525brims Uav Model Compa

download 525brims Uav Model Compa

of 7

Transcript of 525brims Uav Model Compa

  • 8/20/2019 525brims Uav Model Compa

    1/14

    (2003). In the Proceedings of the 12th  Conference on Behavior Representation in Modeling and Si!lation" #$%. Instit!te for Si!lation ' raining.

    Comparing Three Variants of a

    Computational Process Model of Basic Aircraft Maneuvering

     Jerry T. Ball 2

     Kevin A. Gluck 1

     Michael A. Krusmark 2

    Stuart M. Rodgers1

    1ir *orce Research +a,orator- 2+3 Co!nications

    arfighter raining Research /ivision" 030 S. ent St." Mesa" #212%01 4S

    (all eail addresses are  irst.last!"illiams.a.mil )

    e-5ords6

    Cognitive Modeling" no5ledge /ifferences" Strateg- /ifferences" 47

    ABSTRACT6  A key o#$ective o cognitive modeling research at the Air %orce Research &a#oratory's (arighter 

    Training Research )ivision is to #e a#le to e*+lore the eects o #ackground kno"ledge and task strategies on

     +erormance and learning o skills relevant to accom+lishing the Air %orce mission. ,n +ursuit o that o#$ective-

    this +a+er com+ares three variants o a com+utational +rocess model o #asic aircrat maneuvering. All threevariants are em#odied +erormance models im+lemented in the AT/R cognitive modeling architecture. The model 

    variants o+erate a 0redator A Synthetic Task 3nvironment 4ST35. 3ach model variant im+lements a dierent 

    com#ination o #ackground kno"ledge and task strategy or com+leting the #asic maneuvering task. The three

    variants o the model are called Model ariant 0 40erormance only5- Model ariant 0 4ontrol and  0erormance5 and Model ariant %0 4ontrol %ocus and 0erormance5. Model ariant 0 lacks the kno"ledge o 

    control instrument settings ty+ically availa#le to e*+ert +ilots and only considers +erormance indicators in

    com+leting the #asic maneuvering task. Model ariant 0 has kno"ledge o control instrument settings needed to

    accom+lish various +erormance o#$ectives and uses that kno"ledge as +art o a crosscheck strategy "hichincludes attending e6ually to control and +erormance indicators. Model ariant %0 also has kno"ledge o 

    control instrument settings- #ut has a dierent crosscheck strategy "hich includes ocusing on control instruments

    until they are correctly set- in addition to normal crosschecking across control and +erormance indicators. This

     +a+er documents eorts to use these model variants to e*+lore the relative eects o dierences in kno"ledge and task strategy on +ilot +erormance in A #asic maneuvering. 

    1.

    Introduction

    8e- o,9ective of cognitive odeling research at the ir 

    *orce Research +a,orator-:s arfighter raining

    Research /ivision (*R+;plore

    the effects of ,ac8gro!nd 8no5ledge and tas8 strategieson perforance and learning of s8ills relevant to

    accoplishing the ir *orce ission. C!rrentl-" the

    division:s Perforance and +earning Models (P+M)

    Research Progra is foc!sed on the !se of a S-nthetic

    as8 =nvironent (S=) 5hich incl!des a high%fidelit-

    si!lation of a Predator 4ninha,ited ir 7ehicle (47)

    a!gented 5ith ,asic ane!vering" landing and

    reconnaissance tas8s and data collection facilities. e are

    !sing the 47 S= as a test,ed for cond!cting epirical

    research and creating e,odied cognitive odels of 47

     pilot perforance and learning. 4ltiatel-" the goal is to

    !se this and other odels to develop odeling g!idelines

    for detailed and ps-chologicall- realistic representations

    of h!an ,ehavior in cople>" d-naic 5arfighting

    doains.

    his paper 5ill ,egin ,- setting the conte>t for o!r 

    cop!tational cognitive odeling research thro!gh soe ,ac8gro!nd inforation on the S=" piloting a 47" and

    the C%R cognitive odeling architect!re 5hich 5e are

    !sing. It then introd!ces the three odel variants" Model

    7ariant P (Perforance ?nl-)" Model 7ariant CP

    (Control and Perforance)" and Model 7ariant C*P

    (Control *oc!s and Perforance)" and descri,es the

    representations and processes ,!ilt into each variant of 

    the odel. he paper contin!es 5ith a coparison of the

    mailto:[email protected]:[email protected]:[email protected]

  • 8/20/2019 525brims Uav Model Compa

    2/14

     perforance of the odel variants 5ith each other and in

    the case of odel variants CP and C*P" 5ith h!an

     perforance data. he paper concl!des 5ith a disc!ssion

    of the relevance of the research for 5arfighter training.

    2. Background on the UAV STE

    he core of the S= is a realistic si!lation of the flight

    d-naics of the Predator R@%1 S-ste A 47. his

    core aerod-naics odel has ,een !sed to train ir 

    *orce Predator pilots at Indian Springs ir *ield in

     evada. B!ilt on top of the core Predator odel are three

    s-nthetic tas8s6 the Basic Maneuvering  Task- in 5hich a

     pilot !st a8e ver- precise" constant%rate changes in

    airspeed" altit!de and;or heading the  &anding Task   in

    5hich the 47 !st ,e g!ided thro!gh a standard

    approach and landing and the  Reconnaissance  Task   in

    5hich the goal is to o,tain si!lated video of a gro!ndtarget thro!gh a sall ,rea8 in clo!d cover. he design of 

    these s-nthetic tas8s is the res!lt of a !niD!e

    colla,oration ,et5een ,ehavioral scientists and e>pert

     pilots of the 47. he ai in developing the tas8s 5as to

    identif- iportant aspects of the 47 pilot:s overall tas8 

     Easpects that ta> the 8e- cognitive and ps-chootor 

    s8ills reD!ired ,- a 47 pilot. he- are tas8s that lend

    theselves to la,orator- st!d-" -et do not fall pre- to

    oversiplifications. he design philosoph- and

    ethodolog- are descri,ed in F1G. ests !sing ilitar-

    and civilian pilots sho5ed that e>perienced 47 pilots

     perfor ,etter in the S= than pilots 5ho are highl-

    e>perienced in other aircraft ,!t have no 47e>perience" indicating that the S= is realistic eno!gh to

    tap 47%specific pilot s8ill F2G. *ig!re 1 provides a vie5

    of the 47 S=. he 47 S= consists of a t5o

    onitor pilot station 5ith attached stic8 (right hand)"

    throttle (left hand) and r!dder (not sho5n).

    Basic ane!vering is the foc!s of the c!rrent odeling

    effort. he tas8 reD!ires the 47 pilot to fl- seven

    distinct ane!vers 5hile tr-ing to iniiHe root%ean%

    sD!ared deviation (RMS/) fro ideal perforance on

    altit!de" airspeed" and heading. *or each ane!ver" a trial

    starts 5ith a 10%second straight and level lead%in period

    as the pilot prepares to e>ec!te the ane!ver. t the end

    of this lead%in period" the tied trial (either 0 or &0

    seconds) ,egins and the pilot is reD!ired to ane!ver the

    aircraft at a constant rate of change 5ith regard to one or 

    ore of the three flight perforance paraeters. he

    initial three ane!vers reD!ire the pilot to change one

     paraeter 5hile holding the other t5o constant. *or 

    e>aple" in Mane!ver 1 the goal is to red!ce airspeed

    fro $ 8nots to 2 8nots at a constant rate of change"

    5hile aintaining altit!de and heading" over a 0%second

    trial. S!,seD!ent ane!vers increase in cople>it- ,-

    reD!iring the pilot to fl- trials that change in

    co,inations of t5o paraeters. Mane!ver A" for instance" is a constant%rate 1#0°  left t!rn" 5hile

    si!ltaneo!sl- increasing airspeed fro 2 to $ 8nots

    and holding altit!de constant. he final ane!ver 

    reD!ires changing all three paraeters si!ltaneo!sl-"

    decreasing altit!de fro 1300 to 1000 feet" increasing

    airspeed fro 2 to $ 8nots" and changing heading left

    2$0° over a &0%second trial.

    a,le 16 47 Basic Mane!vers

    Mane!ver irspeed ed in the vertical and

    horiHontal center. he hatched line crossing the reticle is

    the horiHon line" 5hich oves !p and do5n relative to

    the reticle to indicate changes in pitch" and also rotates

    aro!nd its center point to indicate changes in ,an8.

    t the end of a trial" the res!lts for the altit!de" airspeedand heading deviations are displa-ed graphicall-" 5ith

    act!al and desired val!es on each perforance paraeter 

     plotted across tie. @!antitative RMS/s provide

    n!erical feed,ac8 for trac8ing perforance. vie5 of 

    the feed,ac8 screen follo5ing copletion of Mane!ver 1

    is sho5n in *ig!re 3.

  • 8/20/2019 525brims Uav Model Compa

    3/14

    *ig!re 16 47 S=

    *ig!re 26

  • 8/20/2019 525brims Uav Model Compa

    4/14

    *ig!re 3 *eed,ac8 screen (right onitor)

    3. Piloting the UAV.

    he instr!entation provided 5ith ost aircraft"

    incl!ding the 47" consists of t5o different t-pes of 

    indicators6 perforance indicators and control indicators.Perforance indicators reflect the ,ehavior of the aircraft

    and incl!de airspeed" heading" altit!de and vertical speed

    indicators. Control indicators reflect the settings of the

    controls 5hich affect the ,ehavior of the aircraft and

    incl!de pitch" ,an8 and engine speed (RPM) indicators.

    d9!stents to the controls have a first order effect on

    the control indicators and a second order effect on the

     perforance indicators. hat is" a control ad9!stent 5ill

    iediatel- ,e reflected ,- the relevant control indicator.

    his control ad9!stent 5ill affect the ,ehavior of the

    aircraft" 5hich 5ill ,e reflected in the perforance

    indicators as a second order effect after soe dela-.

    ccording to the ir *orce Man!al on Instr!ent *lightFG" a 8e- to e>pert flight perforance is 8no5ledge of 

    the appropriate control settings needed to o,tain desired

    flight perforance. *or e>aple" a pitch of 3 degrees and

    an engine RPM of A300 5ill aintain straight and level

    flight of the 47 at $ 8nots over a range of altit!des

    and e>ternal conditions. he e>pert pilot need onl- set the

    appropriate pitch and engine RPM to o,tain the desired

     perforance" s!,9ect to onitoring and ad9!stent ,ased

    on varia,le flight conditions li8e 5ind and air press!re.

    +ac8ing 8no5ledge of control settings" novice pilots !st

    rel- on the perforance indicators to control the aircraft.

    tent ,an8 ad9!stents"

    5hich f!rther coplicates piloting of the aircraft.

    Besides 8no5ledge of control instr!ent settings" e>pert

     pilots are vigilant in aintaining a5areness of the stat!s

    of all indicators. his is t-picall- accoplished ,- eans

    of a crosschec8 across indicators eplo-ing either a h!,

    and spo8e pattern or a ro!nd ro,in pattern or soe

    i>t!re of the t5o. /!ring this crosschec8" it is iportant

    not to foc!s too !ch attention on perforance indicators

    and to 8eep control indicators in the crosschec8.

    4. Background on ACT-R.

    he toic Coponents of ho!ght Rational (C%R)

    cognitive odeling architect!re and developent

    environent FA"G is a po5erf!l" -et ps-chologicall-

    constrained" tool for the developent of cop!tational

    cognitive odels. C%R is ,eing !sed ,- researchers

    aro!nd the glo,e to develop and test cognitive odels

    covering a 5ide range of ,ehaviors.

    C%R 7ersion is a Coon +isp ,ased

    ipleentation of the C%R architect!re. It incl!des a

     prod!ction s-ste integrated 5ith a declarative eor-

    s-ste. he distinction ,et5een proced!ral and

    declarative eor- is a cornerstone of C%R and is

    s!pported ,- e>tensive epirical evidence. C%R is a

    h-,rid architect!re 5hich provides s-,olic prod!ctions

    and declarative eor- ch!n8s and s!,s-,olic

    echaniss for prod!ction selection and declarative

    eor- ch!n8 activation and retrieval. Prod!ction

    selection and declarative eor- ch!n8 activation and

    selection are ipleented as highl- parallel processes"

    ho5ever" once selected" prod!ction e>ec!tion is serialE 

    onl- one prod!ction can ,e e>ec!ted at a tie. hese

    s-,olic and s!,s-,olic coponents and echaniss

     provide the cognitive infrastr!ct!re needed to odel

    h!an ,ehavior at a lo5 eno!gh grain siHe to odel the

    tie co!rse of cognition at the illisecond level. his

    a8es it possi,le to odel real%tie h!an perforance

    in C%R. C%R 7ersion also provides a percept!al%

    otor s-ste for interacting 5ith the e>ternal 5orld

     

  • 8/20/2019 525brims Uav Model Compa

    5/14

    5hich a8es it possi,le to develop e,odied odels of 

    cognition. he interface ,et5een the prod!ction s-ste"

    declarative eor- and percept!al%otor s-ste is

    coordinated ,- a collection of ,!ffers. Recent research

    efforts have foc!sed on apping the C%R architect!re

    to vario!s ,rain str!ct!res and fMRI st!dies have ,een

    cond!cted to validate that apping.

    he 47 pilot odel represents real challenges and

    opport!nities for the !se of C%R in the developent of 

    cop!tational cognitive odels. he challenge of 

    integrating C%R 5ith an e>isting real%5orld si!lation

    environent has not ,een !nderta8en ,- an-

    researchers. he cople>it- of the tas8 of piloting a

    47 is significantl- greater than the cople>it- of the

    la,orator- tas8s that provide the historical roots of the

    C%R architect!re and its odeling co!nit-. he

    opport!nit- to ta8e advantage of recent enhanceents to

    C%R that s!pport e,odied cognition offers the

     prospect of ore ecologicall- valid odels" and in fact

    a8es possi,le the entire research progra of 5hich the

    5or8 descri,ed here is a coponent.

    5. The Three Model Variants

    In the interest of e>ploring the effects of differences in

    doain 8no5ledge and tas8 strategies on perforance" a

    odel of ,asic ane!vering has ,een ipleented 5ith

    three distinct variants. he three variants of the odel are

    called Model 7ariant P (Perforance onl-)" Model

    7ariant CP (Control and Perforance) and Model 7ariant

    C*P (Control *oc!s and Perforance).

    Model 7ariant P lac8s the 8no5ledge of control

    instr!ent settings t-picall- availa,le to e>pert pilots and

    as a res!lt onl- considers perforance indicators and the

     ,an8 angle indicator in copleting the ,asic ane!vering

    tas8. no5ledge of ,an8 angle settings is provided ,-

    e>plicit instr!ction to participants and is ass!ed to ,e

    availa,le to Model 7ariant P. Model 7ariant P centers its

    crosschec8 on the cloc8 and selects fro aong the

    follo5ing indicators6 airspeed" heading" altit!de" vertical

    speed and ,an8 angle. *ig!re Aa sho5s the concept!al

    design of Model 7ariant P. /!ring an attend%assess%ad9!st

    c-cle" Model 7ariant P selects an indicator to attend"

    locates the indicator on the pert pilots fro novice pilots. he first

    coparison (Model 7ariant P vs. Model 7ariant CP) is

    intended to e>plore this 8no5ledge difference.

    ltho!gh Model 7ariant CP has 8no5ledge of 

    appropriate control settings" it does not foc!s on a control

    indicator !ntil the correct val!e is achieved. Rather" ita8es a control ad9!stent and contin!es 5ith the

    noral crosschec8 5itho!t chec8ing to see if the

    ad9!stent has had the intended effect !ntil the ne>t tie

    the control indicator is attended as part of the noral

    crosschec8.

    +i8e Model 7ariant CP" Model 7ariant C*P has

    8no5ledge of appropriate control instr!ent settings.

  • 8/20/2019 525brims Uav Model Compa

    6/14

    he second coparison (Model 7ariant CP vs. Model

    7ariant C*P) is intended to e>plore the effect on

     perforance of !sing a strateg- of foc!sing on control

    instr!ent settings. o soe e>tent" foc!sing on control

    instr!ents disr!pts the noral crosschec8 across

     perforance and control indicators. /espite this

    disr!ption" foc!sing on control instr!ents a- iprove

     perforance since it allo5s the pilot to coplete the

    correct setting for a control instr!ent prior to res!ing

    a noral crosschec8.

     

    *ig!re Aa6 Model 7ariant P *ig!re A,6 Model 7ariant CP

    Model 7ariant P Crosschec8 

    (perforance J ,an8 angle onl-)

    find

    attend

    encode

    setdeviation

    select

    indicator 

    assess;

    ad9!st

    retrieve

    desired

    Model 7ariant CP Crosschec8 

    (perforance;control)

    find

    attend

    encode

    setdeviation

    select

    indicator 

    assess;

    ad9!st

    retrieve

    desired

    =sta,lish%Control

    (control)Crosschec8 

    (perforance;control)

    find

    attend

    encode

    select

    control

    indicator 

    set

    deviation

    find

    attend

    encode

    set

    deviation

    select

    indicator 

    assess;

    ad9!st

    assess;

    ad9!st

    retrieve

    desired

  • 8/20/2019 525brims Uav Model Compa

    7/14

    *ig!re Ac6 Model 7ariant C*P

     

  • 8/20/2019 525brims Uav Model Compa

    8/14

    5.1 Declarative Knowledge

    he odel variants have declarative 8no5ledge in

    addition to that descri,ed a,ove. he ost iportant of 

    these declarative 8no5ledge ch!n8s are disc!ssed ,elo5.

    he goal ch!n8 contains the 8no5ledge" or lin8s to the8no5ledge" needed to fl- the 47. It is organiHed

    hierarchicall- in ters of the top%level 8no5ledge

    relevant to fl-ing in general" 8no5ledge relevant to

    anaging control instr!ents" and 8no5ledge relevant to

    onitoring perforance indicators. coon odeling

     practice in C%R odels is to restrict the siHe of 

    declarative eor- ch!n8s to 3% slots (or pieces) of 

    inforation. In the case of the goal ch!n8 for the 47

     pilot odel" ho5ever" 5e fo!nd this to ,e !nanagea,l-

    restrictive. here is 9!st too !ch inforation a,o!t the

     pilot:s cognitive state and the aircraft:s ph-sical state that

    needs to ,e availa,le in the goal ch!n8 for decision

    a8ing.

    ?n the other hand" having all aircraft state data availa,le

    to the odel variant at all ties 5o!ld ,e too po5erf!l.

    e are not creating an a!topilot. e are creating odel

    variants intended to replicate act!al h!an perforance.

    herefore" the prod!ctions are designed in s!ch a 5a-

    that" at an- one tie" onl- a fe5 slots in the goal ch!n8 

    are act!all- !sed. *or e>aple" if the odel variant has

     9!st attended to airspeed" then the c!rrent%airspeed slot is

    availa,le to the odel variant. Slots 5ith val!es fro

     previo!s attend%assess%ad9!st c-cles are not ass!ed to

     ,e availa,le" and ne5 val!es !st ,e encoded fro the

    indicators or retrieved fro eor-. h!s" altho!gh thegoal ch!n8 has a siHea,le n!,er of slots" onl- a fe5 of 

    the have availa,le val!es at an- one tie.

    /!ring the (noral) crosschec8 attend%assess%ad9!st

    c-cle" each odel variant !st decide 5hich indicator to

    attend ne>t. ransitions fro one indicator to the ne>t are

    odeled 5ith crosschec8%intent ch!n8s that ap fro the

    c!rrent indicator to the ne>t indicator. Retrieval of a

    crosschec8%intent ch!n8 deterines the ne>t indicator to

     ,e attended ,ased on the c!rrent indicator" the ane!ver 

     ,eing perfored" and the tie%segent. llo5ing the

    ane!ver 5hich is ,eing perfored to infl!ence this

    selection res!lts in ore freD!ent attention to indicators

    5hose desired val!es are changing d!ring the co!rse of 

    the ane!ver.

  • 8/20/2019 525brims Uav Model Compa

    9/14

    specified at the start of the c-cle. Retrieval of the desired

    setting for the control instr!ent is onl- necessar- on the

    first loop thro!gh the c-cle. he desired val!e is availa,le

    in the goal ch!n8 on s!,seD!ent loops so long as the

    sae control instr!ent is in foc!s. ?ther5ise" the

    esta,lish%control attend%assess%ad9!st c-cle is patterned

    after the crosschec8 attend%assess%ad9!st c-cle.

    t the ,eginning of a ane!ver" follo5ing the lead%in

     period" all odel variants e>ec!te a series of prod!ctions

    to transition fro the straight and level lead%in period to

    the ane!ver period. hese transitions !st ,e learned

    since the- are ane!ver specific and do not reflect

    general aviator 8no5ledge. pert pilots.

    5.3 Parameter Settings

    variet- of paraeters in C%R can ,e odified to

    infl!ence the ,ehavior of a odel FA" G. Most paraeters

    have esta,lished defa!lts or coonl- !sed val!es. ?ne

    of the long%ter architect!ral goals in the C%R 

    co!nit- is to settle on defa!lt" or at least Kcoonl-

    accepted"L val!es for all paraeters" in order to f!rther 

    constrain odel ipleentations. his research has ,enefited fro that goal.

    he C%R paraeters are organiHed into s!,s-stes.

    s!,s-ste is ena,led in order to ta8e advantage of the

     paraeters it incl!des. In the odel variants descri,ed

    here" the s!,s-,olic cop!tation s!,s-ste and the

     ,ase level learning s!,s-ste are ena,led and the

     paraeters 5ithin these s!,s-stes have ,een set to

    defa!lt or coonl- accepted val!es as sho5n in ta,le 2.

    a,le 26 47 Pilot Model Paraeters

      Paraeter 7al!e

    4tilit- oise (σ  4) 1 (coon)

    oal eight (( ) 1 (defa!lt)

    +atenc- *actor ( % ) 1 (defa!lt)

    Retrieval hreshold (N) 1 (coon)/eca- Rate (d ) 0. (defa!lt)

    ctivation oise (σ  ) 0.2 (coon)

    he ,ase level learning coponent is !sed to esta,lish

    the ,ase level activation of declarative eor- ch!n8s

    corresponding to the location of control and perforance

    indicators. hese ,ase levels correspond to those

    e>pected for e>pert 47 pilots. ctivation noise is !sed

    to add stochasticit- to the selection of the crosschec8%

    intent ch!n8s that deterine 5hich indicator 5ill ,eattended to ne>t. he partial atching and other learning

    s!,s-stes of C%R and their associated paraeters are

    not c!rrentl- ,eing !sed.

    6. Verification of Knowledge and Strategy

    Differences.

    s verification that the three odel variants reall- reflect

    the 8no5ledge and tas8 strateg- differences ,eing

    considered" *ig!re copares the attention fi>ation

    freD!encies of each variant.

    *ig!re 6 ttention *i>ation Mean O for P" CP and C*P

    s *ig!re sho5s" Model 7ariants CP and C*P spend

    significantl- ore tie attending to control indicators

    and !ch less tie attending to perforance indicators"

    relative to Model 7ariant P" t (A1$) #.00" + Q .001 and

    t (A1$) 3#.3&" + Q .001 respectivel-. ltho!gh !ch less

    draatic" Model 7ariant C*P spends significantl- ore

    tie attending to control indicators than Model 7ariant

    CP" t (A1$) .A&"  + Q .001" and significantl- less tie

     

    ie

    Control

    Perforance

       M  e  a  n   O   o

       f   *   i  >  a   t   i  o  n 100O

    $O

    0O

    2O

    0O

    Model

    P

    CP

    C*P

  • 8/20/2019 525brims Uav Model Compa

    10/14

    attending to perforance instr!ents" t (A1$) A.13" + Q .

    001. hese differences in attention fi>ation freD!enc-

    reflect the 8no5ledge and tas8 strateg- coparisons of 

    interest.

    7. Comparing Knowledge Differences

    o copare the effect of 8no5ledge of control instr!ent

    settings on perforance" Model 7ariant P is copared to

    Model 7ariant CP. o eval!ate overall perforance on a

    trial" a coposite eas!re of deviation fro desired

    altit!de" airspeed" and heading 5as cop!ted. ?n each

    trial" RMS/s 5ere cop!ted for altit!de" airspeed" and

    heading. Beca!se these perforance eas!res are on

    different scales" RMS/s 5ere converted to H%scores

    5ithin class. he res!lting standardiHed RMS/s 5ere

    added together for each trial" res!lting in a standardiHed

    s! RMS/.  *ig!re presents the Mean and &O

    Confidence Interval for overall perforance of the odelvariants on 20 trials in each of the seven ,asic

    ane!vers" averaged across trials 5ithin a ane!ver.

    s *ig!re sho5s" a significant ain effect of odel

    variant on perforance is o,served" % (1"2) &A.01"  +

    Q .001" 5ith perforance of Model 7ariant CP

    significantl- ,etter than Model 7ariant P on of the $ane!vers. ?n ane!ver $" the ean perforance of 

    Model 7ariant P is slightl- ,etter" altho!gh 5ithin the

    &O Confidence Interval of Model 7ariant CP.

    *ig!re 6 Model P vs. Model CP

    no5ledge of appropriate control settings is cr!cial to

     pilot perforance on of the $ ane!vers. he

     perforance of Model 7ariant P 5as poor eno!gh on

    an- trials that the aircraft 5as at ris8 of entering an

    !nsta,le state.

  • 8/20/2019 525brims Uav Model Compa

    11/14

    *ig!re $6 Model CP vs. Model C*P

    s *ig!re $ sho5s" a significant ain effect is fo!nd ,et5een Model 7ariants CP and C*P" % (1"2) 3$." + Q .001" 5ith

     perforance of Model 7ariant C*P significantl- ,etter 

    than Model 7ariant CP on ane!vers 2" 3" A" " and $" ,!t

    not on ane!vers 1 and . hen perforing a noral

    crosschec8E5hich Model 7ariant CP is liited toEif an

    ad9!stent is ade" the effect of that ad9!stent is not

    considered !ntil the ne>t tie an indicator 5hich reflects

    the ad9!stent is attended. It is possi,le that several

    seconds a- pass ,efore that occ!rs. If a siHea,le

    ad9!stent is needed" the f!ll ad9!stent is not

    accoplished on a single attend%assess%ad9!st c-cle. (snoted a,ove" this is needed to avoid overcorrecting and

     p!tting the aircraft into oscillation.). his a8es it

    diffic!lt for Model 7ariant CP to recover fro large

    deviations and the perforance of Model 7ariant CP

    degrades significantl- on the ore diffic!lt ane!vers

    (especiall- those involving heading changes).

    9. Comparison with Human Data

    he coparison of odel variants is !sef!l to the e>tent

    that the odel variants acc!ratel- odel the aspects of 

    h!an perforance !nder consideration. Is Model

    7ariant P an acc!rate odel of novice pilot ,ehavior at

    least 5ith respect to the lac8 of 8no5ledge of control

    instr!ent settings /oes the coparison of Model7ariant P 5ith Model 7ariant CP reflect this 8no5ledge

    Is Model 7ariant CP or C*P a ,etter odel of e>pert pilot

     ,ehavior 5ith respect to the control foc!s tas8 strateg-

    coparison of the attention fi>ation freD!encies of 

    Model 7ariant C*P 5ith the e-e fi>ation freD!encies

    collected fro t5o SM=s in preliinar- research on

    ane!vers 1" 2 and 3 s!ggests that Model 7ariant C*P

    foc!ses ore attention on control indicators than is

    evident in the e-e fi>ation data fro the t5o SM=s"

    altho!gh there is considera,le varia,ilit- ,et5een the

    SM=s. Model 7ariant CP" 5hich does not foc!s on

    control instr!ents" coes closer to odeling the

    fi>ation freD!encies of these t5o SM=s. ation data fro these SM=s are -et to ,e anal-Hed. 

    In the interest of 8eeping the three odel variants

    copara,le" all three variants 5ere ipleented 5ith

    inial variation other than the variation relevant to the

    coparisons" nael-" 8no5ledge of control instr!ent

    settings and control foc!s tas8 strateg-. his eant that

    the perforance of Model 7ariant P co!ld not ,e

     

    Mane!ver 

    $A321

       M  e  a  n   S  !  

      o   f   R   M   S   /  s   (  H   )

    3

    2

    1

    0

    %1

    %2

    Model

    CP

    C*P

     

    Mane!ver 

    $A321

       M  e  a  n   S  !    o   f   R   M   S   /  s   (  H   )

    A

    2

    0

    %2

    %A

  • 8/20/2019 525brims Uav Model Compa

    12/14

    iproved ,- changing the initial control ad9!stents at

    the start of a trial to copensate for the fail!re of Model

    7ariant P to achieve straight and level flight d!ring the

    lead%in periodEaltho!gh novice pilots a- learn s!ch

    copensator- settings. Moreover" even

    novice pilots attain reasona,le perforance on the ,asicane!vers given eno!gh trials" ho5ever" there is a h!ge

    learning c!rve" 5ith perforance on earl- ane!vers

    D!ite a,-sal. iven the 5ide range of perforance

    varia,ilit- fro trial to trial and choices ade to 8eep the

    odel variants copara,le" it is diffic!lt to deterine

    5hat h!an data co!ld ,e copared to that of Model

    7ariant P. s a res!lt" Model 7ariant P 5ill not ,e directl-

    copared to h!an data.

    o validate the ,ehavior of Model 7ariants CP and C*P"

    5e copare the perforance of the odel variants

    against the perforance of the seven SM=s in the larger st!d- entioned a,ove. *ig!re # copares the

     perforance of the SM=s against Model 7ariants CP and

    C*P on all seven ane!vers. coposite eas!re of 

     perforance 5as cop!ted to copare overall

     perforance aong SM=s and Model 7ariants. *or each

    SM= on each trial" RMS/s for altit!de" airspeed" and

    heading 5ere screened for o!tliers" converted to H%scores"

    and s!ed to generate standardiHed s! RMS/s. e>t"

    to a8e odel and SM= data copara,le" thro!gh a

    siple linear transforation" odel data for each

     perforance eas!re 5ere converted to H%scores !sing

    the sae eans and standard deviations that 5ere !sed to

    cop!te H%scores for SM= data. he res!lting H%scores

    on RMS/s for altit!de" airspeed" and heading 5ere added

    together. *inall-" aggregate eans 5ere cop!ted on

    standardiHed s! RMS/s for all passed trials of each

    SM= on each ane!ver" and for all trials of each odel

    on each ane!ver. h!s" *ig!re # depicts eans and

    &O confidence intervals of standardiHed s! RMS/s

    for a saple of $ SM=s on passed trials on each

    ane!ver. /epicted for Model 7ariants CP and C*P on

    each ane!ver are eans of standardiHed s! RMS/s

    cop!ted fro the 20 trials copleted. Mean

     perforance of odel variants CP and C*P can ,e

    considered point predictions of perforance of highl-copetent 47 pilots ipleenting different control

    foc!s strategies.

    *ig!re #6 Model 7ariants CP and C*P vs. SM=s

    s *ig!re # sho5s" the ,ehavior of Model 7ariant C*P

    falls 5ithin the &O confidence interval of the SM=s on

    trials 2" 3" A" and $. Model 7ariant C*P is significantl-

     ,etter than the SM=s on ane!ver 1 and significantl-

    5orse than the SM=s on ane!ver . here appears to ,e

    a learning effect not capt!red in the odel in that Model

    7ariant C*P does ,etter than the SM=s on ane!vers 1"2and 3 and 5orse than the SM=s on ane!vers " and $.

    Model 7ariant CP onl- coes close to SM= perforance

    on ane!ver 3 and is significantl- 5orse than SM=

     perforance on all ane!vers e>cept ane!ver 1. It

    sho!ld ,e noted that the SM= data are for passed trials

    onl-E5hich f!rther   avoids pro,les 5ith o!tliers and

    iniiHes learning effectsE5hereas the odel data is

    for all trials. If 5e aggregate across ane!vers as 5ell as

    trials 5ithin ane!ver" 5e see that Model 7ariant C*P

    falls 5ell 5ithin the confidence interval of the SM=s"

    5hereas Model 7ariant CP does not (see *ig!re &).

    *ig!re &6 Perforance data aggregated across ane!vers

    ?verall" Model 7ariant C*P copares !ch ,etter to

    SM= perforance than does Model 7ariant CP. plain the o,servation in o!r preliinar- research that

    Model 7ariant C*P attends to control indicators orethan o!r t5o SM=s on ane!vers 1" 2" and 3 ?ne of 

    those SM=s alost never attended to the RPM indicator.

    pert pilots have 8no5ledge of the position

    of the ar in setting the throttle and stic8 and receive

     proprioceptive feed,ac8 fro these instr!ents. C%R 

     

    C*PCPSM=s

       M  e  a  n   S  !  

      o   f   R   M   S   /  s   (

    1.0

    0.0

    %1.0

    %2.0

  • 8/20/2019 525brims Uav Model Compa

    13/14

    does not c!rrentl- s!pport the odeling of proprioceptive

    feed,ac8. s a res!lt" Model 7ariant C*P (and CP) !st

    rel- e>cl!sivel- on the inforation provided ,- the

    control indicators to anage control instr!ent settings.

     onetheless" foc!sing on control instr!ent settingsE 

    5hether via control indicators or via proprioceptive

    feed,ac8Eis li8el- to ,e iportant to overall perforance.

    10. Relevance to Pilot Training

    he anal-sis of the perforance of the odel variants

    s!ggests that 8no5ledge of control instr!ent settings

    co,ined 5ith a control foc!s strateg- leads to the ,est

     perforance on the ,asic ane!vers. hether or not a

    control foc!s strateg- is !sed" 8no5ledge of control

    instr!ent settings is cr!cial to pilot perforance.

    itho!t this 8no5ledge" Model 7ariant P is !na,le to

    aintain sta,le aircraft ,ehavior on an- trials.

    he coparison of the three odel variants deonstrates

    the !tilit- of !sing cop!tational process odels to

    assess the effects of 8no5ledge and tas8 strateg- on

     perforance. Cop!tational process odels can ,e !sed

    to cond!ct ver- precisel- controlled st!dies of the effects

    of 8no5ledge and strateg- differences on h!an

     perforance in cople> environents. itho!t s!ch

    odels" those st!dies reD!ire h!an participants that

    !st ,e caref!ll- trained to have the desired 8no5ledge

    or !se the desired strateg-. he !se of h!an participants

    is considera,l- ore tie cons!ing and e>pensive" and

    considera,l- less 5ell controlled. he e>perienter has

    less confidence that the h!an participants are !sing the

    desired 8no5ledge;strateg-. ?nce the relative erits of 

    different 8no5ledge and tas8 strategies are assessed via

    cop!tational cognitive odels" the res!lts can ,e !sed

    to a8e changes to training progras in order to iprove

    5arfighter readiness.

    11. Acknowledgments

    e than8 the ir *orce ?ffice of Scientific Research for their s!pport. his research is f!nded ,- *?SR grant

    02=rgonoic Societ-.

    F2G Schrei,er" B. ." +-on" /. R." Martin" =. +." '

    Confer" ico State 4niversit-. t coprehension" and the

    odeling of h!an ,ehavior in cople> environents.

    #V$% &"'C# is a Research Ps-chologist at the ir 

    *orce Research +a,orator-:s arfighter raining

    Research /ivision in Mesa" . /r. l!c8 earned a Ph/

    in Cognitive Ps-cholog- fro Carnegie Mellon

    4niversit- in 1&&&.

  • 8/20/2019 525brims Uav Model Compa

    14/14

    ST' R)*&RS is a +ie!tenant Colonel in the 4nited

    States ir *orce and is the /ep!t- Coander of the ir 

    *orce Research +a,orator-:s arfighter raining

    Research /ivision (