DOT/FAA/AM-02/22 Peter M. Moertl University of Oklahoma John M. Canning White Oak Technologies Scott D. Gronlund University of Oklahoma Michael R.P. Dougherty University of Maryland Joakim Johansson Ericsson, Inc. Scott H. Mills SBC Technology Resources, Inc. December 2002 Final Report This document is available to the public through the National Technical Information Service, Springfield, VA 22161. Aiding Planning in Air Traffic Control: An Experimental Investigation of the Effects of Perceptual Information Integration Office of Aerospace Medicine Washington, DC 20591
Transcript
DOT/FAA/AM-02/22
Peter M. MoertlUniversity of Oklahoma
John M. CanningWhite Oak Technologies
Scott D. GronlundUniversity of Oklahoma
Michael R.P. DoughertyUniversity of Maryland
Joakim JohanssonEricsson, Inc.
Scott H. MillsSBC Technology Resources, Inc.
December 2002
Final Report
This document is available to the publicthrough the National Technical InformationService, Springfield, VA 22161.
Aiding Planning in Air TrafficControl: An ExperimentalInvestigation of the Effects ofPerceptual Information Integration
Office of Aerospace MedicineWashington, DC 20591
N O T I C E
This document is disseminated under the sponsorship ofthe U.S. Department of Transportation in the interest of
information exchange. The United States Governmentassumes no liability for the contents thereof.
9. Performing Organization Name and Address 10. Work Unit No. (TRAIS)1 University ofOklahomaNorman, OK 73109
2 White Oak Technologies,Inc.Silver Springs, MD 20910
3 University of MarylandCollege Park, MD 20742
4 Ericsson, Inc.Stockholm, Sweden
5 SBC TechnologyResources,Inc.Austin, TX 78759
11. Contract or Grant No.
Grant #97-G-03712. Sponsoring Agency name and Address 13. Type of Report and Period Covered
Office of Aerospace MedicineFederal Aviation Administration800 Independence Ave., S.W.Washington, DC 20591 14. Sponsoring Agency Code
15. Supplemental Notes
Work was accomplished under approved subtask AM-A-00-HRR-522. Carol Manning (AAM-500) served as theCOTR.16. Abstract
Prior research examined how controllers plan in their traditional environment and identified various informationuncertainty by perceptually representing important constraints. This included integrating spatial information onthe radar screen with discrete information (planned sequences of air traffic). Canning et al. (1999) and Moertl etal. (2000) reported improved planning performance and decreased workload in the planning aid condition. Thepurpose of this paper was to determine the source of these performance improvements. Analysis of computerinteractions using loglinear modeling showed that the planning interface led to less repetitive, but moreintegrated, information retrieval gave rise to the performance improvements. Potential applications of thisresearch include the design and evaluation of interface automation that keeps users in active control bymodification of perceptual task characteristics.
17. Key Words 18. Distribution Statement
Information Integration, Air Traffic Control, PlanningAids, Strip Marking
Document is available to the public throughthe National Technical Information Service,Springfield, VA 22161
19. Security Classif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price
Unclassified Unclassified 12Form DOT F 1700.7 (8-72) Reproduction of completed page authorized
iii
ACKNOWLEDGMENTS
The support of the Federal Aviation Administration under agreement 97-G-037 is gratefully
acknowledged. We thank Frank Durso (Texas Tech University) and Robert Terry (University of
Oklahoma) for their methodological expertise, and Carol Manning and Scott Shappell (FAA-Civil
Aerospace Medical Institute) for their comments on this manuscript. We also extend our thanks
to Ron Andrei and Paul DeBenedittis (of the FAA Academy) and Duane Duke and Rob Enos ( FAA
Office of Information Services) for their expertise.
1
AIDING PLANNING IN AIR TRAFFIC CONTROL: AN EXPERIMENTAL
INVESTIGATION OF THE EFFECTS OF PERCEPTUAL INFORMATION INTEGRATION
Air traffic control equipment has changed in recentyears as the Federal Aviation Administration (FAA)adapts its procedures to the growing volume of airtraffic across the country. However, two major com-ponents of control equipment have stayed constantover the years. Specifically, generations of air trafficcontrollers have utilized a radar screen and flightprogress strips as separate representations of aircraftentering their controlled sector and cognitively inte-grated those representations. This equipment hasproven to be highly beneficial and, therefore, formsthe foundation from which any innovation to the airtraffic control system should begin.
If controllers are to manage the increasing volumesof air traffic, planning will be of increasing impor-tance. This is evidenced by recent efforts to providecontrollers with plan-aiding technology (e.g., URET,User Request Evaluation Tool, Arthur & McLaughlin,1998; CTAS, Center-Tracon Automation System,Denery, & Erzberger, 1995; ERATO, En Route AirTraffic Organizer, Bressolle, Benhacene, Boudes, &Parise, 2000). Such interfaces offer additional func-tions to the controller like conflict detection algo-rithms (URET), automated traffic advisory functionsfor descending, sequencing, and spacing aircraft(CTAS), and decision aid tools like filtering optionsand problem reminders (ERATO). The approach wetook in this study can enhance plan-aiding technologyby identifying essential informational elements thatsupport air traffic planning and determining the ex-tent to which air traffic planning could be improvedby optimizing the representation of that information.
Air traffic controllers manage a complex flow ofaircraft through their airspace. They maintain strictrules of separation between the aircraft while allowingall aircraft to reach their destinations as safely andexpeditiously as possible. In planning the routes forthe aircraft, two forms of planning can be distin-guished. Controllers make tactical plans when theymake decisions that relate to the current moment andinvolve the separation of (usually) pairs of aircraft thatcould soon violate the separation rules and hence,need immediate action. They make strategic planswhen their plans span longer periods of time (about10 minutes or longer) and typically involve multipleaircraft. An examination of strategic planning in airtraffic control is timely, given future concepts being
proposed. For instance, there have been discussionsregarding the creation of a strategic controller posi-tion (N. Lawson & K. Thompson, personal commu-nication, Dec. 15, 1997; see also Vivona, Ballin,Green, Bach, & McNally, 1996). The proposal pro-vides for one person who would be responsible for amultiple-sector airspace, making decisions about traf-fic in that airspace, and delegating responsibility fortactical decisions to sector-level controllers. A goal ofour project was to develop interface tools for a strate-gic controller position.
Dougherty, Gronlund, Durso, Canning, and Mills(1999) studied how air traffic controllers make strate-gic plans for en route traffic (high altitude, high speedtraffic between destinations) using the radar screenand paper flight progress strips. They identified air-craft sequencing for approach to a common destina-tion as a strategic planning task by analyzing controllerverbalizations and use of flight progress strips. Thespecific sequence of a group of aircraft is determinedby many factors– aircraft speed, altitude, destination,and airspace restrictions. Therefore, sequencing air-craft was a complex cognitive task that involved theconsideration of many aircraft over an extended pe-riod of time. Dougherty et al. (1999) argued thatcontrollers could benefit from an interface that sup-ported planning the sequences of aircraft.
We begin by outlining the relevant aspects of thetraditional air traffic control environment that guidedour interface design. Following that, we describe theelectronic planning aid and outline its design prin-ciples. Finally, we report the evaluation of the plan-ning aid by comparing participants’ planningperformance using the interface to their performancein the traditional air traffic control environment.
Traditional En Route Air Traffic ControlEnvironment
Air traffic controllers primarily use informationfrom two different sources, the radar screen and theflight progress strips. The radar screen shows thespatial position and progress of aircraft, together withsome characteristics of the controlled sector (e.g.,boundaries and airways). The radar screen displaysthe spatial location of aircraft as well as the most vitalflight information (identifiers, altitude, speed, andsometimes, flight destination). Discrete information
2
about an aircraft (e.g., its destination, flight origin,planned altitude) can be read from the strips (seeFederal Aviation Administration, 1995; this informa-tion also is available from the Computer ReadoutDisplay). Flight progress strips are small paper stripsthat are computer-printed and posted next to thecontroller about 20 minutes prior to arrival of thatflight in the sector. Strips also are used to indicate theprogress of traffic, as controllers manually update thestrips when information about an aircraft changes. Inaddition, strips carry identifiers that link them to theirrepresentations on the radar screen. However, thelinkage is not direct because controllers have to searchfor the aircraft on the radar screen to coordinate therepresentations. The radar is of primary importancebecause it tells the controllers where the aircraft cur-rently are. The strips can be useful for determiningwhere the aircraft are going to be.
Dougherty et al. (1999) described how controllersuse strips for strategic planning purposes. Controllersoften classified aircraft into logical groups. Thesegroupings were based on flight plan characteristics(e.g., destination, route, or altitude), and/or flowrestrictions. Then strips were ordered to show theplanned sequence for these aircraft. In other words,controllers used strips as planning aids with whichthey externally represented their planned sequences.For example, if a controller wanted to plan a sequenceof aircraft A, B, and C, the controller would arrangethe three strips in that order. In this way, the stripsphysically hold the planned sequence of aircraft. Incontrast, the radar screen represents the actual airtraffic situation. It is important to note that theplanned and current sequences of aircraft are notlinked to one another, and changes on the radar screenare not reflected by changes to the strips.
The discrepancy between the planned sequencerepresented by the strips and the positions of theaircraft on the radar screen is essential information fora controller. It makes apparent that something mustbe done to update the intended path of the aircraft. Inorder to assess the discrepancy between the plannedand current sequence, controllers have to locate thestrip and the corresponding aircraft representation onthe radar screen and mentally link those representa-tions to one another. In this way, they determine themismatch between the planned sequence and thecurrent air traffic situation. In other words, the con-troller is responsible for linking spatial information(location of the aircraft in the sector, distance to otheraircraft, distance to specific points in the sector) todiscrete information (flight identifier, optimal se-quence of aircraft, flight plan, and aircraft type).
In order to help link the spatial and discrete aircraftrepresentations, controllers have to “declutter” the radarscreen. A block of information (the datablock) is con-nected to each aircraft on the radar containing importantflight information about that aircraft. As more and moreaircraft come into a sector, datablocks overlap and be-come illegible. Therefore, controllers often adjustdatablocks into one of eight different positions aroundthe aircraft target and also adjust the distance betweenthe datablock and its aircraft. However, some informa-tion on the datablock is redundant; controllers can getthe same information from strips by matching a uniqueidentifier on the strip to the radar. Therefore, declutteringthe radar screen to make aircraft distinguishable could beviewed as a hindrance to more important tasks likestrategic planning. Though controllers can utilize radardecluttering to create memory cues, a perceptually-improved interface might supplant the memory de-mands. The necessity for decluttering seems to be causedby the separation of the different information sources.
We addressed this problem by designing a newinterface that took over the task of linking spatial anddiscrete information. The interface should allow con-trollers to perceptually assess the difference betweenthe current air traffic sequence and the planned sequencewithout having to frequently declutter the radar screen.We discuss this electronic planning aid next.
Electronic Planning AidThe electronic planning aid changed the task of
planning air traffic sequences by perceptually repre-senting a major constraint in the controllers’ environ-ment that was not visible in the traditional environment(see http://www.ou.edu/cas/psychology/HTIC/FOPA/ for screenshots of the interface). It linkeddiscrete information to its spatial representation onthe radar screen. This dynamic linkage between theplanning aid and the radar screen was implemented bypresenting the participants with two computer screens:Figure 1 shows a snapshot of the two screens.
The planning aid was linked to the radar screen andthe controller could move the cursor between the twoscreens. Aircraft in a sector were represented on theplanning aid by rectangular tokens as shown in Figure1. The tokens contained the aircraft’s identificationnumber, destination airport, equipment type, andaltitude. Aircraft tokens were marked and unmarkedby clicking on them, in which case a check marksymbol appeared or disappeared on the token. Allaircraft were represented on both screens and could beperceptually located on one screen by moving thecursor over the token on the planning aid or byclicking on its target on the radar screen.
3
The planning aid allowed participants to categorizeaircraft and updated that representation on the radarscreen in the color corresponding to their category.Figure 1 shows two aircraft (UAL 755 and DAL80)that a controller placed into different categories. Be-cause each column of aircraft was in different colors,participants do not have to encode and remember thecategories of each aircraft after they were categorizedbecause they were represented perceptually on theradar screen.
The planning aid also allowed the controller toautomatically sort categorized air traffic according totime or distance to specific points (fixes) along theroute. This information was displayed adjacent toeach aircraft token. Note that this automatic sortingdid not include higher-level conflict information; itsimply was based on distance/time measures and there-fore provided only an initial approximation of aircraftorder. These initial sequences needed manual updat-ing and checking. Participants also could get distance
information between points on the radar screen byusing a distance-measuring tool. This tool was similarto how controllers measure distance on traditionalradar screens.
One essential consequence of this design was thata planned sequence could be perceptually comparedwith the current sequence by sliding the cursor acrossthe sequence on the planning screen. This allowed thecontroller to observe how the sequenced position ofeach aircraft corresponded to its current position onthe radar screen. Any discrepancy between the plannedsequence and the current sequence was therefore madeperceptually salient to the controller. If the plannedsequence differed from the current sequence, thisdiscrepancy signaled the need for modifications toeither the planned sequence or to an aircraft’s path.The discrepancy represented an important constraintas it guided the controller toward the aspects of thesituation where control interventions were needed.
Figure 1. The dynamic linkage between the planner screen and the radar screen. On the radar, everydiamond-shaped aircraft representation is linked to a datablock of flight information (in order from top, leftto right: aircraft identification, altitude in 100 feet, computer identification number, and speed). The twoaircraft on the radar screen (UAL755 and DAL80) are grouped into different categories of traffic (i.e., finaldestination Dallas and Unclassified). Different categories of traffic are represented in different colors.Aircraft are selected by clicking on them (UAL755). Moving the cursor over any token on the planningscreen or a target on the radar screen puts a rectangle around the two representations (DAL80).
4
Do participants rely to the same extent on the radarscreen when using the planning aid as when using thestrips? The answer to this question should provideevidence of the extent to which the participants in ourstudy made differential usage of the same radar func-tions in each condition. For example, they should nothave to declutter the radar as much. From observeddifferences, we can infer how participants used andaccessed information differently in the two conditions.
METHOD
Twelve en route air traffic controllers participatedin the experiment. All participants were full-perfor-mance-level controllers who served as instructors atthe FAA Academy. Full-performance-level controllersare qualified to control traffic in a sector by them-selves. The experiment was conducted at the FAACivil Aerospace Medical Institute in Oklahoma City.All participants were familiar with the airspace (i.e.,sector) used in the experiment.
Each of the 12 participants was given the task ofplanning and maintaining aircraft sequences for ageneric en route sector and to communicate the se-quence orally to a tactical controller (tactician) whoimplemented it. The tactician, another en route airtraffic controller, was the subject-matter expert. Allparticipants performed the same task under two con-ditions: 1) using the electronic planning aid withoutstrips (planner condition) and 2) using paper flightstrips without the electronic planning aid (strips con-dition). Participants used the same radar screen inboth conditions, although some of the radar function-ality was absent without the planning aid (i.e., thecolor coding of aircraft). The two scenarios were busy;up to 44 aircraft were in the sector during 20 minutesand included arrivals, departures, and overflights.The order of the scenarios and the order of the twoconditions (planner or strips) were counterbalancedacross participants.
Each participant completed a 45-minute trainingsession on the use of the planning aid and radar screen.One or two days later, each participant completed a 1-hour practice session. No training was given on theuse of the flight progress strips, as all participants hadextensive experience with them in the field. Theexperiment started a day or two following the practicesession. Immediately prior to testing, participantspracticed with the planning aid for a few minutes tore-familiarize themselves with its operation.
For each scenario, the participant was told aboutany special airspace restrictions in effect and was givensufficient time to set up either the planning aid orpaper strips while the aircraft in the scenario remainedpaused. During this time, they verbally conveyed theirplan to the tactician. The simulation was started afterthe plan was conveyed. The simulation ran for 20minutes, with a brief pause after 10 minutes.
RESULTS AND DISCUSSION
We have previously described the details of theparticipants’ superior performance and reducedworkload when using the planning aid (see Canning,Johansson, Gronlund, Dougherty, & Mills, 1999;Moertl et al., 2000). The advantage for the planningaid condition was especially noteworthy because par-ticipants had, on average, about 10 years experiencewith paper strips and less than 2 hours practice withthe planning aid. In the present study, we were inter-ested in the differential usage of the radar screenbetween the conditions and what that can tell us aboutwhy the planning aid led to superior planning perfor-mance and reduced workload.
Loglinear Modeling of User InteractionsOlson, Herbsler and Rueter (1994) presented
loglinear modeling as a technique for analyzing hu-man-computer interactions. Loglinear models allowthe test of main effects and interactions of models thatfit empirical human-computer interaction frequen-cies and thereby allow one to determine sequentialstructure and structure changes between conditions.Olson et al. (1994) used loglinear modeling on a set ofinteraction data that were semantically categorized byhuman judges. Such a procedure can prove highlyuseful but has the disadvantage of costing consider-able time resources for the manual categorizations ofactions. In contrast, we used this approach to analyzeraw data-files that were not categorized by humanjudges. Therefore, we sidestepped the objection thatthe analysis of categorized data might reflect thejudgments of human judges, in addition to actualempirical patterns in the computer interactions. Incontrast to Olson et al. (1994), our design includedrepeated measurements for which we accounted inour models.
Although participants interacted with the radarscreen more frequently in the strip condition (signifi-cant main effect of experimental condition, Wald
5
χ2(1) = 17.07, p < 0.01), of primary interest wasdetermining the source of these differences. We com-pared the observed frequencies with the expectedfrequencies and determined how participants inter-acted with the radar screen differently in the twoconditions.
Participants performed nine different types of ac-tions on the radar screen (listed in Table 1). Wecompared observed frequencies for each action withexpected frequencies assuming no differences betweenconditions (i.e., standardized residuals for each actionand condition). The standardized residuals were cal-culated as the difference between the predicted andobserved frequency divided by the square root of thepredicted frequency. Table 1 displays the results ofthis analysis; the last column shows the standardizedresiduals. The model predicted the observed frequen-cies satisfactorily (within a 95% confidence interval)for all but three of the actions. This meant that thesethree actions occurred with differing frequencies inthe two conditions. We discuss these three user ac-tions in turn.
Select token. Participants selected significantly moreaircraft on the radar screen in the strip condition thanin the planning aid condition. Selecting an aircraftcreates a border around its datablock that enhances itsvisibility. Participants in the planning aid conditiondid not need this perceptual aid as frequently, pre-sumably because they could rely on the dynamiclinkage between the two screens to perceptually locateaircraft on the radar screen.
Distance measurement. Participants measured thespatial distance between points on the radar screenmore frequently in the strip condition. Distance in-formation was crucial for planning, as it allowedestimation of when aircraft would reach specific pointsin the sector. Participants in the planning aid condi-tion did not measure the distances on the radar screenas frequently, presumably because they could rely onthe time/distance information that was presented tothem next to each aircraft on the planning aid.
Datablock adjustment. Datablock adjustments in-cluded changing and adjusting datablock position.Participants adjusted datablocks more frequently inthe strip condition than in the planning aid condi-tion. As mentioned above, controllers declutter theirradar screen to make datablock information visible.These adjustments are an important index of the usageof the radar screen. In the strip condition, participantshad to get their flight information from the radarscreen and had to declutter the radar screen to get tothis information. However, when using the planningaid, participants adjusted datablock position less fre-quently and instead relied on the planning aid toreview flight information. This was consistent withthe greater ease of information access in the planningaid condition and more time spent in the stripscondition on “housekeeping” functions.
Participants interacted with the radar screen lesswhen they worked with the planning aid. They ma-nipulated aircraft less (visually highlighted or markedaircraft less) and adjusted datablocks less often. They
Table 1.
Observed and Expected Frequencies for a Loglinear Model Assuming no Difference BetweenExperimental Conditions
User actionFrequency
in stripcondition
Frequency inplanner
condition
PredictedFrequency
Standardizedresidual for strip
conditionAdjust Vector Length 3 6 4.5 -0.71
Invalid command (error) 72 83 77.5 -0.62
Zoom in/out 10 11 10.5 -0.15
Move information table 2 2 2 0
Altitude filter 12 10 11 0.30
Invalid track (error) 3 2 2.5 0.32
Select Token 482 408 445 1.75*
Distance measurement 108 56 82 2.87*
Datablock adjustment 661 520 590.5 2.90*
Note. * Observed frequency is outside a 95% confidence interval around the expected frequency.
6
also assessed spatial distance less often. The greaterease of accessing information was the result of replac-ing radar functionality with the functionality of theplanning aid. The prior results examined interactionfrequencies. In what follows, sequences of user inter-actions were examined to provide additional insightsinto how participants used the radar screen differentlybetween the two conditions.
User Interactions Sequence AnalysisDifferences in sequences of interactions between
conditions were also testable using loglinear model-ing. For each condition, the empirical probabilities ofan interaction occurring after another interactionwere compared with the estimated probabilities, as-suming that there were no differences between condi-tions. In a loglinear model, the difference betweenconditions was tested by a three-way interaction of themain factors (antecedent action, consequent action,and experimental condition, see Olson et al., 1994).We found a significant three-way interaction (Waldχ2 (13) = 610.97, p < 0.01), which indicates different,non-random sequences between the conditions. Ac-cordingly, we found significant differences by com-paring specific action sequences between conditions.Again, we calculated the standardized residuals usingantecedent and consequent actions, and their interac-tion. Because there were 115 empirically occurringsequences, only the four sequences that were notpredicted by the model are reported in Table 2. Thesedifferences reinforced the usage differences betweenthe conditions noted above.
Participants had more frequent repetitions of dis-tance measurement and datablock adjustment in thestrip condition. They also cycled between adjustingdatablocks and selecting aircraft, and vice versa, morefrequently in the strip condition. Information retrieval
in the strip condition appeared more repetitive, com-pared with the planning aid condition. Participantsneeded to declutter the radar screen more because theyhad to integrate the strips with the aircraft on the radarscreen. Furthermore, the integration effort seemednot to occur as an isolated event but happened inbursts of repeated decluttering activity. It is hard tosee how these bursts of decluttering activity couldentail any memory-enhancing function. This manualmatching was much less prevalent in the planning con-dition, where participants used fewer interactions inabsolute numbers and fewer self-repeating interactionsto retrieve the necessary information.
CONCLUSIONS AND IMPLICATIONS
Complex cognitive tasks such as planning the se-quencing of air traffic can be supported by integratingdifferent information sources. The planning aid re-sulted in less repetitive, but more integrated, informa-tion retrieval compared with the traditional planningenvironment. Less repetitive interactions were pos-sible because the interface itself provided the physicalintegration of the information. This provided onereason why Moertl et al. (2000) found that plannerworkload was reduced. The integration of the discreteflight information with the radar information allowedparticipants to develop their plan 6.3 minutes faster inthe planning aid condition than in the strip condition.
Not only did the planning aid make planningeasier, it also improved the quality of those plans.Moertl et al. (2000) found no difference in planquality during the first half of the scenario (althoughremember, those plans were developed much faster).However, in the second half of the scenario, whenparticipants were changing and updating their plans,planning performance was superior in the planning
Table 2.
Sequences of User Actions. The Positive Standardized Residuals Indicate That User ActionSequences Occurred in the Strip Condition More Frequently Than Expected by the Model
Note. * Observed frequency is outside a 95% confidence interval around the expected frequency.
7
aid condition. The planning aid especially was benefi-cial in this situation because adapting plans to thecurrent traffic situations depended strongly on theintegration of planned sequence information with thecurrent air traffic situation. The planning aid wasdesigned to do exactly that. Participants could see onthe radar the sequence of aircraft that they had pro-posed on the planning aid. By sliding the cursor overthe corresponding aircraft on the planner, they couldsee how the plan was progressing. This visual displayof a planned sequence on the radar gave the controlleran important indication of where changes were re-quired. Aircraft that were out of sequence and did not“light-up” where they should have would focus theparticipant on relevant decision points.
The current interface has many characteristics of anecological interface (e.g., Effken, Kim, & Shaw, 1997;Lintern, 2000; Pawlak & Vicente, 1996; Rasmussen& Vicente, 1990; Vicente & Rasmussen, 1990). Eco-logical interface design argues for a perceptual formu-lation of user goals within the interface. The interfacethen facilitates actions as the user perceives his or hergoals mirrored in the affordances of the interface. Inthis way, the interface guides users’ interactions with-out major intrusions or the need for automation. Itreplaces effortful cognitive processes with parallel,perceptual processes.
Perceptual information integration proved a suc-cessful design principle when we examined the cogni-tive task of planning air traffic in isolation. Futureexperiments should be directed at integrating strate-gic planning with other controlling tasks (e.g., tacticalplanning). Only then can it be determined if theplanning aid can replace strips, or if other controllertasks still require the availability of paper flight progressstrips. Also, it is important to keep in mind our focuson strategic planning and the accompanying decisionto isolate the strategic planning tasks from the tacticalplanning tasks by assigning these responsibilities totwo different individuals. It is possible that a singlecontroller responsible for both tactical and strategicplanning would not find the planning aid useful.However, that was not the goal of our project; the goalwas to develop an interface for a possible futurestrategic controller position. A different interface mayhave resulted if our goal had been to develop aninterface to enhance the strategic planning capabili-ties of a controller working a sector alone. An impor-tant next step will be to compare the planning aid withconflict probe software to determine what aspects ofthe air traffic control task can best be accomplishedthrough information re-organization and what can bebest handled by automation.
Recent research suggests the advantage of activecontrol over passive monitoring in air traffic manage-ment (e.g., Metzger & Parasuraman, 2002). Ourfindings can be applied to the design and evaluation ofinterfaces that keep the controller in-the-loop. Spe-cifically, reliance on perceptual processes could serveas an alternative to outsourcing plan development toa piece of software. This allows the controller to dowhat humans are good at (parallel perceptual process-ing) while allowing the computer to do what it is goodat (organization and linking information). Simplemodifications to the perceptual properties of an inter-face will decrease task difficulty and increase humanperformance without impinging on higher-level cog-nition. Furthermore, the perceptual optimization ofinterfaces should be accompanied by an empiricalanalysis of behavioral differences between the newand old interface. As a result, tasks that are bestaccomplished by a human operator could be delin-eated from those more appropriately left to a com-puter. The design and evaluation of interfaces wouldbenefit from this process.
REFERENCES
Arthur, W. C., & McLaughlin, M. P. (1998). User requestevaluation tool (URET) interfacility conflict probeperformance assessment, MP 98W0000204, TheMitre Corporation, McLean, VA.
Bressolle, M. C., Benhacene, R., Boudes, N., & Parise,R. (2000). Advanced decision aids for air trafficcontrollers: Understanding different working meth-ods from a cognitive point of view. Paper presentedat the 3rd International Air Traffic ManagementSeminar, Napoli, Italy.
Canning, J. M., Johansson J., Gronlund, S. D.,Dougherty, M. R. P. & Mills, S. H. (1999).Effects of a novel interface design on strategicplanning by en route controllers. Proceedings of theInternational Symposium on Aviation Psychology,10, 528-33.
Denery, D. G., & Erzberger, H. (1995). The center-TRACON automation system: Simulation andfield testing, NASA Technical Memorandum110366, Moffett Field, CA, NASA-Ames.
Dougherty, M. R. P., Gronlund, S. D., Durso, F. T.,Canning, J. M., & Mills, S. H. (1999). Plangeneration in air traffic control. Proceedings of theInternational Symposium on Aviation Psychology,10, 522-7.
8
Effken, J. A., Kim, N. G., & Shaw, R. E. (1997).Making the constraints visible: Testing the eco-logical approach to interface design. Ergonomics,40, 1-27.
Federal Aviation Administration. (1995). Air traffic con-trol handbook (7110.65J). Washington DC: Fed-eral Aviation Administration Air Traffic Opera-tions Program.
Lintern, G. (2000). An affordance-based perspective onhuman-machine interface design. Ecological Psy-chology, 12, 65-9.
Metzger, U. & Parasuraman, R. (2001). The role of theair traffic controller in future air traffic manage-ment: an empirical study of active control versuspassive monitoring. Human Factors, 43, 519-28.
Moertl, P. M., Canning, J. M., Johansson, J., GronlundS. D., Dougherty, M. R. P., & Mills, S. H. (2000).Workload and Performance in FOPA: A StrategicPlanning Interface for Air Traffic Control. Pro-ceedings of the Human Factors and Ergonomics Soci-ety, 44, 65-8.
Olson, G. M., Herbsleb, J. D., & Rueter, H. H.(1994). Characterizing the sequential structureof interactive behaviors through statistical andgrammatical techniques. Human-Computer In-teraction, 9, 427-72.
Pawlak, W. S., & Vicente, K. (1996). Inducing effectiveoperator control through ecological interface de-sign. International Journal of Human-ComputerStudies, 44, 653-88.
Rasmussen, J., & Vicente, K. J. (1990). Ecologicalinterfaces: A technological imperative in high-techsystems. International Journal of Human-ComputerInteraction, 2, 93-110.
Vicente, K. J., & Rasmussen, J. (1990). The ecology ofhuman-machine systems II: mediating “direct per-ception” in complex work domains. EcologicalPsychology, 2, 207-49.
Vivona, R.A., Ballin, M.G., Green, S.M., Bach, R.E, &McNally, B.D. (1996). A system concept for fa-cilitating user preferences in en route airspace,NASA Technical Memorandum 4763, Ames Re-search Center, Moffett Field, CA.
