August 2004 97

I N V I S I B L E C O M P U T I N G

The emergence of convergedmobile devices with a widerange of computing, commu-nications, entertainment, andsensing capabilities repre-

sents a major step in the evolution ofwireless computing. Such devicesincreasingly shift the decision-mak-ing power from the user to themachine, which has the capacity tobe better informed about the currentenvironment and can respond morequickly.

People spend much of their timeplanning where they’re going, what todo along the way, and recalling wherethey’ve been. Trinity College Dublin’sHermes project (http://hermes.dsg.cs.tcd.ie) is exploring innovativeways to enhance these activities withproactive, context-aware mobile appli-cations.

Drawing on previous work with theEU-funded Global Smart Spaces pro-ject (www.gloss.cs.strath.ac.uk), weuse the notion of a trail to capture amobile user’s daily activities. At themost general level, a trail is a collectionof locations, together with associatedinformation about these locations anda recommended order for visitingthem.

The trails metaphor makes it possi-ble to explore adaptive characteristicscommon to all mobile, context-awareapplications. Adaptation in this con-text involves altering the set of interestpoints on a trail and their visiting orderwith timeliness, accuracy, and rele-vance while remaining in tune withuser expectations.

PROPERTIESThe most common and obvious

characteristic of an adaptive trails-based application is providing supportfor planning an appropriate route. Thisinvolves processing a rich set of con-textual information such as currenttraffic congestion along different road-ways, mode of transport, weather,route characteristics such as scenic orfastest, crowd density, time of day, andso on.

Activities at different locations alongthe trail also influence the aggressive-ness of application adaptation. If thetasks are deemed mandatory or highpriority, the application should makeit difficult for the user to ignore them.If the tasks are optional, more subtleadaptation is appropriate.

Reliable and timely dissemination ofcontextual information and services isanother key property of adaptableapplications. A mobile device withsensing capabilities can determine theinformation and services available inthe current environment and dissemi-nate what is appropriate for the user’sactivity.

Trails-based applications can alsoincorporate an element of team-based

communication or cooperative work.To this end, models that support adhoc communication between mobileusers provide a useful feature.

APPLICATIONSMobile, context-aware trails-based

applications range from single-usersystems that focus on individual dailyactivities to multimedia groupwarethat supports a wide collection of userrequirements.

For example, a simple route plannercan help a business traveler in an unfa-miliar city get from the airport to ameeting location. The application,deployed on a PDA, calculates the bestroute based on factors including theuser’s current location, personal trans-portation preferences, real-time roadtraffic data, and bus and train sched-ules. It also presents a digital map ofthe area, overlaid with currently feasi-ble routes and their estimated traver-sal time. Dynamic route adaptationmay occur if traffic conditions make adifferent mode of transport or routemore appropriate.

Other applications can manage theactivities of multiple users. For exam-ple, a delivery company can equipoperatives on bicycles or motorcycleswith PDAs running a trails-basedinformation system administered bystaff in a traditional office. Admini-strators receive client requests andassign them to mobile couriers basedon their current location, mode oftransport, workload, schedule, andshift end time.

The application notifies couriers ofnew jobs via vibrotactile alerts and dis-plays a trail on their PDA showing the

Context-AwareTrailsSiobhán Clarke and Cormac Driver, Trinity College Dublin

The trails metaphor makes it possible to exploreadaptive characteristicscommon to all mobile,context-aware applications.

98 Computer

I n v i s i b l e C o m p u t i n g

optimal route from their current loca-tion to the package pickup point andon to the drop-off point, as well as tex-tual and audio information about thejob. Dynamic route adaptation mayoccur based on factors such as assign-ment urgency and traffic conditions,and courier monitoring is also possible.

A more relaxed trail, such as thatshown in Figure 1, is appropriate tosupport tourists’ interests. For exam-ple, we are currently developing a con-text-based application that dissemi-nates multimedia-based “stories”about previous inhabitants of an oldpart of Dublin to mobile users.

Finally, trails-based applications canbe used to support multiplayer games.For example, consider a treasure-huntgame involving teams of PDA-equippedmobile players dispersed throughout acity or other large ubiquitous comput-ing space. The game’s objective is toarrive at an initially unknown destina-tion as fast as possible by solving a setof cryptic riddles. Each clue hints at alocation, which contains the next clue.Clues are tailored to team members’interests and vary in difficulty, delivery

format (video, audio, or text), language,and topic.

The game uses trails to determineoptimal routes between clue locationsin real time, based on mode of trans-port, street congestion, team progress(very successful teams may be hinderedslightly to keep the game interesting),and the location of other team mem-bers (if, say, two or more team mem-bers are required to reveal the clue), aswell as more typical contexts such aslocation, weather, and time.

CHALLENGESThere have been significant advances

in the understanding of human-com-puter interaction as users performtasks of greater variety and complex-ity (J.M. Carroll, Human-ComputerInteraction in the New Millennium,Addison-Wesley, 2002). However, theform factor and limited interface capa-bilities of mobile devices such as PDAsinhibit the representation of thedynamic adaptation of trails-basedapplications.

In addition, poor data quality com-plicates context acquisition. Con-

textual information may be stale oruncertain in terms of fuzziness andtrust, while power and signal strengthlimitations, together with frequent usermobility, also contribute to its unrelia-bility. Frequent re-adaptation to undoprevious bad decisions is unacceptable.Sensor data fusion, a large researchfield, addresses these and related issues.

Further, mobile networks requirenew protocols for reliable and consis-tent sharing of information amonggroups of users who come together in an ad hoc fashion (J. Wu and I.Stojmenovic, “Ad Hoc Networks,”Computer, Feb. 2004, pp. 29-31).

Moreover, the limited processing andpower capabilities of PDAs challengealgorithm design for dynamic adapta-tion. A balance must be found betweenthe time it takes to reconfigure the cur-rent trail and application responsive-ness. The Lancaster GUIDE project(www.guide.lancs.ac.uk) provides agood example of a mobile, context-sen-sitive system experiencing this problem.

Determining whether the user is fol-lowing a similar or previous path canenhance trail reconfiguration. Patternrecognition may be used to discoversuch situations, though the inevitableinexactness of repeated user behaviorpresents challenges for recognition andmatching algorithms. The journalIEEE Transactions on Pattern Analysisand Machine Intelligence explores theseand related issues.

Finally, a recent Invisible Computingcolumn (B. Schilit et al., “Wireless Lo-cation Privacy Protection,” Computer,Dec. 2003, pp. 135-137) outlined sev-eral privacy protection challenges thatapply to trails-based applications.

TECHNOLOGYIn collaboration with Intel Re-

search, we are developing a genericsoftware framework that addressesthese challenges and provides a con-figurable platform for developers ofmobile, context-aware applications.Building such a framework involvesboth exploring and creating appropri-ate technologies.

Figure 1. Context-aware trail application for mobile users. A recommended tourist trailthrough the Trinity Dublin Campus includes optional high-priority and time-constrainedactivities.

August 2004 99

• generating a highly customizedoptimal trail, and

• application responsiveness.

The first requirement negates theusefulness of classic route-generationtechniques, such as shortest-path hillclimbing, which do not consider users’personal route-planning preferences.Current best practices in route plan-ning are knowledge- and experience-based. These approaches generateroutes most likely to conform to theuser’s current preference.

For unfamiliar situations, proposedcollaborative case-based approachesgenerate a route based on similar users’experiences. Notwithstanding thesemantic difficulty of identifying simi-larities among different users, thesealgorithms’ recursive nature furtherstrains system resources and therebyslows application responsiveness.

Trail reconfiguration quality can beenhanced by pattern matching recentcontextual data—for example, loca-tion, user activity, and time—against adatabase of previous trails. This is sim-ilar to established pattern-matchingtechniques such as Lemple-Ziv com-

Sensor fusionConsider, for example, classical sen-

sor fusion technology, which has diffi-culty satisfying context-sensing re-quirements because

• source configuration is highlydynamic,

• inexpensive sensors are inferior tohuman perceptual capabilities,and

• sensor numerical parameters mustbe converted to meaningfulsemantics.

Among the proposed solutions tothis problem include those based onDempster-Schafer theory, a generalizedform of Bayesian theory. In addition,many projects have addressed how tofuse the output of numerous, low-cost,multimodal sensors for context aware-ness, including Technology EnablingAwareness (www.teco.edu/tea), Media-Cup (http://mediacup.teco.edu), andSmart-Its (www.smart-its.org).

Trail generation The trail-generation process involves

two conflicting requirements:

pression as well as memory referencestring prediction. A solution based onthe later technology would require aprobabilistic model of which trails aremost likely.

M obile, context-aware applica-tions have far to go to realizetheir potential. Progress will

remain slow if developers must con-front the full set of technical challengesfor each individual application. Anapplication framework that provides areusable, configurable set of solutionsto such challenges will give developersa head start. �

Siobhán Clarke is a lecturer in theDepartment of Computer Science atTrinity College Dublin. Contact her atsiobhan.clarke@cs.tcd.ie.

Cormac Driver is a PhD candidate inthe Department of Computer Scienceat Trinity College Dublin. Contact himat cormac.driver@cs.tcd.ie.

Editor: Bill Schilit, Intel Research, Seattle; bill.schilit@intel.com

Get accessto individual IEEE Computer Society documents online.

More than 100,000 articles and conference papers available!

$9US per article for members

$19US for nonmembers

www.computer.org/publications/dlib