ArtiÞcial Intelligencepstone/Courses/393Rfall13/resources/week12-kurt.pdfTraffic Level (vehicles/s ) Stop Sign Traffic Light Minimum FCFS Optimal Tuesday, November 10, 2009. Vehicle-to-Vehicle

Artificial Intelligence

Tuesday, November 10, 2009






Driving Is Easy

Eating, phone calls, texting, sleeping

Drunk driving

Aggressive driving


Driving Is Hard!

Distance and velocity estimation

Physical dexterity

Piloting vs. navigating

Split-second reactions










To what extent and how can a multiagent intersection control mechanism take advantage of the capabilities of autonomous vehicles in order to

make automobile travel safer and faster?

The Question


DesiderataAutonomyLow communication complexitySensor model realismProtocol standardizationDeadlock/starvation avoidanceIncremental deployabilitySafetyEfficiency


MetricsMeasuring Efficiency

Delay: increased travel time due to intersection

Throughput: total vehicles/time/lane


Simulator

“aim3” http://code.google.com/p/aim3

~20K lines of Java

Discrete time (0.02 s)

Non-holonomic vehicle motion

Point-to-point/broadcast communication

Vehicle spawned using Poisson process


Vehicle-to-IntersectionDriver agents call ahead to reserve a region of space-time

Intersection manager approves or denies based on an intersection control policy

Vehicles may not enter the intersection without a reservation

Driver agents trust the intersection manager in the intersection


I’m arriving at time t...


Sounds good to me!


ProtocolREQ

UEST

source

_id

destin

ation_

id

vehicl

e_leng

th

vehicl

e_widt

h

maximu

m_acce

lerati

on

minimu

m_acce

lerati

on

minimu

m_velo

city

front_

wheel_

displa

cement

rear_w

heel_d

isplac

ement

max_st

eering

_angle

max_tu

rn_per

_secon

d

emerge

ncy

traver

sal_pr

oposal

s

arriva

l_time

arriva

l_velo

city

maximu

m_velo

city

CANCELsource_id

destination_idreservation_id

DONEsource_iddestination_id

AWAY

source_id

destination_idACZ

_REQUE

ST

source

_id

destin

ation_

id

start_

lane

target

_lane

vehicl

e_leng

th

ACZ_CA

NCEL

source

_id

destin

ation_

id

ticket

_numbe

r ACZ_ENTEREDsource_id

destination_id

ticket_number

ACZ_EXIT

source_i

d

destinat

ion_id

CONFIRMsource_iddestination_id

reservation_idarrival_timeearly_errorlate_error

arrival_lanedeparture_lane

arrival_velocity

acz_distanceaccelerations

REJECTsource_iddestination_id

next_communicationreason

EMERGENCY_STOPsource_iddestination_id

ACZ_CONFIRMsource_iddestination_idticket_numberstart_lanetarget_laneacz_distance

ACZ_REJECTsource_id

destination_idreason


The Important PointsProtocol

Set of messages and rules

Digitally signed

Agent implementations do not matter

Assume communication failure

Current mechanisms subsumed


The FCFS Policy

“First come, first served”

Primary policy

Grid of reservation tiles

Internal simulation of vehicles’ trajectories



Reservation Tile

“Granularity”


Time t


Time t


FCFS Video


0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0 0.5 1 1.5 2 2.5

Dela

y (s

)

Traffic Level (vehicles/s)

Stop Sign

Traffic Light Minimum

FCFS

Optimal


Vehicle-to-Vehicle

Driver agents broadcast a claim

Define relations over claims:

✦ Conflict

✦ Priority

✦ Dominance

Permissibility



Uh, I’m also arriving at time t...


Time t for me too...

No, I’m arriving at time t...


1 2 3 4

source_id1

message_id13

ixn_id5

stopped_at_ixnfalse

arrival_lane1

departure_lane2

arrival_time128479

departure_time128523

source_id2

message_id6

ixn_id5

stopped_at_ixnfalse

arrival_lane1

departure_lane2

arrival_time128479


source_id3

message_id10

ixn_id5

stopped_at_ixnfalse

arrival_lane1

departure_lane2

arrival_time128479


source_id4

message_id4

ixn_id5

stopped_at_ixnfalse

arrival_lane1

departure_lane2

arrival_time128479


Claim Claim Claim Claim


1 2 3 4

Claim Claim Claim Claim


1

32

4

“Dominance”


41 3

Permissible Nonpermissible

2






V2V Video


The BenefitsHuman Usability

Some people enjoy driving

Classic cars

Transition period

Concepts extend to cyclists, pedestrians


The FCFS-Signal PolicyAutonomous vehicles use protocolHuman-driven vehicles use signalsPolicy contains a signal modelUses state of relevant signal at arrival time:✦ Green: accept✦ Yellow: reject✦ Red: FCFSSet aside off-limits tiles during green phases


✔


✘


All-Lanes



Single-Lane



0

10

20

30

40

50

60

0 0.5 1 1.5 2 2.5

Dela

y (s

)

Traffic Level (vehicles/s)

100% Human

10% Human

5% Human

1% Human

Fully Autonomous


Failure Mode Analysis

Enable collision detection

Trigger incidents, examine aftermath

Construct crash log




10

20

30

40

50

60

70

80

90

100

0 10 20 30 40 50 60

Cars

Cra

shed

Time (s)

6 Lanes3 Lanes


Mitigating Catastrophe

Assume intersection manager can detect

Reaction:

✦ Refuse future reservations

✦ Emergency-Stop message

Oblivous vs. passive vs. active

What if vehicles do not receive?



1

1.5

2

2.5

3

3.5

0 2 4 6 8 10 12 14

Cars

Cra

shed

Time (s)

Passive20% receiving40% receiving60% receiving80% receiving

100% receiving


0

25

50

75

100

Oblivious20%

60%100% Receiving

Average Number Of Crashed Vehicles

6 Lanes 3 Lanes


0

1

2

3

4

Passive40%

80%

Average Number Of Crashed Vehicles

6 Lanes 3 Lanes


best case # of cars today: 1

worst case # of cars involved: 4.5

accidents due to driver error: ~96%

1 - (4.5 * 0.04) : 82%


What’s The Big Deal?Multiple Intersections

Protocol considerations

Downstream effects

Driver agent navigation

Upstream effects



Admission Control Zone(ACZ)

ACZ Distance



ACZ Capacity



ACZ Capacity




ACZ Capacity

✔



ACZ Capacity

✔



ACZ Capacity

Sounds good to me...



ACZ Capacity

I’m arriving at time t′...



ACZ Capacity

✘



ACZ Capacity

Sorry! Can’t do that!



ACZ Capacity

I’m away!



ACZ Capacity

I’m away!


Multi-intersection Video


Other Results

Effects of multiple intersections

Emergency vehicles

On-the-fly policy switching

Learning policy selection


Intelligent VehiclesRelated Work

Object detection and tracking✦ Stereo far-IR/fusion (Mählisch et al. 2005)✦ Gray-valued video (Gepperth et al. 2005)Lane following✦ NN for Road Departure Warning (Kohl et al. 2006)✦ “No Hands Across America” (Pomerleau 1995)✦ Robust to lighting/road conditions (Watanabe and

Nishida 2005)✦ Unmarked roads (Ramström and Christensen 2005)Adaptive cruise control (Jaguar, Honda, BMW, Nissan, Toyota)


Traffic SignalsRelated Work

TRANSYT (Robertson 1969)

SCOOT (Hunt et al. 1981)

Cooperative traffic signals (Roozemond 1999)

Q-learning (Abdulhai et al. 2003)

Learning Classifier Systems (Bull et al. 2004)

MAS + game theory (Bazzan 2005)

History-based (Balan and Luke 2006)


Autonomous Vehicles at IntersectionsRelated Work

“Potential collision points”(Rasche and Naumann 1998)

Steering algorithms/collision avoidance(Reynolds 1999)

Platoons (Clement 2002, Hallé and Chaib-draa 2005)

Physical robots (Kolodko and Vlacic 2003)


Mixed SimulationFuture Directions


Proteus RobotsFuture Directions


Exploring AsynchronicityFuture Directions


Exploring AsynchronicityFuture Directions


??

?

?

??

?

?

?

?

?

?

?

Questions?


Documents

ArtiÞcial Intelligencepstone/Courses/393Rfall13/resources/week12-kurt.pdfTraffic Level (vehicles/s ) Stop Sign Traffic Light Minimum FCFS Optimal Tuesday, November 10, 2009. Vehicle-to-Vehicle