acquisition and retention oft uavc di tif team uav coordination … · certt developer: sandia...
TRANSCRIPT
Acquisition and Retention f T UAV C di tiof Team UAV Coordination
Skill.Skill.Nancy Cooke1,2, Nia Amazeen2, Dee Andrews3, Jasmine
Duran1 Jamie Gorman1 Harry Pedersen1 Leah Rowe1, 2, 3Duran , Jamie Gorman , Harry Pedersen , Leah Rowe , , , Amanda Taylor1,
and Jennifer Winner1,2
Cognitive Engineering Research Institute1
Arizona State University2
Air Force Research Laboratory3Air Force Research Laboratory
21-23 May 2007
CERI’s Human Factors of UAVsCERI s Human Factors of UAVs Workshop
Chandler, AZ
C ll b tCollaborators
AFRLDee Andrews , Ph.D.P t Fit ld Ph D
ASUNia Amazeen, Ph.D.
Pat Fitzgerald , Ph.D.Kevin Gluck , Ph.D.Jerry Ball , Ph.D.
Research Assistants Jasmine Duran
Jamie C. Gorman, Ph.D.CERTT Developer: Sandia ResearchSteven M. Shope , Ph.D.
Jamie C. Gorman, Ph.D.Harry K. Pedersen
Leah RoweEugene SlutskiySteven M. Shope , Ph.D. g yAmanda Taylor Jennifer Winner
Overview of TalkOverview of Talk
T C di ti d UASTeam Coordination and UAS ControlT C iti & C di tiTeam Cognition & CoordinationExperiment 1: Retention of Team C di tiCoordinationExperiment 2: Training Team C di tiCoordinationConclusions
Team Coordination and UAS Control
2 Operators: 1 UAV2 Operators: 1 UAV
Team Coordination and UAS ControlControl
Team Coordination and UAS C t lControl
TEAM COGNITIONAND
COORDINATION
Team Cognition in Practice
Interdependent groups ofInterdependent groups of individuals who plan, decide,
perceive, design, solve problems, and act as an integrated unitand act as an integrated unit.
What is team cognition? How can we measure it? How can we improve it?
Our Context: Uninhabited Air Vehicle Synthetic Task Environment
(CERTT Labo ato )(CERTT Laboratory)
In our UAV STE three operators must coordinate over
headsets in order to maneuver their UAV to take pictures of groundtake pictures of ground
targets
CERTT UAV TeamsCritical Features: Heterogeneous,Critical Features: Heterogeneous,
Interdependent
AVOPilot
Common Display
Current/Next Target Controls UAV airspeed headingPilotCurrent/Next Target
Heading
Altitude
Controls UAV airspeed, heading, and altitude and monitors air vehicle systems
PLODEM
Altitude
Airspeed
PLODEM
Sensor Operator
NavigatorControls camera settings, takes photos, and monitors camera
Plans route from target to target under constraints monitors camera
systemstarget under constraints
T C itiTeam Cognition
Cognitive activity at the team level (planning, decision making, perceiving, situation assessment)Relevant to C2 performance and pother cognitive team tasksTraining and technologicalTraining and technological interventions can facilitate
Team Cognition Team Level Perception, Attention,Team Level Perception, Attention,
Thinking, and ActionA Measure individuals
+ +
Inside the Head ViewA Measure individuals and aggregateIncreasing similarity or convergence over time is associated withis associated with better performance or implicit coordinationAssumptions of homogeneity
Team cognition as collective cognition
Interactionist ViewB
g y
It is more than the sum of the cognition of individual B
Team cognition as holistic
gteam membersIt emerges (Gestalt - like) from the interplay of the individual cognition of
h t bcognition each team member or cognitive entity
Team Cognition Team Level Perception, Attention,Team Level Perception, Attention,
Thinking, and Action
Interactionist ViewBTeam cognition is in the interactions among team membersCommunication, coordination,
h d ll f i f i
Team cognition as holistic
push-and-pull of info. is team-level cognitive processingFocused on team process (vs. team member knowledge)as holistic
cognitionteam member knowledge)Team cognition (or team mind) is observable in the team’s interactions
Results That Support H li ti T C itiHolistic Team Cognition
600
Acquisition of Team Skill
Performance and process change with team-level experience and periods of nonuse 300
400
500
600
am P
erfo
rman
ce
Tm 1Tm 2Tm 3Tm 4Tm 5Tm 6Tm 7Tm 8
Knowledge is relatively stable after training Manipulations (dispersion, knowledge sharing cross
0
100
200
1 2 3 4 5 6 7 8 9 10
Mission
Tea Tm 8
Tm 9 Tm 10Tm 11
knowledge sharing, cross training) affect knowledge, but not performanceOther factors (team familiarity, 450
500
Team of internet video gamers exceed typical performance levels on UAV task
Other factors (team familiarity, video game experience) correlate with performance, but not knowledge.
200
250
300
350
400
450
perfo
rman
ce ra
te BenchmarkExp1Exp2Exp3Exp4
150
200
1 2 3 4
UAV mission
Di l iDisclaimer
We are NOT saying that there is nothing th hil i id th h d fworthwhile inside the heads of
individuals….But we ARE saying that in our studies theBut we ARE saying that in our studies the interactions among individuals provide more predictive power (are more relevant to team performance) than individual cognition
Implications of Holistic or Interactionist View
Measures & MetricsFocus on process over knowledgeProcess-based measures e.g., team SA, coordination
Theories and ModelsTheories and ModelsInspired by ecological psychology Dynamical systems modelsDynamical systems models
InterventionsProcess vs. knowledge orientedProcess vs. knowledge oriented
Team Coordination in C d d C t lCommand-and-Control
Coordination, a team-level cognitive process,Coordination, a team level cognitive process, is a type of team cognitionTeam coordination is the timely and adaptive exchange of information among teamexchange of information among team members/cognitive entitiesIn the command-and control context, team ,coordination means getting the right information to the right person and the right timet eObservations in our lab and of C2 problems in general have identified team coordinationas central to team performanceas central to team performance
Exp. 1 Method: Experimental Design117 males(92) & females(25) divided into 39 3-person (unfamiliar) Session 2 teamsTwo between subjects conditions
3-5 weeks 10-13 weeks
Retention Interval
(retention interval and familiarity) randomly assigned with scheduling constraintsParticipants randomly assigned to one of three roles
9 Teams10 Teams
Sam
e
mpo
sitio
n
three rolesSession 1: 5 40-min missionsSession 2: 3 40-min missions
10 Teams10 Teams
Mix
edCom
i C i iS C diti Mixed Condition
Session 1 Session 2AVO PLO DEMPC AVO PLO DEMPC
Same Condition
Session 1 Session 2 AVO PLO DEMPC AVO PLO DEMPC
RetentionRetention
IntervalInterval
Exp. 1 Hypotheses
H1: Teams in the Long retention-interval condition will demonstrate performance and cognitive (knowledge, process, p g ( g , p ,coordination, SA) deficits compared to teams in the Short retention-interval condition.
H2: Mixed teams will demonstrate performance and cognitive deficits compared to teams in the Same condition.
H3: Retention interval and familiarity should interact, whereby the deleterious effects of unfamiliar team members are more severe at the short retention interval compared to the long one (team member familiarity will decline with time)
Exp. 1 Results: Team PerformancePerformance
Retention IntervalRetention Interval
Exp. 1 Results: Team PerformancePerformance
Retention IntervalRetention Interval
Exp. 1 Results: No Team Performance DecrementNo Team Performance Decrement
for Short –Same TeamsNo decrement for short-same teams (H1) 1.0
1.5
ence
Mixed teams show significant decrement after -0.5
0.0
0.5
man
ce D
iffer
eor
es Short Mixed
Long Mixed
Short Same
break (H2)
No improvement for short-same teams -2.0
-1.5
-1.0
am P
erfo
rm Sco Short Same
Long Same
short same teams-2.5
6 7 8
Mission
Te
Exp. 1 Results: Team ProcessTeam Process
Mixed teams show improved process after the break.
0.4
0.5
ore
.
MIXED
Mission x Familiarity effect
S
0.2
0.3
fere
nce
Sco
Same teams showed no process change after the interval
0
0.1
Diff after the interval
Mixed teams did not show a
SAME
-0.16 7 8
Mission
decrement and in fact, improved across missions
MEASURING &MEASURING & MODELING O G
TEAM COORDINATIONTEAM COORDINATION
Target ProcedureGoal: Take picture of a ground target
Target Procedure
AVO Information (I)
Actors Activities
DEMPC AVO
Interdependencies
PLO
DEMPC Feedback (F)
Negotiation (N) AVO PLO
PLO AllDEMPC Feedback (F) PLO All
AVO
I N FAVO
AVO AVO
PLO
DEMPCDEMPC
1 2 3AVO PLOPLO
Recording Time Stamps of the dTarget Procedure
UAV task elements
I
NFF
Intrinsic Geometry (IG) y ( )Metric of Coordination
IF −NF −
F
Landmarks + Intrinsic G i l S iGeometry = Trial Series
Landmarks = targets
IG 1
targets
(TRIAL SERIES)
Mixed Teams are More Flexible/More Exploratory
0.70.80.9
1
exib
ility
ent) Flexible
SAME
MIXED
(M = .57)
0.30.40.50.6
rdin
atio
n Fl
eH
urst
Exp
one
RandomSAME
(M = .34)
(M .57)
00.10.2
Coo
r (H Rigid
Same Mixed
Coordination Flexibility was correlated with post-retention performance decrement
(r (30) = -.39, p < .03)
Mixed Teams are More Stable To Perturbation
0.04
0.05
ity nt) SAME
0.01
0.02
0.03
atio
n St
abil
ov E
xpon
en
Flo
UnstableMIXED
(M 005)
(M = .025)
0 03
-0.02
-0.01
0
Coo
rdin
a(L
yapu
no Flow
Stable(M = -.005)
-0.03
Same Mixed
Coordination Stability was correlated with post-retention numberCoordination Stability was correlated with post retention number of Team SA roadblocks overcome (rho (36) = -.36, p < .03)
Summary of Experiment 1 D iDynamics
SAME MIXED
MORE STABLELESS STABLE
E i t 1 Fi diExperiment 1 Findings
Long retention intervals and changes in team composition result in short-livedteam composition result in short lived team performance decrementsHowever these same factors (especiallyHowever these same factors (especially changes in team composition) result in more adaptive (flexible stable) teamsmore adaptive (flexible, stable) teams in terms of coordination
Experiment 2 Research Question Can we devise a training protocol that mimics the dynamics of Mixed teams tomimics the dynamics of Mixed teams to produce adaptive team coordination?
Procedural: Procedural Training – rigid, prescriptive I N Fprescriptive I N F
Cross-Trained: Shared Mental Models –cross train team memberscross train team members
Perturbation: constrain possible interactions to force new patternsinteractions to force new patterns
Experiment 2 Designp g
26 3 ( f ili )26 3-person (unfamiliar) teams completed all 9 missions (six additional teams did not) Session 1 Session 2)Three training interventions (between subjects condition) 8-10 weeks
AVO PLO DEMPC AVO PLO DEMPC
10 Procedural
Nine missions (within subjects) with an 8-10 week interval separating the 5th and 6th missions
Interval8 Knowledge
8 Perturbedthe 5 and 6 missionsTeams randomly assigned to condition with constraints and
8 Perturbed
participants randomly assigned to one of three roles
Experiment 2 Dynamics ResultsKnowledge Process Perturbed
Correlations between coordination stability (Lyap.) and team performance in t t it ti dbl kresponse to team situation awareness roadblocks
Overcome Roadblock Perturbation:
Session 1 r (24) = -.40, p < .05( ) , p
Session 2 r (24) = -.38, p < .06
Relaxation Time to Perturbation:
Session 1 r (24) = .26, p < .10
Session 2 r (24) = .41, p < .05
Experiment 2 Results:Team PerformanceTeam Performance
KnowledgeProcess
Perturbed
Experiment 2 Results:Team Performance
Perturbed advantage
Team Performance
Perturbed advantage
Knowledge
P
Process disadvantage
Process
Perturbed
CONCLUSIONSCONCLUSIONS
ConclusionsConclusions
TheoreticalProcess view of team cognitiongFocus on interactions over knowledgeEcological, Gibsonian
EmpiricalImportance of team process over knowledgeSh t t di d t l t ti d tShort term dips due to long retention and team mixing, but adaptive in long termSuggests perturbation training for adaptivitySuggests perturbation training for adaptivity
ConclusionsConclusions
MethodologicalTeam cognition metricsTeam cognition metricsSA, Coordination, CommunicationFacilitates on-line measurementFacilitates on line measurement
ModelingDynamical Systems predictionsy y pNext steps…ACT-R AVO Agent
Thank You!
Contact:Web Sites:
cerici orgContact:
Nancy Cooke
www.cerici.orgwww.certt.com
[email protected] Thanks to sponsors AFOSR andAFOSR and
AFRL
BACK-UPSBACK UPS
Cognitive Engineering Research Institute
Independent OrganizationNot-for-profit A partnership of
Academia
ASU PolytechnicOne of 4 distributed ASU
campusesAcademiaGovernmentIndustry
Mesa, AZ
campusesFormerly Williams AFBApplied PsychologyMesa, AZMesa, AZ
Sponsors
Air Force Office of Scientific ResearchAir Force Office of Scientific ResearchAir Force Research LaboratoryOffice of Naval Research
US 2004 Olympic Basketball TTeam
"We still have a couple of pdays, but I don't know where we are," replied USA head coach Larry Brown to a questionBrown to a question Wednesday on where his team was in its preparations. "We have p pgood moments and bad, but I've got a pretty good understanding of who needs to play Now the jobneeds to play. Now the job is to get an understanding of how we have to play."
US 1980 Olympic Hockey Team
Team Cognition ProblemsTeam Cognition Problems
INCIDENTSVincennes 1989Challenger & Columbia 9/11 Response 2001Friendly Fire IncidentsyResponse to Katrina 2005Remotely Operated Vehicles mishapsVehicles mishaps
T C iti P blTeam Cognition Problems
CHALLENGESAutomationComplex dynamic systemsComplex dynamic systemsFatigue/experienceFast TempoUncertain informationUncertain information
And…Distributed teamsAd hoc teamsAd hoc teamsInteragency coordinationCommunication needs
No Performance Decrement for Short-Same
575 2
350
425
500
orm
ance
0
1
2
Diff
eren
ce S
core
)
200
275
350
Per
fo Short-SameLong-SameShort-MixedLong-Mixed
-3
-2
-1
sqrt
(Per
form
ance
Short-Same
Long-Same
Short-Mixed
Long-Mixed
1 2 3 4 5Mission
6 7 8Mission
Team Situation Awareness MetricMetric
Inside the Head
How many targets have you photographed?
Measuring Team Situation Awareness
( di d f hCAST (Coordinated Awareness of the Situation by Team) Metric
P i d fi t h d P i d fi t h d
PLO AVO DEMPC PLO
Perceived first-hand:
Coordinated perception:
AVO
AVO DEMPC
Perceived first-hand:
Coordinated perception:
AVO
X X X
X X
DEMPC
AVO
DEMPC PLO
Coordinated action:
AVO
PLO
Coordinated action:
X X
AVO
DEMPC PLO
AVO
DEMPC PLO X
X
X
X
X X
Overcome roadblock? Overcome roadblock? YES NO YES NO XX
Exp. 1 Results: Team Situation Awareness
Teams responded to CAST roadblocks (1 per mission; change that requires coordinated response)Events associated with coordinated perception and action are scoredscoredOptimal degree to which team members are involved in coordinationScore based on match between observed coordination and optimal model (hits and false alarms)
Exp. 1 Results: pTeam Situation Awareness
00.05
Scor
es Short MixedLong MixedShort Same
Overall and continued decrease in false alarm rate
-0.15-0.1
-0.05
ffere
nce
S
Long Samein false alarm rate after the break (reduction in unnecessary
-0.3-0.25-0.2
6 7 8
F.A
. Dif MIXED interactions)
Mission x Composition x6 7 8
Mission
Composition x Retention Interval effect
M id fMost evident for Long-Mixed teams
COMMUNICATIONCOMMUNICATION ANALYSISS S
Team Communication Metrics(ONR Sponsor)
Communication is one form of team coordination Real time embedded in the taskReal-time, embedded in the taskObservable; Think aloud “in the wild”Rich multidimensional (amount flowRich, multidimensional (amount, flow, speech acts, content)Reflects team cognition at the holistic level;Reflects team cognition at the holistic level; for us this is team cognitionAutomated methods needed to exploit data prichness and accelerate data analysis
Automating Communication AnalysisAnalysis
Communication flow data captured by Comlog toolI t l ti d f h i t lki tInterval time record of who is talking to whom, when, and for how long (8 Hz)
Logged automatically by intercom system and Comlog software
Comlog analyzed by FAUCET methods
Communication content captured on audio tapeDiscourse on audio tape is manually transcribed
Transcripts are analyzed using Latent Semantic Analysis (LSA)
ProNet-- Procedural Networks
Specific Patterns: Conflicts
BEFORE AFTERPb
Pbeg
Pend AendAendPend
Pbeg
Pend Aend
DbegDbeg AbegAbegDbeg
DendDend
Dend
C i ti R ltCommunication Results
Distinguish high from low performing UAV teams
Co located vs distributedCo-located vs. distributedLow vs. high workload
Content coding of macrocognitive processes R2 52R2=.52.Enron email: unique chains detected surrounding Enron-critical eventssurrounding Enron critical eventsExploring patterns in chat dataUse for coordination coding
SummarySummary
Long retention intervals and mixing causeLong retention intervals and mixing cause temporary hiccups in team performance (H1, H2)However, there may be long-term benefits to these perturbations of coordination in terms ofthese perturbations of coordination in terms of adaptive, flexible teams (better process, team SA).C di ti t i i di t th t di tiCoordination metric indicates that coordination variability/stability may be most predictive of performance.Possible tradeoff between immediate performance and long-term adaptability
SSummary
Model provides a qualitative look at team coordination differences. Same teams achieve selfSame teams achieve self-organization, but not mixed teams.Implications for improving team
+p p g
coordination through training interventionsCoordination metric has generalCoordination metric has general appeal for team assessment (especially if automated through communication analysis)communication analysis)
SSummary
Perturbed training resulted in better performance in novel/high workload missions. Process - poorest performance.pInteresting relationships between flexibility/stability/adaptabilityAdaptive teams may suffer some performance loss along theAdaptive teams may suffer some performance loss along the way, but may be stronger in the endTeam training can focus on
Rigid procedural training (PROCESS) rigid teamRigid procedural training (PROCESS) rigid teamor
Presentation of deliberate and varied coordination experiences (PERTURBED) adaptive team
Coordination EventsCoordination EventsIdentified coordination events of good and bad teamsIdentified coordination events of good and bad teamsDeveloped coordination logger
Experimenter–recorded observations of information passing at each waypointpassing at each waypoint Examples:
AVO was told restrictionsPLO was told effective radiusPLO was told effective radiusAVO/PLO coordinated target airspeed
Record message present/absent, sender/ receiver repetition timingrepetition, timing10% of missions were coded by two raters; Weighted Kappa =.675
Eventually automate using communication metricsEventually automate using communication metrics