Were they in the loop during automated driving? Links between visual attention and collisions

25
Were they in the loop during automated driving? Links between visual attention and collisions 11 August 2016 ICTTP 2016, Brisbane, Australia 1 Tyron Louw @tyronlouw researchgate.net/profile/Tyron_Louw Institute For Transport Studies University Of Leeds, UK

Transcript of Were they in the loop during automated driving? Links between visual attention and collisions

Page 1: Were they in the loop during automated driving? Links between visual attention and collisions

Weretheyintheloopduringautomateddriving?Linksbetweenvisualattentionandcollisions

11 August 2016 ICTTP 2016, Brisbane, Australia1

TyronLouw@tyronlouw

researchgate.net/profile/Tyron_Louw

InstituteForTransportStudiesUniversityOfLeeds,UK

Page 2: Were they in the loop during automated driving? Links between visual attention and collisions

Acknowledgements

11 August 2016 ICTTP 2016, Brisbane, Australia2

• CO-AUTHORS• Natasha Merat

Oliver CarstenRuth Madigan

• AND COLLEAGUES AT ITS LEEDS:Gustav MarkkulaAnthony HorrobinMichael Daly

Page 3: Were they in the loop during automated driving? Links between visual attention and collisions

Introduction

Thefirstgenerationofpartially-automatedvehicles(SAELevel2)isalreadyonourroads(SAE,2014).

Driversarestillrequiredtosupervisethesystemandresumecontrol.

Butwhatwilltheyactuallydo?

11 August 2016 ICTTP 2016, Brisbane, Australia3

Page 4: Were they in the loop during automated driving? Links between visual attention and collisions

PossibleOutcome1– PassiveFatigue

Prolongedmonitoringtakesdriversoutoftheloop (Merat etal.,2014) inducingpassivefatigue (Desmond&Hancock,2001).

Reducesdrivers’attentionalcapacity(Desmond&Matthews,1997)

andabilitytodetect,evaluateandrespondtocriticalevents.

Increasingthelikelihoodofcrashes(Endsley,1995;Hollnagel & Woods,2005;deWinteretal.,2014).

11 August 2016 ICTTP 2016, Brisbane, Australia4

Page 5: Were they in the loop during automated driving? Links between visual attention and collisions

PossibleOutcome2– TaskDisengagement

Driversmaychoosetoengageinnon-drivingrelatedactivities(Carstenetal.,2012).

Distractsfromthesupposedprimarytaskofmonitoringthevehicle.

11 August 2016 ICTTP 2016, Brisbane, Australia5

Page 6: Were they in the loop during automated driving? Links between visual attention and collisions

Consequenceofoutcomes

Passivefatigueandtaskdisengagement mayhamperdrivers’abilitytosafelyresumecontrolfromanautomateddrivingsystem(Neubauer etal.,2012;Merat etal.,2012)

Suddenchangestotheroadenvironmentcapturedrivers’attention,resultinginreducedvisualscanningofthesceneandincreasedfixationstowardschanges(Chapman& Underwood,1997;Velichkovsky etal.,2002) whichlinkdirectlytoanincreaseincrashes(Crundall, Shenton& Underwood,2004).

11 August 2016 ICTTP 2016, Brisbane, Australia6

Page 7: Were they in the loop during automated driving? Links between visual attention and collisions

Experimentalsetup

Outoftheloop(OOTL)manipulationtechniques

Alteringdrivers’visibilityoftheroadsceneduringautomationwhiletheycompletedvisualandnon-visualtasks

Aim:varyingdrivers’levelofawarenessandengagementinthedrivingtask(Louw etal.,2015)

11 August 2016 ICTTP 2016, Brisbane, Australia7

NO FOG + n-back

e)

Page 8: Were they in the loop during automated driving? Links between visual attention and collisions

ManipulationAims

NoFog: ControlConditionNoFog+n-back: Assesstheeffectofacognitivetask.LightFog: Simulateaprocesswherebylimitedvisual

attentionwasdirectedtowardsthescreen.HeavyFog: Simulatesituationswherethedriverislooking

completelyawayfromtheroadandisunawareofthetrafficconditions.

HeavyFog+Quiz:Assesstheeffectofavisualtaskwithoutaphysicaldistraction.

11 August 2016 ICTTP 2016, Brisbane, Australia8

Page 9: Were they in the loop during automated driving? Links between visual attention and collisions

Researchquestions

(i) HowdotheOOTLmanipulationsaffectthelocationofdrivers’firstfixationaftertheuncertaintyevent

(ii) Howaredrivers’fixationsdistributedovertime

(iii) Whatistherelationshipbetweenfixationsduringtheuncertaintyalertsandcrashfrequency.

11 August 2016 ICTTP 2016, Brisbane, Australia9

Page 10: Were they in the loop during automated driving? Links between visual attention and collisions

Participants

11 August 2016 ICTTP 2016, Brisbane, Australia10

• 75 drivers (41 male)

• 21-69 years (M=36.16, SD=12.38)

• Normal or corrected-to-normal vision

• Average annual mileage of 8290.46 (SD=6723.08)

• Years holding driving licence (M=16.22, SD=12.92)

Page 11: Were they in the loop during automated driving? Links between visual attention and collisions

UniversityofLeedsDrivingSimulator

11 August 2016 ICTTP 2016, Brisbane, Australia11

Jaguar S-type cab in a 4m spherical projection dome

300° field-of-view projection system

v4.5 Seeing Machines faceLAB eye-tracker recorded eye movements at 60Hz

Page 12: Were they in the loop during automated driving? Links between visual attention and collisions

DriveDesign

11 August 2016 ICTTP 2016, Brisbane, Australia12

NoFog HeavyFogLightFog HeavyFog+Task

Leadvehicle

a AutomationOn

b ScreenManipulationsOn

c DroneMovesIntoLane

d ScreenManipulationsOff/UncertaintyAlert

e LeadVehicleAction

Non-critical Critical

1 2 3 4 5 6

≈150s

a b d ec

Egovehicle

NoFog+NBack

100s 3s 3s

Not a takeover request, but a request to monitor and intervene if deemed necessary

Page 13: Were they in the loop during automated driving? Links between visual attention and collisions

AreasofInterestCentre:6° circularregionoftheroadcentreareaFixations:200msthresholdwasusedwithastandarddeviationofgazepositionbelow1o

11 August 2016 ICTTP 2016, Brisbane, Australia13

Visual attention regions (Carsten et al., 2012)Carsten O, Lai FCH, Barnard Y, Jamson AH, Merat N. Control task substitution in semi-automated driving: Does it matter what aspects are automated? Hum Factors 2012;54,747–761.

Page 14: Were they in the loop during automated driving? Links between visual attention and collisions

Wheredodriversfirstlookafterthemanipulationsend?

11 August 2016 ICTTP 2016, Brisbane, Australia14

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

NoFog LightFog HeavyFog HeavyFog+Quiz

NoFog+n-back

Inde

xofDisp

ersio

n

Percen

toffixations

Centre Top Left Bottom Right IndexofDispersion

Eachbar:Percentoftotalnumberoffixations(Firstfixationof15driversfor6events=90fixations)ineachAreaofInterestforonegroup.

Indexofdispersion(id):Calculationofhowevenlyfixationsaredistributedamongtheareasofinterest

Page 15: Were they in the loop during automated driving? Links between visual attention and collisions

LookingattheRoadCenter inthefirst3s

6X3X5mixedANOVA

EventNumber(1-6)andTime(1-3)aswithin

ScreenManipulation(5Conditions)between

11 August 2016 ICTTP 2016, Brisbane, Australia15

No effect of Screen Manipulation but interaction of Time*Screen Manipulation = 1st Second differences

0%10%20%30%40%50%60%70%80%90%

100%

1s 2s 3s

PercentR

oadCenter

****

Effect of Time

** p < .001

Page 16: Were they in the loop during automated driving? Links between visual attention and collisions

LookingattheRoadCenter inthefirst3s

11 August 2016 ICTTP 2016, Brisbane, Australia16

0%10%20%30%40%50%60%70%80%90%

100%

Event1 Event2 Event3 Event4 Event5 Event6

PercentR

oadCenter

Event1 Event2

**

***

Non-critical Critical

Effect of Event Number

* p < .01, ** p < .001

Page 17: Were they in the loop during automated driving? Links between visual attention and collisions

Fixationsduringthetransition

11 August 2016 ICTTP 2016, Brisbane, Australia17

NoFog HeavyFogLightFog HeavyFog+Task

Leadvehicle

a AutomationOn

b ScreenManipulationsOn

c DroneMovesIntoLane

d ScreenManipulationsOff/UncertaintyAlert

e LeadVehicleAction

Non-critical Critical

1 2 3 4 5 6

≈150s

a b d ec

Egovehicle

NoFog+NBack

100s 3s 3s Crash

ü

💥

No Crash

VS

6X2mixedANOVATimeWindow(1s-6saftermanipulationsended)aswithinfactorCrashOutcome(collision/nocollision)betweenfactor

Page 18: Were they in the loop during automated driving? Links between visual attention and collisions

CrashOutcomes

11 August 2016 ICTTP 2016, Brisbane, Australia18

Critical Event 1 Critical Event 2 No Crash Crash No Crash Crash

No Fog (N=15) 10 5 15 0 Light Fog (N=15) 14 1 14 1

Heavy Fog (N=15) 8 7 12 3 Heavy Fog + Quiz (N=15) 13 2 15 0 No Fog + n-back (N=15) 11 4 15 0

Total 56 19 71 4

Focus on Critical Event 1

Page 19: Were they in the loop during automated driving? Links between visual attention and collisions

FixationsbasedonCrashOutcomeinCriticalEvent1

11 August 2016 ICTTP 2016, Brisbane, Australia19

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

110%

120%

1 s before 1 s 2 s 3 s 4 s 5 s 6 s

Perc

ent

Road

Cen

ter

No Collision in Critical Event 1 Collision in Critical Event 1 No Collision in Critical Event 2

**

*

SCREEN MANIPULATIONS END BRAKE LIGHT ONSET

**

* p<.05, ** p<.005

ü No Crash (N=54)

3 s after brake light

3 s after manipulations

💥 Crash (N=19)

Effect of Time Window

No effect of Crash Outcome

Interaction of Crash Outcome*Time Window

No Difference 1s before

Page 20: Were they in the loop during automated driving? Links between visual attention and collisions

Summary

Whatdriversdidduringautomationinfluencedwheretheyfirstlooked,butdifferencesdisappearedwithin2s

Drivers’visualattentionallocationstrategiesadapted totheuncertaintyalert

11 August 2016 ICTTP 2016, Brisbane, Australia20

Page 21: Were they in the loop during automated driving? Links between visual attention and collisions

Summary

Non-collidersidentifiedhazardearlyandhadconsistentscanningpattern

Colliderslatetoidentifyhazardandexperiencedperceptualtunnelling

11 August 2016 ICTTP 2016, Brisbane, Australia21

Page 22: Were they in the loop during automated driving? Links between visual attention and collisions

Conclusion(1)

Driversactivelyseekouthazardsandanticipatetheirresponse

Thisisdemandinginmanual,butmoresoinautomation.

11 August 2016 ICTTP 2016, Brisbane, Australia22

Page 23: Were they in the loop during automated driving? Links between visual attention and collisions

Conclusion(2)

Withincreasingautomation,drivers’decisionswillshiftfrombeingaboutwhen toacttowhether toact

Implication

Driversneedtopossessaconfidentunderstandingoftheirroleandthecapabilitiesoftheirhighlyautomateddrivingsystems

11 August 2016 ICTTP 2016, Brisbane, Australia23

Page 24: Were they in the loop during automated driving? Links between visual attention and collisions

Conclusion(3)

Thetypeofthehazardhasastronginfluenceonwhatisfixatedandhowquicklyitisprocessed

Implication

Highlyautomateddrivingsystemsshoulddiscriminatehazardsleadingtoautomationdisengagement,anddirecttodrivers’visualattentiontotheminsufficienttimetorespondappropriately

11 August 2016 ICTTP 2016, Brisbane, Australia24

Page 25: Were they in the loop during automated driving? Links between visual attention and collisions

[email protected]

Formoreinformationlookoutfor:

11 August 2016 ICTTP 2016, Brisbane, Australia25