estate, allowing for the convenient

Figure 5: MIThril soft mount.

Figure 6: MIThril cable anchor.

placement of computing cores, microphones, cam-eras, breathing and heart rate sensors, accelerome-ters, directional tag readers,etc. The package dis-tributes the weight of the system evenly across theshoulders, resulting in a comfortable, balanced feelvery similar to ordinary clothing.

All MIThril components are mounted on the linerusing the “soft mount” system, a combination of Vel-cro and tie-on fabric panels (see Figure 5). The soft-mount system allows components to be easily re-moved and repositioned to accommodate reconfig-urations or washing the liner.

Cables are routed on the body using Velcro andneoprene cable anchors (Figure 6). The anchors holdthe cables securely but are easy to open for main-tenance and fail gracefully under excessive load.Components are placed and cables are routed toavoid the contact-points along the back and shoul-ders, allowing the wearer to sit comfortably, movenaturally, and even fall down without damaging theequipment — as one author inadvertently demon-strated during a cell-phone-induced bike crash (fur-ther discussion in Section 5).

4

Figure 7: MIThril body bus junction.

MIThril’s clothing-integrated design facilitatesthe development of context-aware applicationslike the Memory Glasses by creating a comfort-able, light-weight package virtually indistinguish-able from (and integrated into) the wearer’s existingclothing. This package imposes the minimal socialand physical burdens on the user, while maximizingthe useful on-body real estate.

3.3 Sensors and the MIThril body-bus

On-body connectivity for sensors, peripherals, andpower is provided by the MIThril body bus. Thebody bus, composed of body-bus cables and junc-tions (Figure 7), is a branching single-connectionpower/data bus that provides regulated 5V and un-regulated 12V power, USB, I2C, the Dallas Semi-conductor one-wire protocol, and three currently un-used twisted-pair connections through a stranded 16conductor cable. The body-bus connectors are lock-ing, strain-relieved, high-torque Hirose 3500 con-nectors (similar to those used for cell phones) andare rated for 20,000 insertion cycles.

Through its branching structure and multi-protocol design, the MIThril body bus greatly sim-plifies on-body device placement and cable rout-ing. The connectors and cables are mechanically andelectrically stable, robust, and locking. All compo-nents are flat and small, making it easy to integratethe MIThril body bus with existing clothing or newdesigns. The modular design and robust construc-tion makes mechanical failures unlikely and allowsfor easy reconfiguration and troubleshooting in thefield.

The USB protocol provides medium-to-high (12megabit per second) bandwidth and compatibilitywith a range of off-the-shelf USB cameras and mi-crophones. However, the USB protocol is complexto implement for custom sensors and peripherals.

At the time of this writing, we have functioningUSB support for a number of off-the-shelf camerasand audio devices, including the D-Link DSB-C300camera (OmniVision OV511 chipset) and audio de-vices by Tenex.

We selected the I2C protocol to compliment USBon the body-bus. The Philips Inter-IC or I2Cprotocol[8] is a multi-device two-wire serial proto-col commonly used in industrial and embedded ap-plications. The I2C protocol provides less band-width than USB (400 kbps in High-speed mode)but is easier to implement and suitable for a wide-range of low-bandwidth microcontroller-based sens-ing applications. We have implemented a numberof I2C-based MIThril sensors, including an IR tagreader (for use with the Crystal tag system[13, 12]),a three-axis accelerometer, a smart battery board forhot-swappable battery power, and an I2C to RS232Cconverter for low-bandwidth legacy serial devices.

The Dallas Semiconductor one-wire protocol isused by a variety of single-chip devices, includ-ing unique ID tags, temperature sensors, and bat-tery monitor/chargers. At the time of this writing,we have a microcontroller-based implementation ofthe Dallas Semiconductor protocol and are workingon an FPGA-based “native” implementation for theBSEV computing core, described in Section 3.4 be-low.

The body-bus design facilitates the developmentof Memory-Glasses applications by providing an in-tegrated framework for combining a range of customand off-the shelf sensors with an aggregate usablebandwidth of approximately 10 megabits per second.Future high-bandwidth context-sensing applications(on the order of 100 megabits per second) will besupported by running an additional high-speed pro-tocol such as IEEE 1394 or USB2 over the undesig-nated twisted-pair body-bus conductors.

3.4 Computing cores and the MIThrilbody network

On-body computing resources are provided bya peer-to-peer network of one or more high-performance low-power computing cores. The onlyrequirement the MIThril body network imposes onthe choice of computing core is that all devices mustbe capable of 10 megabit per second Ethernet andrun off of 5V regulated power. (Ergonomics andpower-consumption impose additional practical con-straints on the choice of computing cores.) In addi-tion, at least one core must support the the MIThrilbody network and provide the required user interac-tion resources.

The MIThril body network combines 10 megabitper second Ethernet and 5V regulated power, pro-

5

Figure 8: A MIThril body-net hub.

viding a single-connector power/data connection toall body-network devices. A 10-base Ethernet hubmodified to distribute power as well as data (See Fig-ure 8) ties the computing cores together, and pro-vides the option of bridging the body network towired off-body networks through a crossover cable.

The current MIThril prototype employs threecores: the BSEV core, the CerfBoard, and an 802.11wireless bridge, which are described in more de-tail below. However, the flexibility of the bodynetwork provides a range of choices, from dedi-cated video or audio processors to specialized net-work devices. This flexibility allows performance,bandwidth, and power consumption to be tailoredto the specific needs of the application,e.g.a high-bandwidth Memory Glasses implementation mightemploy multiple body-bus enabled cores to increaseon-body sensing bandwidth, whereas a simple low-bandwidth implementation might employ a singlecomputing core and no other body-network devices.

The BSEV core is a combination of theipEngine1, an MPC823 based single-board com-puter from Brightstar Engineering, and a MIThrilbody-bus/video driver board (See Section 3.5) thatprovides the body-bus terminus and drives the headmounted display. The ipEngine1 provides a 66MIPS MPC823 (a PowerPC-derived processor withintegrated peripherals), 16 MB of RAM and anAltera FPGA. Combined with the body-bus/videodriver board, the result is a moderately powerful andvery flexible sensing and user interaction computingcore that consumes less than two and a half watts ofpower. The BSEV core with display and networkconsumes less than three watts of power.

The SA 1110 Strong-Arm based Intrinsyc Cerf-Board is a low-power high-performance computingresource, and in combination with a one gigabyteIBM Microdrive acts as the on-body NFS file server.The CerfBoard provides 200 MIPS of computing

power and 32 MB of RAM, allowing it to run com-plex applications. The total power consumption ofthe CerfBoard with Microdrive ranges from less thanone watt to as high as three depending on disk activ-ity.

The 802.11 bridge is an repackaged Lucent Wave-LAN Ethernet converter and Orinoco card, whichprovides 11 megabit per second wireless network-ing. Due to the high power consumption of this de-vice (a sustained two watts regardless of network ac-tivity) we are investigating the possibility of using asecond CerfBoard and a compact-flash based 802.11interface, such as the Symbol LA4137.

We are collaborating with Urs Anliker, PaulLukowicz, and Gerhard Troester of the WearableComputing Lab at ETH in Zurich in the developmentof the WearARM, a modular high-performance low-power computing core which will provide a super-set of the functionality of the Brightstar ipEngine1,the CerfBoard, and the 802.11 bridge. When com-bined with a body-bus/video-driver board similar tothe one currently in use in the BSEV core, it willprovided a single-core solution for a wide range ofMIThril applications. At the time of this writing, theWearARM core exists as functioning hardware buthas yet to be integrated into a functioning MIThrilsystem.

The scalable, heterogeneous computing environ-ment provided by the MIThril body network allowsthe tailoring of computing resources to meet appli-cation needs. For instance, a simple, low-bandwidthimplementation of the Memory Glasses might relyon a a single BSEV computing core, whereas ahigher bandwidth implementation might rely on ded-icated vision or audio processors,

3.5 User Interface Hardware

The choice of user interface is extremely important,since it is only through the interface that the user ex-periences the functionality of the application. Theuser interface places the highest demands on theuser’s attention and represents the user’s primary ex-perience of the system as a whole. An importantconstraint on interface design is the functionality andergonomics of the interface devices; the MIThril ar-chitecture provides both visual and auditory chan-nels for delivering information to the wearer, and au-dio and key-entry based explicit input.

We chose the the MicroOptical Clip-on QVGA,which is based on the Kopin CyberDisplay 320Color chip, to be the MIThril head-mounted display.This is a small, light display that clips to the wearer’sglasses and provides a full-color quarter-VGA dis-play. The MicroOptical is unobtrusive and obscureslittle of the user’s field of view.

6

Figure 9: MicroOptical clip-on without IR tagreader.

Figure 10: Brightstar ipEngine1 with bodybus/video driver board.

The BSEV body bus/video driver board uses thedirect output of the MPC823’s on-board LCD driver,bypassing VGA signal conversion. This eliminatesthe need for the bulky VGA driver board and re-duces power consumption by almost four watts. Inaddition, the display driver provides direct control ofbacklights and signal timing, allowing for software-switching between 60 Hz field-sequential color and180 Hz high-brightness gray-scale modes.

A range of USB audio devices, such as the micro-phone/speaker headset made by Tenex, may be usedfor audio input and output. For key entry, our currentMIThril implementation allows for hot-swappingbetween a modified Twiddler chording keyboard anda modified Palm folding keyboard, and a range ofoff-the-shelf USB or custom microcontroller-basedI2C devices may also be used.

The flexibility provided by MIThril’s interfacehardware provides a solid platform for multi-modalinterface design.

4 System Software

We chose Linux as the operating system for MIThrilbecause it is open source, has a large and active de-

velopment community, and is widely deployed inthe embedded device market. Linux provides excel-lent performance and a fully integrated native de-velopment environment. Excellent package man-agement is available through the Debian packagetools, and literally thousands of software packagesare available pre-compiled for both the StrongARMand PowerPC.

We are currently using the Debian Arm distribu-tion for the CerfBoard and WearARM, and a mod-ified Brightstar/Hardhat distribution for the BSEVcore, though we are in the process of moving to De-bian for that platform as well. The result is a consis-tent, high-performance operating environment withexcellent development tools.

In addition to device drivers for the MicroOpti-cal, I2C and USB, we have focused on the devel-opment of a prototype Memory Glasses application,utilizing the MicroOptical display for output, the IRtag reader for localization and object/person recog-nition, and the three-axis accelerometer for motiondetection and classification. We are in the process ofintegrating mixed audio-visual feedback and voicerecording for spoken reminders.

5 Evaluation and Experience

The success of the MIThril architecture can be mea-sured both in terms of the goals we set for its devel-opment and in terms of what we have learned in theprocess of developing the system.

At the time of this writing, MIThril has been indaily use for over half a year, during which time wehave greatly refined its packaging, ergonomics, reli-ability, sensor/peripheral support, software, and gen-eral ease-of-use. We have learned a great deal aboutengineering reliability and ergonomics in “soft”body-worn computing systems.

One of the first lessons learned was the impor-tance of strain relief and high reliability connec-tors for on-body connections. Unlike a hard chas-sis, clothing is constantly flexing and shearing withuse. Although the instantaneous torque placed onbody-bus connectors is ordinarily low, the cumula-tive strain from constant motion will cause failuresin low-torque-rated low-duty-cycle connectors. Ourfirst MIThril prototype was retired after less than ayear of use due to connector failure, prompting acomplete reengineering of the body-bus connectorsystem.

Another lesson is the importance of high-qualitydesign and fabrication of the textile components ofthe system. Although “soft,” the textile packag-ing must be carefully designed and fabricated tomeet demanding engineering, comfort, and fashion

7

requirements. For instance, the MIThril vest linercombines multiple fabric types and structural beltingto maintain its shape and support the equipment loadwithout sagging or hanging away from the body.Many design iterations went into development of thesoft-mount system and the placement and routing ofcomponents.

Two particularly interesting and seemingly uncon-nected hypothesis found support in an unplanned ex-periment involving the author, a cell phone, a pro-totype MIThril system with functioning 3-axis ac-celerometer, and a bicycle. The first hypothesis isthat placing relatively small, low-aspect-ratio “hard”parts near to front and sides of the torso would pro-tect them without the addition of a hard shell or extrapadding. The second is that sudden interrupts (suchas answering a ringing cell phone) are far more dis-ruptive than constant low-level demands on one’s at-tention, such as a small HMD in one’s peripheral vi-sion displaying a constant stream of accelerometerdata1.

The results of this experiment included minorbruises and contusions, a slightly damaged bicy-cle, and a completely unscathed MIThril system thatcontinued to log accelerometer data throughout theduration of the ride and crash. Perhaps this data setwill one day be used to train context-aware applica-tions to recognize the important “wearer experienc-ing cell-phone induced bike crash” context.

In addition to unintentional experiments in ter-minal ballistics, we are collaborating with theRochester Center for Future Health to deploy andtest a Memory Glasses implementation at the Med-ical Smart Home[11] to study the benefits and dif-ficulties of building Memory Glasses applicationsfor amnesia and agnosia patients, as well as healthyunimpaired research subjects.

Several simple Memory Glasses demos have beendeveloped, with more sophisticated applications onthe way. MIThril is the primary development andtarget platform for the Enchantment[5] wearableuser interface framework.

MIThril is also in full-time use as a general-purpose wearable computer, supporting daily usefunctions such as recording grocery lists and movierecommendations, conversational note taking, mes-saging and email, and the writing and typesetting ofthis paper in LATEX.

1It should be noted that the author has ridden his bicycle manytimes before and since while wearing a MIThril system withoutcell-phone interruptions and without ill effect.

6 Conclusions

After almost a year of development, the MIThrilarchitecture has prove itself useful as a platformfor the development and testing of Memory Glassesapplications and general-purpose wearable comput-ing research. We have learned a great deal aboutbody-worn systems engineering and the develop-ment of context-aware applications. The modular-ity and flexibility of the MIThril architecture hasallowed us to continually refine our design and in-corporate new features, with many exciting develop-ments (such as the inclusion of the ETH WearARMcore) just around the corner. Our work with ETH andthe Rochester Center for Future Health demonstratesMIThril’s value in fostering collaboration, and weare working to encourage this trend by making thefull plans of MIThril freely available on the worldwide web.

Having developed MIThril to the point of usabil-ity, our real work is only now just beginning. As weturn our attention toward the research applicationsfor which MIThril was developed, its ultimate valuewill be measured in the degree to which it facilitatesthe exploration of the last frontier of wearable com-puting: the domain of everyday life.

References

[1] Urs Anliker, Paul Lukowicz, ClaudioJeker, Steve Schwartz, and AndreasThiel. Weararm: Modular low-powerwearable based on the strongarm proces-sor. http://www.wearable.ethz.ch/projects/research/weararm.html, 2001.

[2] Jan-Derk Bakker, Erik Mouw, Marc Joosen,and Johan Pouwelse˙ The LART pages.http://www.lart.tudelft.nl/, 2001.

[3] Brian P. Clarkson. Recognizing user’s contextfrom wearable sensors. Technical Report 519,Vision and Modeling Group, MIT Media Lab,January 2000.

[4] Richard W. DeVaul, Steven J. Schwartz, andAlex ”Sandy” Pentland. The MIThril projectwebsite. http://www.media.mit.edu/wearables/mithril/, 2001.

[5] Xybernaut Inc. Xybernaut mobile assistant.http://www.xybernaut.com/, 2001.

[6] Edward Keyes. The Enchantment win-dow manager: A wearable computing userinterface, http://albireo.mit.edu/enchantment.html. Technical report, MIT,2000.

8

[7] Jason Pascoe. Adding generic contextual ca-pabilities to wearable computers˙ InDigest ofPapers. Second International Symposium onWearable Computers, pages 92–99. IEEE, Oc-tober 1998.

[8] Steven J. Schwartz and Alex Pentland. Thesmart vest: Towards a next generation wearablecomputing platform. Technical report, MITMedia Lab, 1999.

[9] Philips Semiconductors. The i2c-bus specifica-tion version 2.1. Technical report, 2001.

[10] Thad Starner. Lizzy: MIT’swearable computer design 2.0.5.http://wearables.www.media.mit.edu/projects/wearables/lizzy/, 1999.

[11] Thad Starner, Bernt Schiele, and Alex Pent-land. Visual context awareness in wearablecomputing. InDigest of Papers. Second Inter-national Symposium on Wearable Computers,pages 50–57. IEEE Computer Society, October1998.

[12] Rochester University. Center forfuture health: Advanced medi-cal technology on a human scale.http://www.futurehealth.rochester.edu:8080/.

[13] Roy Want, Andy Hopper, Veronica Falccao,and Jonathan Gibbons. The active badge loca-tion system.ACM Transactions on InformationSystems, 10(1):91–102, January 1992.

[14] Josh Weaver, Ari Y. Benbasat, andRichard W. DeVaul. The crystal tag sys-tem. http://www.media.mit.edu/∼rich/, 1999.

[15] Adam Gordon Wiggins. The PLEB/photonproject home page. http://www.cse.unsw.edu.au/∼pleb/projects.html,2001.

9