:\..
1.1;!
3 ||.

H15)... 1..
53.! .
lid}:
3‘

 

z. 2 4....
iaflflafiuwunmafi
53*
a...

 

if... . . V . . V anvidfirfimffi,“

17‘.

 

0(0‘

0
a -1

This is to certify that the
dissertation entitled

IMPACTS OF COMMUNICATION DELAYS ON WIRELESS-

BASED AUTOMATED VEHICLE CONTROL SYSTEMS

presented by

YU LIU

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Civil Engineering

 

 

5W D1?»

Merss‘or’s Signature
4/ 2 7/ 2007

Date

MSU is an affirmative-action, equal-opportunity employer

.I-o-o-I-o-I-0-0-0-I-I-0-I-I-u-a-o-l— -

 

LIBRARY
Michigan State
University

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/07 p:/ClRC/Date0ue.indd-p.1

IMPACTS OF COMMUNICATION DELAYS ON WIRELESS-BASED
AUTOMATED VEHICLE CONTROL SYSTEMS

By

YuLm

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY

Department of Civil and Environmental Engineering

2007

ABSTRACT

IMPACTS OF COMMUNICATION DELAYS ON WIRELESS-BASED
AUTOMATED VEHICLE CONTROL SYSTEMS

By

Yu Liu

While existing adaptive cruise control systems and collision warning systems are
designed for comfort and safety benefits, both applications are associated with potential
problems related to their reliance on radar sensors in detecting nearby vehicles and
objects. The most concerned problem is that only neighboring vehicles can be detected,
yielding a very limited view of surrounding traffic conditions. A remedy for this
limitation is to feed individual vehicles with traffic data of other vehicles via vehicle—to—
vehicle communication. Although additional information may provide opportunities for
performance improvements, wireless communication delay must still be considered for

its potential impact on the safety of vehicle occupants.

This study primarily focused on the development of vehicle control algorithms robust to
information delays. To this end, a current assessment was conducted on the state-of—the-
art in vehicle-to-vehicle communication and driver assistance systems and the existing
research on the impacts of wireless communication delay. A microscopic traffic
simulation model was then developed to enable the simulation of Intelligent
Transportation Systems (ITS) applications using onboard vehicle sensors or wireless
communications. This simulation model has been used to assist with the development of
a vehicle control algorithm making vehicle acceleration/deceleration decisions based on

delayed information from multiple downstream vehicles.

Upon completion of the vehicle control algorithms, various evaluations were conduct to
assess its behavior under various levels of communication delays, alternative multi-
vehicle anticipative control environments, and various proportion of vehicles equipped
with the proposed control algorithms within a stream of vehicles otherwise controlled by
ordinary drivers. These evaluations were carried out using the traffic simulation model
that has been developed and looked more specifically at vehicle responses within three
car-following scenarios: single-file platoon accelerating from standstill, vehicles
responding to a slowdown by the platoon’s lead vehicle, and vehicle responding to the

sudden immobilization of the platoon’s lead vehicle.

The simulation results clearly demonstrate that information delay has impacts on vehicle
control algorithm and that this parameter should not be overlooked when vehicle safety
and stability could be affected. Two evaluation scenarios have shown significant
influence from the level of information delay considered. The simulation results further
indicate the ability of the proposed algorithms to effectively reduce the sensitivity of
vehicle control decisions to information delay. In particular, it is demonstrated that
multi-anticipative vehicle control algorithms offer significant safety benefit, particularly
in harsh situations. The presence of manually controlled vehicles within a stream of
automated vehicles showed a clear negative impact on the performance of the vehicle
control algorithm. Although manually controlled vehicles are not being fully considered
in the decision making process, such evaluation results may still help determine the
degree to which additional benefits may be obtained by any increase in the market

penetration of the proposed algorithm.

ACKNOWLEDGMENTS

First and foremost, I wish to thank Dr. Francois Dion, my academic advisor, for his
generous support and insightful guidance throughout my graduate study, and for having
been a constant source of inspiration for this research. I have learned a lot from his

approach and professional skills of conducting research projects.

I wish to extend my gratitude to other committee members -- Dr. Richard Lyles,
Dr. Ghassan Abu-Lebdeh and Dr. Subir Biswas -- for their invaluable suggestions,
comments, and feedback. I would like to thank them for the time and care they have
taken to guide this work. I would thank them for the encouragement they gave me during

this critical moment of my life.

My parents have always supported me in so many ways and I wish to thank them
infinitely, for everything. Last, but not least, I would thank my wife, Michelle (Hechuan),
who has supported my decision to fulfill my academic goals and has shared with me the

same understanding of leading a prosperous life.

iv

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................... viii
LIST OF FIGURES ......................................................................................................... x
KEY TO SYMBOLS ..................................................................................................... xii
CHAPTER 1 INTRODUCTION .................................................................................... 1
1.1. Vehicle-Based Driver Assistance Applications ..................................................... 3
1.2. ITS Applications Enabled by Vehicle—to-Vehicle (V2V) Communications ............ 7
1.3. Research Needs .................................................................................................. 13
1.4. Structure of Dissertation ..................................................................................... 14

CHAPTER 2 STATE-OF-THE-ART IN VEHICLE-TO-VEHICLE

COMMUNICATIONS AND ITS DRIVER ASSISTANCE APPLICATIONS .............. 15
2.1. Data Communication Flows in National ITS Architecture ................................... 15
2.2. Current Mobile Wireless Communications Applications and Standards .............. 17
2.3. Current DSRC Standards .................................................................................... 24

2.3.1. Elements Targeted by Standards ................................................................... 24
2.3.2. 915-MHz North America DSRC Standard .................................................... 26
2.3.3. 5.9-GHz North America DSRC Standard ..................................................... 28
2.3.4. Other Developments around the World ........................................................ 32
2.4. Existing ITS Driver Assistance Applications ...................................................... 33
2.4.1. Adaptive Cruise Control Systems ................................................................. 33
2.4.1.1. Constant Spacing Control ...................................................................... 34
2.4.1.2. Constant Time Headway Control ........................................................... 36
2.4.1.3. Variable Time Headway Control ........................................................... 37
2.4.1.4. Performance Evaluations of Current Adaptive Cruise Control Systems .38
2.4.2. Collision Warning and Collision Avoidance Systems ................................... 43
2.5. Envisioned Wireless-Based Driver Assistance Applications ................................ 50
2.5.1. Cooperative Adaptive Cruise Control ........................................................... 51
2.5.2. C00perative Collision Warning .................................................................... 54
2.6. Other Automotive Applications Enabled by Wireless Communications .............. 57
2.7. Summary of Literature Review ........................................................................... 63

CHAPTER 3 PROBLEM DEFINITION ....................................................................... 64
3.1. Envisioned Problems with Systems Implementing Multiple Applications ........... 64
3.2. Problem Specifications and Potential Solutions ................................................... 67

3.2.1. Coordinated Decision-making Algorithms .................................................... 67
3.2.2. Impacts of Wireless Communications Delay ................................................ 69
3.2.3. Coexistence of Vehicles with Different Level of Automation ....................... 70
3.2.4. Data Management Strategy .......................................................................... 72
3.2.5. Data Fusion algorithms ................................................................................ 72

CHAPTER 4 METHODOLOGY .................................................................................. 76

4.1. Research Tasks ................................................................................................... 77
4.1.1. Literature Review and Problem Definition (Task 1) ..................................... 77
4.1.2. Testbed Development (Task 2) ..................................................................... 77
4.1.3. Implementation of Communication Delay (Task 3) ...................................... 80
4.1.4. Development of Decision-making Model (Task 4) ....................................... 85
4.1.5. Evaluation of Impacts of Communication Delay and Mixed Vehicle Type
(Task 5) ................................................................................................................. 86
4.1.6. Conclusions and Dissertation Preparation (Task 6) ....................................... 87

4.2. System Evaluation .............................................................................................. 87
4.2.1. Acceleration Profile ..................................................................................... 88
4.2.2. Number of Vehicles Collided, Collision Velocity ......................................... 89
4.2.3. Time-to-Collision ......................................................................................... 90
4.2.4. Time Exposed Time-to-Collision ................................................................. 91
4.2.5. Time Integrated Time-to—Collision ............................................................... 92
4.2.6. Cumulative Threatening Frequency .............................................................. 93
4.2.7. Cumulative Jerk Frequency .......................................................................... 93
4.2.8. Acceleration Noise ....................................................................................... 94

CHAPTER 5 TESTBED DEVELOPMENT .................................................................. 95

5.1. Need for New Simulation Testbed ...................................................................... 96

5.2. Testbed Development Strategy ........................................................................... 99

5.3. Open-Source Intelligent Driver Model .............................................................. 100
5.3.1. General Description ................................................................................... 100
5.3.2. IDM Control Flow ..................................................................................... 102
5.3.3. IDM Program Structure .............................................................................. 104

5.4. CATSIM Model ................................................................................................ 107
5.4.1. CATSIM Control Flow .............................................................................. 109
5.4.2. CATSIM Program Structure ....................................................................... 110
5.4.3. Implementation of Wireless Communication Delay .................................... 114

5.5. Testbed Validation ............................................................................................ 116

5.6. Summary .......................................................................................................... 117

CHAPTER 6 LONGITUDINAL VEHICLE CONTROL ALGORITHM ..................... 119

6.1. Longitudinal Vehicle Control Functional Needs ................................................ 120

6.2. Sliding-Mode Controller for Single-Lead Scenario ........................................... 122

6.3. Sliding-Mode Controller for Multiple-Lead Scenario ........................................ 126

6.4. Multiple-Lead Vehicle Control Algorithm Refinement ..................................... 128

CHAPTER 7 IMPACT OF COMMUNICATION DELAY, MULTI-ANTICIPATIVE
VEHICLE CONTROL AND MANUALLY CONTROLLED VEHICLES .................. 131

7.1. Evaluation Scenarios and Assumptions ............................................................. 132

7.2. Impacts of Information Delay ........................................................................... 136
Scenario 1: Standstill to equilibrium .................................................................... 136
Scenario 2: Emergency braking ........................................................................... 138
Scenario 3: “Brick wall” collision ........................................................................ 141

7.3. Impacts of Multiple Lead Vehicles ................................................................... 143

vi

Scenario 1: Standstill to equilibrium .................................................................... 143

Scenario 2: Emergency braking ........................................................................... 144
Scenario 3: “Brick wall” collision ........................................................................ 146

7.4. Impacts of Manually Controlled Vehicles ......................................................... 147
Scenario 1: Standstill to equilibrium .................................................................... 148
Scenario 2: Emergency braking ........................................................................... 153
Scenario 3: “Brick wall” collision ........................................................................ 157

7.5. Summary .......................................................................................................... 162
7.5.1. Impacts of Wireless Communication Delay ................................................ 162
7.5.2. Impacts of Multiple Lead Vehicles ............................................................. 164
7.5.3. Impacts of Manually Controlled Vehicles ................................................... 165
CHAPTER 8 CONCLUSIONS ................................................................................... 168
8.1. Summary of Research ....................................................................................... 169
8.2. Research Main Conclusions .............................................................................. 171
8.3. Future Research Directions ............................................................................... 176
REFERENCES ............................................................................................................ 178

vii

LIST OF TABLES

Table 2.1 Near-Range Wireless Data Communication Standards ................................... 19
Table 2.2 Wireless Communications Alternatives. ......................................................... 22
Table 2.3 Existing DSRC Standards .............................................................................. 27
Table 2.4 Assumed Host Vehicle Maximum Braking Capability. .................................. 46
Table 2.5 Selected Field Test Results ............................................................................ 48
Table 2.6 Warning Levels of the CarTALK2000 System. .............................................. 56
Table 2.7 Envisioned Wireless-based ITS Applications. ................................................ 60
Table 7.1 Vehicle Control and Simulation Parameters. ................................................ 134
Table 7.2 Time to Reach Equilibrium from Standstill, Single Leader. .......................... 137
Table 7.3 Surrogate Safety Measures, Emergency Braking, Single Leader ................... 139
Table 7.4 Colliding Speed, “Brick Wall” Collision, Single Leader ............................... 142
Table 7.5 Surrogate Safety Measures, “Brick Wall” Collision, Single Leader. ............. 143
Table 7.6 Time to Reach Equilibrium from Standstill, Multiple Leaders. ..................... 144
Table 7.7 Surrogate Safety Measures, Emergency Braking, Multiple Leaders. ............. 145

Table 7.8 Surrogate Safety Measures, “Brick Wall” Collision, Multiple Leaders. ........ 147

Table 7.9 Time to Reach Equilibrium from Standstill, Two Leaders, 0.3 3 Delay ......... 149

Table 7.10 Vehicle Equilibrium Speed, Two Leaders Scenario, 0.3 5 Delay ................. 149
Table 7.11 Time to Reach Equilibrium from Standstill, Two Leaders, No Delay. ........ 150
Table 7.12 Vehicle Equilibrium Speed, Two Leaders, No Delay .................................. 150

Table 7.13 Safety Surrogates, Emergency Braking, Two Leaders, Manual = Gipps. 154

viii

Table 7.14 Safety Surrogates, Emergency Braking, Two Leaders, Manual = Modified
CACC. ....................................................................................................... 156

Table 7.15 Surrogate Safety Measures, “Brick Wall” Collision, Two Leaders. ............ 159

ix

LIST OF FIGURES

Figure 1.1 Elements of Intelligent Transportation Systems .............................................. 3

Figure 1.2 A Typical “String” Connection between Vehicles with Cooperative Control. 10

Figure 1.3 Another Connection Pattern between Vehicles with Cooperative Control ..... 11
Figure 2.1 National ITS Architecture ............................................................................. 16
Figure 2.2 Open Systems Interconnection Model ........................................................... 26
Figure 2.3 Data Rate vs. Range of DSRC ...................................................................... 30
Figure 2.4 5.9 GHz DSRC Band Plan with 10 MHz Channels ....................................... 31

Figure 2.5 Cumulative Distribution of Deceleration Rates under Normal Driving

Conditions .................................................................................................... 50
Figure 2.6 Data Flows in Integrated Transportation Systems ......................................... 58
Figure 3.1 Dissembled Vehicle Control Modules ........................................................... 74
Figure 4.1 Data Structure of Traditional Simulation ....................................................... 81
Figure 4.2 Interim Improvement on the Data Structure of Traditional Simulation .......... 83

Figure 4.3 Three-dimensional Data Structure for the Implementation of Communication

Delay ........................................................................................................... 84
Figure 4.4 Sample Acceleration Profiles ........................................................................ 89
Figure 5.1 Wireless-based Application Modeling with Existing Simulation Models ....... 98
Figure 5.2 IDM Model Screenshot ............................................................................... 101
Figure 5.3 IDM Model Control Flow ........................................................................... 103
Figure 5.4 IDM Model Program Structure ................................................................... 105
Figure 5.5 CATSIM Screenshot ................................................................................... 109
Figure 5.6 CATSIM Control Flow ............................................................................... 111

Figure 5.7 CATSIM Program Structure ....................................................................... 112

Figure 5.8 CATSIM Vehicle Numbering Scheme ........................................................ 115
Figure 5.9 Simulated Speed—Flow Diagram based on GIPPS Model ............................ 117
Figure 6.1 Non-filtered Vehicle Acceleration Commands ............................................ 129
Figure 6.2 Filtered Vehicle Acceleration Commands ................................................... 130
Figure 7.1 Vehicle Speed Profiles of Evaluation Scenarios .......................................... 133
Figure 7.2 Vehicle Acceleration Noise, Single Leader, with VS. without Delay ........... 140
Figure 7.3 Vehicle Trajectories in the Emergency Braking Scenario ............................ 141

Figure 7.4 Vehicle Acceleration Noise, Single VS. Multiple Leaders, with Information
Delay ......................................................................................................... 146
Figure 7.5 Time to Reach Equilibrium from Standstill with Varying Market Penetration

Levels ........................................................................................................ 152

xi

adesnm
aCCn(t)

a]:
(ICC n(t)

**
acc n“)

ammn
amax

n

an-i(t)
ddes
fig

fh,av)

Gdn(s)

(3Vn(s)

Gan(s)

KEY TO SYMBOLS

Desired acceleration of vehicle n at time t

Acceleration command of vehicle It at time t

Candidate acceleration of vehicle n at time t, if preceding vehicle did not
stOp after next time interval

Candidate acceleration of vehicle n at time t, if preceding vehicle stopped
after next time interval

Maximum deceleration (braking capability) of vehicle n
Maximum acceleration (full throttle) of vehicle n
Acceleration of the if]? vehicle ahead of vehicle n at time t

Desired distance between two moving vehicles, constant spacing model
Coefficient to incorporate vehicle acceleration in constant time headway
model

Adjustment factor for the desired time headway of vehicle n, based on its
i’h preceding vehicle at time t

Transfer function of vehicle spacing error from vehicle n-l to vehicle n
Transfer function of speed differential between neighboring pairs of
vehicles

Transfer function of acceleration differential between neighboring pairs

vehicles

xii

hdes

hdes(n_,i)

hdes, adj n

h"(t)

k1, k2, k3

[n-i

Tend
Tr
Tstart
Vn(t)
Vn-i( t)
vndes
xn( t)

xn-i(t)

Desired time headway between two moving vehicles, constant time

headway model

Desired time headway between vehicle n and the ith vehicle ahead

Adjusted desired time headway for vehicle n, constant time headway
model

Time headway to maintain between two vehicles, variable time headway
model

Calibration coefficients

Vehicle length of the ith leader in front of vehicle It

Simulation time step

Simulation end time

Driver or system reaction time,

Simulation start time

Speed of vehicle It at time 1

Speed of the ith vehicle ahead of vehicle n at time t
Desired speed of vehicle n

Position of vehicle n at time t

Position of the ith vehicle ahead of vehicle n at time t

AxdeSm—imt) Desired distance between the position of vehicle n and the ith vehicle

ahead at time t

Axb(n—)i)(t) Minimum required distance between the positions of vehicle n and the ifh

vehicle ahead at time t to allow safe braking

xiii

Ath—n-m

AXS(n—>i)

E(n-i)(t)
xi

Minimum required distance between the positions of vehicle n and the ith
vehicle ahead at time t to maintain a constant desired time headway hdes
between them

Minimum required distance between the positions of vehicle n and the ith

vehicle ahead to provide enough queuing capacity for the vehicles located

between them

Error between the positions of vehicle n and the ith vehicle ahead at time t

Sliding coefficient in sliding mode control model

Vehicle delay coefficient in first—order actuator delay model

xiv

CHAPTER 1

INTRODUCTION

The general goal of surface transportation systems is to provide people and goods with
safe, efficient and environmentally sustainable mobility. However, while decades of
infrastructure building has allowed the development of an extensive network of roads
stretching throughout the country, such construction efforts have not resolved the
problems of traffic accidents and increasing road user delay. In particular, it is estimated
that more than 6 million traffic accidents still typically occur every year in the United
States. These accidents further cause 3 millions injuries per year and cost the lives of
over 42,000 persons (NHTSA, 2005). Monetary costs of these fatalities, injuries and
property damages amount to 239 billion dollars. Costs attributed to traffic congestion in
the country’s 85 largest metropolitan areas further add another 63 billion dollars due to
loss productivity and excessive fuel consumption during idling (Schrank and Lomax,

2004).

While the traditional solution to the problem of road congestion has been to build new
infrastructures, this solution has become difficult to keep viable in recent years due to
growing constraints against road construction. Constraints against infrastructure
expansion reside in four aspects: land use constraints in urban areas, limited funding
resources to satisfy all the construction needs, growing environmental regulations, and
local geographical obstacles. Within this context, there is a growing emphasis on trying
to develop solutions in which traffic congestion problems can be solved without the need
for large infrastructure additions, i.e., through increases in the efficiency of utilization of

existing infrastructures.

The first efforts in trying to improve the efficiency of utilization of existing highway
infrastructures date back to the 19603 and 19703. This period saw the first deployments
of inductive loop detectors aimed at providing transportation officials with an automated
traffic monitoring capability. This was followed in the 19705 and 19803 with the
introduction of actuated and real-time traffic control systems enabling the development of
traffic management strategies that would be responsive to changes in traffic demands. In
the early 19903, all applications that adopt modern electronic, communication, sensing,
and control technologies to solve problems in transportation were grouped together under
the initial name of Intelligent Vehicle Highway Systems (IV HS), now known as

Intelligent Transportation Systems (ITS).

While early promoted ITS applications focused primarily on the need to develop new
capabilities for traffic management systems and the need to move towards an ultimate
automated vehicle highway system, the applications that are now being promoted cover
virtually all elements of the transportation network. Based on the capabilities provided
by current and emerging traffic surveillance and vehicle monitoring systems, applications
are being proposed to enhance traffic management systems, transit operations, trucking
operations, travel in rural areas, vehicle control systems, as well as to help better inform
the travelers, as shown in Figure 1.1. A3 is outlined below, the research presented in this
dissertation specifically focuses on the use of driver assistance systems as a tool to

improve traffic flow, alleviate traffic congestion and reduce traffic accidents.

 

nee'dTransportntion

 

h Management Systemv .

       
 

   

' Y‘- anced Traveler . ,
Information8yst .. "

  
   
 
    

l

" r. anced Public _

 

Operations (9V 0

~ “'1" -'Vehicle Control and
Safety Systems (AV ‘ '

Traffic and Vehicle Monitoring Systems
Data Communication System

s - Rural Transportation _ _.
Systems (ART -

    

 

 

1

Figure 1.1 Elements of Intelligent Transportation Systems

1.1. Vehicle-Based Driver Assistance Applications

Vehicle-based driver assistance systems refer to all automotive vehicle control functions
that may help human drivers travel safer and with more comfort. Characteristics of such
systems are usually represented by their usage of modern sensing and control technology
to mimic human perception and reaction in achieving automated driving. Depending on
the level of involvement in vehicle control, driver assistance systems can either perform
partial driving tasks or totally release human driver from the need to be behind the wheel
of a vehicle. Automated vehicle control applications that are currently offered or adopted

by vehicle manufacturers include:

a Cruise control systems;
:1 Adaptive cruise control systems; and

Collision warning systems.

Cruise control features are widely available in today’s vehicles. The feature enables a
vehicle to automatically maintain a constant speed preset by the driver. The main intent
of these systems is to release the driver from the load of longitudinal control in situations
in which the traffic is moving at constant speed. This theoretically allows automatic
speed control, which can reduce driver fatigue, or at least, the ‘lead-foot’ syndrome in
which drivers tend to accelerate as their feet get fatigued. However, traveling at constant
speeds is not a universally valid case and is only achievable when traffic is light. Most of
the time, disturbances exist in traffic flows as preceding vehicles brake or other vehicles
are merging in the traffic stream. Cruise control systems are unable to take these events
into account and adjust the speed accordingly. It is still the drivers’ duty to keep
checking traffic conditions, apply the brakes when necessary, and update the preset cruise
control speed to a suitable value. Consequently, on heavily loaded roadways, the
required driver interruptions to follow an ever-changing traffic flow usually become so

frequent that cruise control systems make the driving tasks more complicated than easier.

Adaptive cruise control supplements elementary cruise control systems with active
monitoring and braking functions. Such systems are currently offered as option on many
vehicles (Bengtsson, 2001; Lee and Peng, 2002; I-car, 2004; IEEE, 2007a). In this case,
vehicles are equipped with on-board radar sensors that provide them with the ability of
detecting slowly moving vehicles or stationary obstacles in their path that could cause a
collision if undetected, late detected, or misinterpreted by a driver. Upon the detection of
a slow moving vehicle, an adaptive cruise control system will slow down the host vehicle
until its speed matches the speed of the vehicle ahead. Compared to vehicles with

elementary cruise control, adaptive cruise control systems alleviate drivers from the need

to watch for slow-moving vehicles and constantly perform cruise speed settings
adjustments in response to the detection of slow-moving vehicles. In particular, this
alleviates the driver from the need to disengage the active cruise control module when
approaching a slower moving vehicle, activate the brakes to decelerate until the speed of
its vehicle correspond to the speed vehicle ahead, and reset the cruise control module to

the new appropriate speed.

Collision warning systems are similar in principle to adaptive cruise control systems, but
without the function of maintaining given vehicle speeds. These systems typically
operate with the objective to maintain a safe stopping distance with other vehicles ahead.
Their goal is to help avoid rear-end collisions with slow moving vehicles, stopped
vehicle, or other stationary obstacles in the path of a host vehicle. Systems under
development further attempt to expand the monitoring capabilities to vehicles traveling
on side streets. In simple collision warning systems, visual and/or audible messages are
issued to drivers upon any obstacle detection, leaving to the driver the task to take
appropriate actions in reaction to the detection. More advanced systems may also feature
automatic braking functions that would take over the vehicle’s control when the vehicle
is sensed to be at a short distance from an obstacle and no action had been taken by the
driver to avoid it. Another important distinction between adaptive cruise control and
collision avoidance systems is in the type of braking exerted. While adaptive cruise
control systems typically slow down vehicles by engine braking, collision avoidance
systems generally directly use the vehicle’s braking system to achieve much higher

deceleration rates.

Collision warning systems started to appear on commercial vehicles in North America in
1993. These systems are currently widely used as a safety option by heavy vehicle
commercial trucking companies to protect their fleets from the danger of long braking
distances (RoadRanger, 2000). Such popularity is explained by the demonstrated
benefits associated with these systems. Evaluations of collision warning systems
conducted by NHTSA in 1996 indicated a 51% reduction potential in collision rate.
More recently, data collected by one of the collision warning manufacturers indicated a
minimum accident reduction rate of 76% (RoadRanger, 2000). For passenger cars,
forward-looking collision warning systems have only been available as an option on
certain models since 2006 (USDOT, 2006). However, rear—looking collision warning
systems are increasingly being offered to help drivers with the task of backing up their
vehicle, particularly on vehicles where the rear window may not allow the driver to see

children behind the vehicle.

Although adaptive cruise control and collision warning systems tend to benefit the users
in the form of enhanced comfort and safety ratings, the two systems are not valid
solutions when addressing traffic congestion and accidents within a traffic network. The
main problem with both applications comes from their headway requirement. To ensure
that safe breaking distances are always available, both systems typically require that
significantly large headways be maintained with the vehicles ahead. On one hand, the
requirement for large headways may result in a reduction in the number of vehicles that a
roadway segment can carry. In turn, these reduced highway capacities may result in
higher congestion levels. On the other hand, the requirement for large headways may

also lead to increased safety risks, particularly in heavy traffic, by allowing aggressive

drivers to constantly cut in front of the vehicles equipped with adaptive cruise control or

collision warning systems.

1.2. ITS Applications Enabled by Vehicle-to-Vehicle (V2V) Communications

The long headway requirements associated with proposed vehicle assistance systems are
imposed by the technical limitations of the radar sensors currently installed on vehicles.
However, the provision of more sensitive radars may not be a cost-effective solution as
the higher cost of such radars may increase vehicle prices beyond a point that may
negatively affect driver attraction to such systems. Faced with the need to keep
improving vehicle safety at reasonable cost, automobile manufacturers have been looking
into the use of new technologies to provide drivers and vehicle control systems with
information surrounding their environment. Within this objective, wireless
communications are increasingly being viewed as a potential means to provide pertinent
information to a moving vehicle at reasonably low cost. Based on the available vehicle-
based communications platform, ITS applications relying on wireless communications
are for instance currently being developed to perform a series of desired functions such as
cooperative driving, cooperative hazard warning, and. large-scale traffic optimization

(e. g., shockwave reduction).

Two types of wireless communications are currently being proposed to help enable new
vehicle control applications. These two types of communication are distinguished based
on the terminals for wireless signals. The first type comprises communications between
vehicle and roadside equipment, while the second type comprises communications

between different vehicles. Vehicle-to-roadside communications allow wireless

communication devices mounted on the side of the road to upload data from passing
vehicles or download data to these vehicles. The messages uploaded from vehicles can
for instance be used to estimate travel time between known points, thereby converting
each passing vehicle into a probe vehicle. At the opposite, information reaching vehicles
may provide travelers with local maps and business directories, inform travelers of
construction zones and congested traffic conditions ahead, and propose alternative routes.
In contrast, vehicle-to-vehicle communications provide direct information channels
between neighboring vehicles. As such, this type of communication allows direct data

sharing between moving vehicles.

One of the most influencing factors in judging one communication type over the other is
their respective implementation cost. While applications based on vehicle-to-vehicle
communications can operate wherever equipped vehicles go, applications supported by
vehicle-to-roadside wireless connections have to be supported by an appropriate roadside
infrastructure communication network similar to the communication network supporting
cellular phones. Significant cost difference will then exists between the two types of
wireless—based applications, particularly when considering applications in rural area,
where communication demand from light traffic is low and it becomes challenging for
vehicle-to-roadside communications to provide comparable coverage as vehicle-to-
vehicle communications can easily do. Because of the lower requirements in
infrastructure deployment, vehicle-to-vehicle based applications are currently seen as

being more versatile and cost-efficient than applications requiring roadside equipment.

When compared to radar-based adaptive cruise control systems, the benefits of wireless
vehicle-to-vehicle communications can generally be found in two aspects. Firstly, the

8

additional communication capability can extend the monitoring scope beyond what can
be offered by on-board radar sensors. In particular, this creates a chance to improve
traffic safety through improved reactions to downstream obstacles that are currently out
of radar detection reach. The expanded detection range provided by wireless
communications not only allows to see beyond immediately neighboring vehicles, but
also irnproves sensing abilities at locations where radar sensors are likely to give
erroneous results or have difficulty operating. A typical example of such a location is
sharp curves, as radar signals cannot bend along curves. In contrast, wireless
communication may allow the collection of information from vehicles in or beyond a
curve. This would effectively enable drivers unaware of downstream hazards because of
roadside obstructions to receive warning messages sent by vehicles who have just past
the hazard or are within sensing range of it. This extended monitoring range would in
this case translate into an increase potential for reducing the number and severity of
accidents. Secondly, the additional sensing capabilities provided by wireless
communication could improve roadway capacity by allowing a compression of
headways. Besides these major benefits, vehicle-to-vehicle communications may also
help improve stability of both individual vehicle and traffic flow, improve the efficiency

of infrastructure utilization, and reduce air pollution.

Examples of foreseen applications based on vehicle-to-vehicle communications currently

include, among others:

Cooperative adaptive cruise control.
Cooperative collision warning systems.

Emergency vehicle preemption.

Vehicles equipped with cooperative adaptive cruise control functions have the ability to
communicate with each other when they are at relatively close distance (typically up to a
few hundred meters apart). A common connection pattern for multiple vehicles is a
string, which links vehicles one by one according to their spatial order, as shown in

Figure 1.2.

Ad-hoc . Neiwork ‘

,Gd ‘‘‘‘‘‘‘ ad ,. ad , “““““““ *

Veh n
x \ ~ _______ ,. I \\\ _______ /< 3:“; ., _______ if”.
—-—+ Radar detection — - — -> Wireless info of Veh n—3
- ----- ‘> Wireless info of Veh n—2 m ~-~-------> Wireless info of Veh n—l

Figure 1.2 A Typical “String” Connection between Vehicles with Cooperative

Control

Besides connecting vehicles next to each other, wireless communication also allows
direct data transfer arcing over a certain number of vehicles, as shown in Figure 1.3.
Such pattern without mandatory information hop on all vehicles along the path of
communication has the potential to reduce communication delay by removing
unnecessary data packet hops. In this case, the range at which communication links can

be established between two vehicles will depend on the communication technology used.

10

 

——> Radar detection - — - -> Wireless info of Veh n-3
- ----- 1» Wireless info of Veh n-2 ------------- D Wireless info of Veh n-l

Figure 1.3 Another Connection Pattern between Vehicles with Cooperative Control

Among the interconnected vehicles, pertinent information from each vehicle, such as its
speed, acceleration rate and warning messages, can be transmitted to its immediate
neighbors via wireless communication. Information traveling along a string would thus
hop from one vehicle to another. As a result of this hopping pattern, each vehicle within
a string would then have access to the status of downstream or upstream vehicles that are
currently outside its field-of-view. With the information received from surrounding
vehicles, the headways maintained by individual vehicles under cooperative driving
could thus theoretically be reduced to a certain level without sacrificing safety, thus
resulting in a potential increase in roadway capacity. Accidents risks could further be
reduced through the ability to obtain advanced warning about sudden changes in traffic
conditions ahead. At the same time, higher levels of driving comfort and traffic stability

may be achieved through less frequent abrupt decelerations.

Similar to the improvements from adaptive cruise control to cooperative adaptive cruise
control, cooperative collision warning and avoidance systems can be viewed as an
enhanced version of current collision warning and avoidance systems. With vehicle-to-

vehicle communication, the major benefits of cooperative collision warning and

11

avoidance systems comprise their ability to detect roadway hazards before they come into
a driver’s field of vision, to redirect the attention of distracted drivers to their driving
task, and to finally reduce driver reaction time. Their usefulness is particularly supported
by a study conducted by the National Transportation Safety Board (NTSB) that has
indicated that 60 percent of rear-end collisions happening in the United States every year
could be reduced if driver reaction times were to be reduced by 0.5 second (NTSB,

2001).

A third envisioned ITS application based on vehicle-to-vehicle communications is
emergency vehicle preemption. This is a novel application. The goal of emergency
vehicle preemption systems is to amend emergency vehicles with communication
capability, so that they can send warning messages directly to the drivers of other
vehicles that are in their path. With respect to visual and audio warnings disseminated by
flashlights and sirens, the superiority of this communication—based warning first lies in
the ability to target the warning messages only to the vehicles that are in the path of
emergency vehicles. Alerting drivers with on-bard warning messages also removes the
confusion that often arises in situations in which a driver cannot determine from which
direction a siren is emitting. This translates into reduced accident risks, as well as lower
delays for vehicles not in the path of an emergency vehicle that may have otherwise
slowed down until they could determine where the vehicle is. With vehicle-to-vehicle
communications and emergency vehicle preemption systems, there is therefore a
potential for emergency vehicles to arrive at incidents faster and to minimize their

impacts on normal traffic.

12

1.3. Research Needs

While the ideas of wireless-based ITS applications provide opportunities for solving
some of current traffic problems without encountering infrastructure expansion
difficulties, it is still necessary to realize that most vehicles currently running on streets
are not typically equipped with functions more advanced than an elementary cruise
control. Before the envisioned ITS applications can be brought into use in every vehicle,
a large gap between application concepts and application reality needs to be bridged. The
most distinguishing characteristics of future driver assistance systems with respect to the
systems that are currently being tested and researched is the number of applications that
can be expected to be used by a given vehicle. While most of the current research
focuses on single applications, at one point all these applications will need to be installed
on a vehicle, thus resulting in a situation in which multiple ITS functions will likely be
running simultaneously. In particular, the coexistence of ITS applications collaborating
with each other is often seen as a means to achieve all the desired functionalities of
vehicle-based ITS systems. However, simultaneous and collaborative applications mean
that individual application will have to compete for limited data communication and
processing resources. This leads to potential operational problems. As will be detailed
later, solutions to these problems will be explored by the development of decision-
making algorithms attempting to consider the delays potentially incurred with data
communications. The research outlined in this dissertation focuses more specifically on

the following issues:

a Impacts of wireless communications delays on ITS vehicle control applications

Effective vehicle control in the presence of information delays

l3

D

Management and fusion of data with varying latency

Coexistence of vehicles with multiple levels of automation

1.4. Structure of Dissertation

The remaining of this dissertation research report is arranged as following.

Chapter 2 provides a summary review of the professional literatures on issues
related to the state-of-the-art in wireless communication for ITS applications and
automated vehicle control techniques.

Chapter 3 defines problems that were researched in the dissertation and explains
their significance with regards to vehicle-based ITS applications.

Chapter 4 describes the tactical plan that was deve10ped to guide each step of the
research to a successful conclusion.

Chapter 5 introduces a traffic simulation testbed that considers wireless
communication delay.

Chapter 6 develops a vehicle control logic that integrates traffic data from
multiple lead vehicles for decision—making.

Chapter 7 demonstrates the impact of wireless communication delay, multi-
anticipative vehicle control logic and the presence of manually controlled vehicles
on the performance of longitudinal vehicle controller.

Chapter 8 finally summarizes the findings of this research and gives

recommendation for further research directions.

14

CHAPTER 2
STATE-OF-THE-ART IN VEHICLE-TO-VEHICLE COMMUNICATIONS AND

ITS DRIVER ASSISTANCE APPLICATIONS

This chapter presents a review of the state-of-the—art in wireless communication standards
and vehicle-based driver assistance systems, as well as a portrait of current efforts to
develop enhanced driver assistance systems based on wireless communications. Starting
from a review of the envisioned data flows between the various physical entities of the
proposed National ITS architecture, the chapter presents the various mobile wireless
communication standards for vehicles that have been developed to date and that can be
used to support ITS applications. In particular, a detailed explanation of the proposed
Dedicated Short-Range Communications (DSRC) standards is provided, as these
standards are currently envisioned to play a central role in the development of wireless-
based ITS applications. This description is followed by a presentation of the various
types of vehicle-based driver assistance systems that are currently available. This is then
followed by a discussion of how wireless communication capabilities can be used to
expand the functionalities provided by the existing driver assistance systems and to

develop novel applications.

2.1. Data Communication Flows in National ITS Architecture

With the objective to provide a coordinated framework for the planning, design and
deployment of ITS applications, the US. Department of Transportation has developed the

general architecture shown in Figure 2.1 based on inputs from the public, private and

15

academic sectors (USDOT, 2003). Within this architecture, four types of communication

capabilities are seen as playing a vital role in the development of new ITS applications:

 

' Emissions . Archived nan
Service Freight
Provider M11!!!- » _ Menu. J

 

 

 

 

 

 

 

Toll
Collection

 

Figure 2.1 National ITS Architecture

Source: (USDOT, 2003).

a Wireline communications between fixed points, typically via telephone or
fiber-optic lines, to allow data exchanges between transportation centers, roadside
equipment, traveler information kiosks and personal computers at fixed locations,
as well as exchanges between fixed systems and wide area wireless broadcast
systems.

a Wide area communications, either through broadcast or interactive two—way
communication, to enable wireless data exchanges between infrastructure—based

equipment and mobile receivers installed in vehicles or carried by persons.

16

Another link not shown in Figure 2.1 is the increasing possibility to directly
exchange data between transportation centers, roadside equipment, travelers and
vehicles through wide area wireless communications.

Dedicated short-range communications (DSRC), to allow close-proximity
wireless data exchanges between roadside equipment at fixed location and
emitters/receptors in moving vehicles.

e- VehicIe-to-vehicle communications, to allow information exchanges between

moving vehicles.

Fixed points wireline and wide area wireless communications are currently standardized
by National Transportation Communications ITS Protocols (NT CIP). These standards
provide the communication rules and vocabulary necessary to allow traffic control
equipment from different manufacturers to operate with each other as a system and
reduce the reliance on specific equipment and software vendors (NT CIP, 2004).
However, NTCIP standards are confined to center—to-center and center-to-field
communications (Pline, 1999). Communications between roadside devices and vehicles
passing by, as well as between vehicles within a certain range, are not currently supported
by NTCIP but by various new emerging standards. These new standards are introduced

in the next two sections.

2.2. Current Mobile Wireless Communications Applications and Standards

The automotive and transportation industry has researched for many years the use of
wireless technologies to enhance services to motorists. This interest is emphasized by the

fact that wireless communications are currently seen as providing the most effective

17

means to send and collect information from moving vehicles. Below are the two main

wireless-based applications that are currently being used in a vehicle:

a Wireless phones. Wireless phones introduced the possibility for drivers to
receive and send information to any person or system connected by phone while
driving. In essence, this technology eliminated the isolation of motorists while
traveling. A prominent example of a system initially based on the use of wireless
phone communication is the OnStar system.

a Vehicle-roadside wireless services. One of the earliest applications of vehicle-
roadside wireless communications is the introduction of electronic toll collection
in the late 19803 to allow vehicles equipped with an onboard transponder to pay
tolls without having to stop. Nowadays, radio—frequency, microwave, infrared
and optical systems are being used for a range of applications requiring the

identification of passing vehicles.

Significant efforts are currently being directed toward the development of information
systems allowing data exchanges with moving vehicles. In support of this need, various
standards for wireless communications between moving objects have been developed.
As indicated in Table 2.1, five Wireless Local Area Network (WLAN) standards now
exist. These are the 802.15 (Bluetooth), 802.154 (ZigBee) and three sub-standards of the
general Wireless Fidelity (WiFi) family, the 802.11a, 802.11b and 802.11g standards. In
parallel to the WLAN standard development, two North-America DSRC standards were
further developed for transportation applications. The first standard, the 915-MHz
DSRC, was developed from industry proprietary standards, while the latter is based on
the 802.11a standard.

18

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Gama
33 oz K 009 mmmd I mw.m 33 sic e9
33% scram
oweafifioam
@8832—
.E: 02 no 8 v 23 a3 hymns
_ _
fl: 02 an 832 am 88 Wanna“
56 I new 953V
0 I
33 m w em com Om mm.m I 26 33 «SSE
A v A7255
, - . I . ES, 3.8.592
awe: oz : 8N o2 84 N 9. N 32 £38 8.2 =83
mac—oh?
as w.
33 oz mmd Om an 33 “immumwh
8 o:
33 oz _ 9 Sam I ova 33 Afimfiwacw—e
Gems: E: 3::
gage—5 682525 33— San «weed wean c8932 Ewen—Sm :e:3=&<
8.5.5.:35 banana Eewmm 8:55.32 335:0 55:3; 53>

 

 

mega—=35 :emaaofiafifieu San moo—oh? «yams—«oz fin ~35.

l9

In terms of communication features, the Bluetooth standard offers a maximum data rate
of 1 million bits per second (Mbps) over a communication range of 10 m. The ZigBee
standard provides a slightly longer range (30 m) but with a significantly lower data
transfer rate (0.25 Mbps). It also allows significantly more communication nodes within
a network (216 nodes compared to 8 for Bluetooth), making it better suited for
applications requiring communications between large numbers of vehicles in
geographically restricted locations, such as parking lots. Since both standards feature
relatively short communication range, they also provide relatively low interference risks

from other applications.

The 802.11b standard extends the communication range to 200 m and the data exchange
rate to 11 Mbps. Because of the use of the unregulated 2.4-GHz frequency range,
communications with this standard are subject to interferences from microwave ovens,
cordless phones and other appliances using the same frequency. By operating in the
more regulated 5-GHz range, communications based on the 802.11a standard are much
less susceptible to interferences. Interferences in this range only include military radar
and satellite communication systems. While data rates of up to 54 Mbps are possible, the
power required for higher frequency reduces the communication range to 200 m. The
more recent 802.11g standard finally attempts to combine the benefits of both 802.11a
and 802.11b standards by supporting data rates up to 54 Mbps with ranges up to 300 m.
However, applications based on this standard are still subject to interferences associated

with the 2.4-GHz frequency.

The desire to provide broadband wireless connectivity across metropolitan areas has
further resulted in the development of the Worldwide Interoperability for Microwave

20

Access (WiMAX/802.16) standards. Within this family, the 802.16 standards address
frequencies from 10 to 66 GHz, while the 802.16a standard addresses the 2 to 11 GHz
range. WiMAX standards provide up to 50 km of service area without direct line-of-
sight requirement, with typical cell radius of 6-10 km. Data transmission rates of up to
70 Mbps are further allowed on a single channel that can support thousand of users,

allowing the standard to adapt to available spectrum and channel widths or to be licensed

to different service providers (IEEE, 2002).

Recent development of communication capabilities between roadside equipment and
vehicles have more specifically focused on DSRC standards. Initial work conducted for
the development of electronic toll collection systems and commercial vehicle clearance
systems focused on the 915-MHz frequency. While this was sufficient for the intended
applications, it did not support the high data rates or communication ranges demanded by
emerging ITS applications. In response to this need, attention has more recently shifted
on a new standard in the 5.9-GHz range specifically designed for transportation
applications and which offers communication rates of up to 27 Mbps, communication
range of up to 1000 m, and a much lower potential for interferences. This standard is
referenced as the Wireless Access for the Vehicular Environment (WAVE) or 802.11p

(IEEE, 2007b).

Table 2.2 lists seven categories of applications that are currently being considered and the
various communication standards that enable these applications (Robinson, 2004). For
many applications, particularly those aimed at improving transportation safety or

providing automated vehicle control, an important consideration in the selection of an

21

Table 2.2 Wireless Communications Alternatives
Source: (Robinson, 2004).

 

 

Application Category DSRC WiF i WiMAX
Vehicle-to-vehicle Communications Based B‘ _ _
Applications
Roadside-to-vehicle Active Safety B - -
Highway Information A” A -
Electronic Payments (secure payment transactions) A A A
Active Highway Information (location—based

. A A A
warnings / controls)
Driver Information (Situational information,

. . A A A

entertainment, data servrces)
Fleet Management (location-based / tracking A A A

services)

 

' The Best choice.

" Different Alternatives.

appropriate communication standard is the ability to avoid large communication delays
and frequent communication failures. Other elements such as interoperability with other
systems, data communication security, interference potential, and deployment costs may
also be considered. For vehicle—to-vehicle communications and safety applications using
vehicle-roadside communications, DSRC is currently viewed as the best choice due to its
long communication range (1000 m), its 27 MBps data transfer rate, and low likelihood
of interferences. For applications requiring shorter communication ranges or not
affecting the safety of a vehicle, such as traveler information or fleet management
applications, either DSRC or WiFi standards are suitable. Long-range communication
through WiMAX standards are also suitable for applications aimed at providing
information to drivers or supporting fleet management operations. WiMAX standards are

however not deemed suitable for applications based on vehicle-to-vehicle or safety

22

applications using vehicle-roadside communications as the large communication cell size
associated with this type of wireless communication creates a potential for increased

communication delays and communication breaks due to data packet collisions.

Deemed as the best means of wireless communication for vehicle-based applications,
DSRC also helps keep implementation cost affordable for vehicle-to—vehicle
communication based applications, including driver assistance functions within the scope
of this research. The basis of low-cost implementation is the independence of DSRC
vehicle-to-vehicle applications on roadside infrastructure. Communications between
vehicles over DSRC is achieved by allowing the establishment of wireless ad—hoc
networks, also sometimes referred to as mesh networks. These are randomly occurring
communication networks that activate when two or more entities need to pass
information between themselves (Leader, 2004). Such networks exist as long as a need
for communication exists. They are typically designed to establish communication links
without having to rely on fixed architecture, though fixed transmission and receiving
points may be used. Each vehicle is viewed as a node that can be used to route a piece of
information between a given source and a given destination. Information routing paths
will thus be developed based on the number and locations of vehicles present in an area
when the communication is initiated. The topology of the communication network will
consequently constantly change as vehicles leaving the communication area void the
established paths and new arrivals create new paths. This capability to dynamically
adjust data routing paths thus make ad-hoc networks particularly well adapted for
applications requiring communications with moving vehicles. Therefore, the presence of

at least two vehicles equipped with communication capabilities at a given location would

23

be sufficient to establish a communication link, and the implementation of vehicle-to-

vehicle applications would not require the development of a costly infrastructure.

2.3. Current DSRC Standards

The development of DSRC standards is closely related to efforts to standardize wireless
communications in ITS applications to achieve interoperability between different
applications, i.e., allow vehicles to interact with roadside equipment and other vehicles
wherever they may be encountered (IEEE, 2004). Among the organizations involved in
the development of the standards are the Federal Communications Commission (FCC),
the American Society for Testing and Materials (ASTM), the Institute for Electrical and
Electronics Engineers (IEEE), the Society of Automotive Engineers (SAE), the American
Association of State Highway and Transportation Officials (AASHTO), the International
Organization for Standardization (ISO) and ITS America. Other stakeholders include
vehicle manufacturers and original equipment manufacturers (OEM’s). Transportation

agencies also have an interest due to their role as users of proposed wireless applications.

2.3.]. Elements Targeted by Standards

DSRC devices typically include transceivers mounted on vehicles or on roadside
equipment and designed to initiate communication with neighboring transceivers upon
demand. Based on the Open Systems Interconnection (OSI) communication framework,
the operation of these devices can be specified by a seven-layer profile, as shown in

Figure 2.2 (Moslehi, 1998):

24

The Physical Layer defines fiequency usage, modulation and coding
specifications for the necessary communication.

The Data-Link layer provides the functional and procedural means to transfer data
between network entities and to detect and possibly correct errors that may occur
in the Physical layer. Within this layer, the Medium Access Control layer defines
the protocols governing the use of the wireless networks while the Logical Link
Control layer manages the data flows.

The Network Layer provides switching and routing functions and creates logical
paths for data transmission from node to node.

The Transport Layer ensures complete data transfer by managing end-to-end
control, such as determining whether all data packets have arrived, and
performing error checking.

In The Session Layer sets up, coordinates, and terminates conversations, exchanges,
and dialogs between the applications at each end. It deals with session and
connection coordination.

:I The Presentation Layer converts incoming and outgoing data from one
presentation format to another.

The Application Layer finally incorporates the specific user program in ITS

applications.

Standardization efforts typically attempt to define the functionality and characteristics of
the various layers described above. However, most of the efforts to date have only

focused on the Physical, Data—Link and Application layers, as indicated by the summary

25

 

 

 

 

 

 

 

 

 

A Application
presentatwn I
s
3 Network
' Data Link
N (Logical Link Control +
Medium Access Control)
:_ Physical

 

 

 

Figure 2.2 Open Systems Interconnection Model

of DSRC standards provided in Table 2.3. In the remaining of this section, the two North
America DSRC standards will be described in details, followed by a general description

of the DSRC standards developed in European Union and Japan.

2.3.2. 915-MH2: North America DSRC Standard

Efforts to develop DSRC standards date back to 1992, when ITS America (then IVHS
America) began discussions with the FCC on communication requirements of electronic
toll and traffic management systems (Pietrzyk and Mierzejewski, 1993). Prior to these
discussions, the absence of a dedicated communication spectrum for transportation led
most of the early experimental electronic toll collection systems and commercial vehicle
operation applications to be designed for operation in the unlicensed 902 — 928 MHz

band dedicated to industrial, scientific, and experimental usages. This led to applications

26

Table 2.3 Existing DSRC Standards

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Country] . Year Intended
. . I . . .
Region Standard No Tit e (last Versron) Applicatrom
Road Transport and Traffic
Telematics Electronic Fee
ISO ISO 14906 Collection — Application Interface 2004 ETC
Definition for DSRC
Standard Specification for DSRC
United StateSI ASTM E 2158 Physical Layer Using Microwave in 1999 ETC/Other“
the 902-928 MHz Band
Standard Specification for
Telecommunications and
Information Exchange Between
United States ASTM E 2213 Roadside and Vehicle Systems — 5 2002 ETC/Other
GHz Band DSRC Medium Access
Control and Physical Layer
Specifications
DSRC for Transport Information
Japan ARIB STD-T55 and Control System 1997 ETC
Japan ARIB STD-T75 DSRC System 2001 ETC/Other
Japan ARIB STD-T88 DSRC Application Sub~Layer 2004 N/A
European DSRC Physical Layer Using 1997
Union EN 12253 Microwave at 5.8 GHz (2004) ”A
European EN 12795 DSRC Data-Link Layer: Medium 1997 N / A
Union (under review) Access and Logical Link Control (2002)
European EN 12834 . . 1997
Union (under review) DSRC Application Layer (2002) ETC
European EN 13372 DSRC Profiles for Road Transport 1999 N / A
Union (under review) and Traffic Telematics Applications (2004)

 

* Other applications include ATIS, CVO, AVCS, E'I'I‘M, APT S and ATMS.
Abbreviations:
APT S: Advanced Public Transportation Systems

ARIB: Association of Radio Industries and Business
ASTM: American Society for Testing and Materials
ATIS: Advanced Traveler Information Systems

ATMS: Advanced Transportation Management Systems
AVCS: Advanced Vehicle Control Systems

CVO: Commercial Vehicle Operations

ETC: Electronic Toll Collection

E'ITM: Electronic T011 and Traffic Management Systems
180: International Organization for Standardization

27

 

that were potentially subject to interferences from other applications using the same

frequencies, such as garage door openers and cordless phones.

Early efforts to dedicate a communication spectrum for transportation usage focused on
the 915 MHz band. In 1996, the US. Department of Transportation started working on
the development of a short-range dedicated communication standard for commercial
vehicle operations to allow electronic tags purchased in one region of the country to be
operable in other regions. In response of this need, Hughes Aircraft (now Raytheon
Systems), made public a number of proprietary protocols they had used for the
development of a 915-Mhz transponder. This protocol was submitted to ASTM as a draft

standard but was not initially approved.

Revisions of the proposed standard were approved in 1999. This standard (ASTM
E2158-01) specifies the communication processes between vehicles randomly entering a
communication zone and the roadside equipment within this zone designed to
communicate with the on-board transponders for both multi-lane and lane-based
applications. It also enables accurate and valid message delivery between moving or
stationary vehicles and fixed or portable roadside communication equipment through a
wireless interface. The applications envisioned for this standard include electronic toll
collection, advanced traveler information, commercial vehicle operations, advanced

public transportation and advanced transportation management systems.

2.3.3. 5. 9-GHz North America DSRC Standard

In response to the need for high-speed data exchanges between roadside equipment and

vehicles, as well as between vehicles, ITS America submitted a petition to the FCC in

28

1997 to allocate a 75-MHz band in the 5850—5925 GHz range for ITS applications while
retaining the 915-MHz band for near-term electronic toll and commercial vehicle systems
already using it. Another basis for the petition for dedicated communication channels
was that the development of related standards would create incentives for the
transportation industry to build devices for the largest possible market, and thus create
assurance for users that particular technologies would not be rendered obsolete later by
different technologies. This petition was approved in 1999 and led to the adoption of the
first 5.9-GHz standard (ASTM E2213—02) in 2002. This standard is based on the 802.11a
WiFi standard to allow usage of built-in security mechanism already widely accepted by
business agencies, to provide product interoperability with existing equipment, and to
reduce the cost of designing and manufacturing DSRC-based products. Since the current
standards only defines the physical characteristics of the equipment using the 5.9-GHz
band, efforts are thus still under way to define the associated protocols and other

communication mechanisms.

As shown in Figure 2.3 (IEEE, 2004), the performance envelope of the 5.9-GHz band is
designed to cover a wide variety of applications not supported by the older 915 MHz
standard. The new standard specifically extends the effective communication range from
30 m to up to 1000 111, when transmitters with appropriate power are used. This allows
for the development of long—range ITS applications. Data rates are further increased from
0.5 Mbps to a range of 6 to 27 Mbps. This enables the development of data intensive
real-time ITS applications, in addition to providing opportunities for high-speed in-
vehicle internet services. The sloped upper edge of the performance envelope further

indicates that a tradeoff exists between the data rate and radio range. Here, increases in

29

data transmission rate typically result in a reduction in the range at which data can be
broadcasted over (MeshNetworks, 2004). This translates into a need to balance
communication range and costs, as the use of higher transmission power levels typically

increases communication costs.

   
 
   
  
    

54

33 .

30 Data Transfer and

Internet Access Services

27
5’
3" 24 5850 - 5925 MHz Band
3 Performance Envelope
2 21 / (Approximate)
U
3

ergency thde Services

Toll and Payment Services

 

0 ‘~~ 902 - 923 MHz and Performance Envelope
0.5Mbps Rangelfl)

Figure 2.3 Data Rate vs. Range of DSRC

 

Source: (IEEE, 2004).

Another important feature of the new standard is its multichannel configuration. While
the 915-MHz standard only provided two communication channels, the 5.9-GHz standard
defines seven channels, each covering a lO-MHz band, as illustrated in Figure 2.4. One
channel is set aside for control applications (Ch178). Two other channels are dedicated
to public safety applications, with one targeting intersection applications (Ch184) and the

other applications using vehicle-to-vehicle communications (Ch 172). The four

30

remaining channels are opened for both public and private applications using short-range

and medium-range vehicle-roadside communications (Zhu and Roy, 2003).

     

Control
_

 

MeqiljngSenJ/gce Short Rng Semce

VehNeh htersections
l— _ l

 

 

 

 

 

 

Power Limit 44.8 dBm 40 dBm
Power Limit 33 dBm
Power Limit 23dBm
U link 1 A
I l . , _W p- l
I I v v I
Downink ‘
‘ """""""""""""""" P GSII'c """ PLBiE'm'PIIBI'IE """"""" r31} 5II'c""'b‘uiaiié'"i=hiliié‘s'aié&""
Safety Safety/ Safety/ Control Safety/ Safety] Intersections
Veh-Veh Private Private Channel Private Private
Ch 172 Ch 174 I‘M Ch 178 Ch 180 Ch 1_82 (£31933:
I I I l l I l I I I I | I I I I I I l I I I I l I I I I l I I I I I I I I I I
In 1.0 o In 0 to o m o In 0 In to o In to o LO
aaasaaaaassa.§asssass
to I15 I15 “.5 In In In In In In In In In to In In In In In In
Frequency (GHz)

Canadian Special License Zones‘

Figure 2.4 5.9 GHz DSRC Band Plan with 10 MHz Channels

Source: (IEEE, 2004).

As indicated earlier, the 5.9-GHz range features lower interference risks from the
relatively few non-transportation applications operating in this range. Another important
consideration is interferences from other transmitters operating the same frequency range.
With the 915—MHz frequency, two communication devices have to be separated by more
than 930 m to reduce the potential for interference (IEEE, 2004). When shorter separation
distances are used, as with electronic toll systems, backseatter-type tag readers with at
least 2 MHz frequency separations for the electronic equipment installed on adjacent

traffic lanes have to be used. For the 5.9-GHz band, the minimum separation between

31

devices is reduced to 50 m, thus making the standard more easily applicable to

transportation applications requiring communications between entities in close proximity.
2. 3.4. Other Developments around the World

The European Union started the development of its DSRC standards in 1991 with the
objective to standardize communications in electronic toll collection systems. These
standards are based on a 5.8-GHz frequency that is not compatible with the North
American standards. These standards further cover data communication exchanges at
distances of up to 100 m at rates varying between 0.6 and 1.0 Mbps. Specific standards
were defined in 1997 for the Physical (EN 12253), Data Link (EN12795) and Application
(EN12834) layers, while a standard detailing DSRC applications (EN13372) was adopted

in 1999 (RI 1 I, 2004).

Japan’s standardization efforts started in 1994. Similar to European efforts, standards
based on the use of a 5.8-GHz frequency were developed. The first standard (ARIB
T55) was adopted in 1997 to provide electronic toll collection services (ARIB, 2004;
Tokuda, 2004). It provided 4 communication channels allowing data transmission rates
of up to 1 Mbps at ranges of up to 30 m. A second standard (ARIB T75) was adopted in
2001 and a third standard (ARIB T88) was adopted in 2004 to make possible the
provision of information services within DSRC, such as payments for purchases at gas
stations, exchange of various information related to parking management, or logistics
applications. For this purpose, the new standard expanded the number of channels to 14
and the data transmission rates to 4 Mbps while retaining the 5.8-GHz band and a 30-

meter communication range.

32

2.4. Existing ITS Driver Assistance Applications

While the 915 MHz DSRC has been established for years, there have been little efforts to
use this communication spectrum to enable wireless communication between vehicles
and deve10p new driver assistance applications. As indicated earlier, this communication
spectrum has been mainly used for the development of electronic toll collection systems.
In particular, one of its main disadvantage with respect to driver assistance systems is the
low data transfer rate it offers (0.5 Mbps). While the 5.9 GHz DSRC standards offer the
promise to resolve the data transfer rate issue, standardization efforts are still currently
underway. As a result, existing vehicle—based driver assistance systems are primarily
relying on radar sensors to amend the limited sensitivity and reliability of human driver.
A review of these systems is presented here. The next section will present how wireless

communications can be used to expand the capabilities offered by the current systems.

2.4.]. Adaptive Cruise Control Systems

Currently, adaptive cruise control systems are the most advanced type of intelligent
vehicle control in commercial use. As indicated earlier, such systems employ range-
sensing technology to adjust the speed of a vehicle to the speed of the vehicle
immediately ahead when the two vehicles are in relative proximity. While range sensing
technology is used to detect the relative position and relative speed of the vehicle
immediately ahead, how this information is used to generate vehicle control commands
for the host vehicle depends on the type of car-following law adopted. Three control

laws are typically considered, as detailed in the following subsections.

33

2.4.1.1 . Constant Spacing Control

In the first control law, the objective is to maintain a specified physical distance with the
vehicle ahead. This law is generally implemented using a control policy of the form of
equation 2.1. In this case, the acceleration command outputted for a vehicle It at a given
time is a function of the vehicle’s current speed differential with the lead vehicle and how
close the vehicle is to the lead vehicle with respect to a desired following distance. The
two constraints of equations 2.2 and 2.3 are further imposed to ensure realistic vehicle
behavior. The first constraint states that the resulting acceleration command cannot
exceed the acceleration and braking capabilities of the host vehicle, while the second

constraint indicates that the accelerations cannot lead to a negative speed or a speed that

would be above the host vehicle’s desired speed, vndC’S .

d
accn (H'Tr) = kl [Viz—l (t) " VII (t)i+k2l:dn—1 (t) "d es] (21)
subject to a,’,’”" < 066,, (t) < a?“ (2.2)
O < v” (I) < vges (2.3)

The philosophy behind the constant spacing policy is intuitive. With proper values of k]
and k2 , the second term considering the relative distance between the two vehicles

should be predominant. An inter-vehicle distance larger than the desired value would
then result in acceleration commands, while following distances that are too close would

result in deceleration commands. The relative speed term, vn_1(,)-vn(,) , is used to adjust

the acceleration command mildly so that steady state can be achieved. If the relative

speed were not considered, significant fluctuation on the following vehicle might exist.

34

For example, consider a fast vehicle that is approaching a slower vehicle from behind. In
typical scenarios, a human driver would gradually reduce the speed of his vehicle to
adjust it to the speed of the vehicle ahead as he closes in on it. In contrast, under
automatic control without consideration on relative distance, an adaptive cruise control
system would first keep accelerating the vehicle until the desired separation distance is
passed. Reacting to the proximity with the vehicle ahead, the adaptive cruise control
system would then decelerate the vehicle until the desired separation is met. However, if
the deceleration pushes the vehicle too far behind, then another cycle of
acceleration/deceleration occurs. Depending on the vehicle’s response on control
command, this fluctuation will diminish, continue, or even amplify with time. Therefore,
the relative speed term can help approaching the desired vehicle following distance
smoothly. However, even with the relative speed term included, other concerns about

constant spacing policy still exist.

First, the desired following distance is fixed and not related to vehicle speeds. For safety

issues, this distance must be chosen with respect to high speed situations. Typically, the

desired spacing with the vehicle immediately ahead in equation 2.1, dd“ , should be set
to allow for safe stopping at all speeds at which the vehicle control system may be
operated. This means adopting a desired spacing between successive vehicles that would
be long enough to allow a full stop at the highest possible speeds. Since the desired
speed remains constant regardless of the current travel speed, this leads to non-ideal
situations when traveling at low speeds as larger than necessary spacing would then be
imposed, thus causing potential roadway capacity reductions. Moreover, the constant

spacing control policy can not guarantee string stability, 3 property that prevents the

35

 

upstream propagation of perturbation introduced by a sudden break of the leading vehicle
or a slow moving vehicle entering the traffic stream (Swaroop and Rajagopal, 1998). A
formal definition of string stability is provided in Section 2.4.1.4. However, to make a
string of controlled vehicles stable with a constant spacing policy is not totally
unachievable, but certain type of vehicle-to-vehicle communication has to be provided to

feed the control law with information from a reference vehicle (Foster, 1979).

2.4.1.2. Constant Time Headway Control

Instead of the constant spacing approach of equation 2.4, most existing adaptive cruise
control systems adopt a constant time headway policy to determine the speed of a
following vehicle. As exemplified by the model of equations 2.4 — 2.6, this approach
seeks to maintain a fixed time interval between two vehicles. Compared to the constant
spacing approach, this control law provides more reasonable car-following operation
since it correctly considers that vehicles traveling at higher speeds typically need a longer

distance to safely follow other vehicles.

“CC" (I+Tr) = k1(v,,_1(,) — Vn (t))+ kzldn—l (I) - (km .1», (Oil (14)
subject to afzm." < ace” (0 < a?” (2.5)
0 < Vn (t) < v3“ (2.6)

Another important feature of the constant time headway control law over the constant

spacing law is that it may guarantee string stability if the calibration parameters k] and

k2 of equation 2.4 and the desired headway behind the lead vehicle, hdeS , are

appropriately configured (Swaroop and Rajagopal, 1998; Liang and Peng, 1999). A

36

string of vehicles is deemed stable if a perturbation in the acceleration command issued
for one vehicle does not amplify when propagating to following vehicles. To achieve
string stability, the desired time headway should be reasonably large enough for a given
k2 and a parameter k] whose values are typically limited by hardware specifications, such
as range sensor noise. Values of the above three parameters have to satisfy the following

relationship (Liang and Peng, 1999):

des2

 

In most of the currently available adaptive cruise control systems, the typical desired time
headway ranges from 1.0 to 2.0 3. Based on these parameter settings, Shladover et al.
(2001) suggests the use of a typical desired time gap of 1.4 s, as such a value would
represent the mid-point in the range of desired time headways considered by current

adaptive cruise control systems.
2.4.1.3. Variable Time Headway Control

In the third control law, the constant time headway is replaced by a variable time
headway. This control law was first proposed by Yanakiev and Kanellakopoulos (1995)
to deal with low braking and acceleration capabilities of heavy trucks. The resulting

control law is expressed by the model of equations 2.8 — 2.11. In equation 2.8, the only
difference with equation 2.4 is the replacement of the constant parameter hdes by the
variable parameter hm) . In this case, equation 2.9 indicates that the time headway to

maintain between successive vehicles remains dependent of a constant desired headway

37

but isadjusted up or down according to the speed differential observed between the lead

and host vehicles.

066;: (t+Tr) = k1 (VII—I (I) - Vn. (t) )+ k 2 [dn—I (I) - (h ’20 "VII (I) J] (2.8)

- d
Wlth 1120 = h es -k3 [vn (t) "_Vn-l (0] (2.9)
subject to azu." < accn (t) < ohm“ (2.10)
o < v" (t) < v5“ (2.11)

Compared to the constant time headway control law, the variable time headway model
allows for the issuance of gentler deceleration commands when a vehicle is temporarily
too close from a lead vehicle that is currently moving at speed faster than its own speed.
The model also allows for the implementation of lower acceleration rates to allow a
vehicle approaching another one traveling at slower speed to gradually catch up with the
lead vehicle until a safe spacing is attained. However, since this model requires more

effort to calibrate, it remains relatively unused in current ACC systems.

2.4.1.4. Performance Evaluations of Current Adaptive Cruise Control Systems

String stability has been considered by many researchers in their performance evaluations
of adaptive cruise control algorithms. In addition to string stability, other measures
including traffic flow behavior, traffic safety, and changes in fuel consumption and
environment pollution have also been used to measure the effects of adaptive cruise
control systems. This section summarizes the usage of these measures in past studies. At

the end, limitations of adaptive cruise control systems are also described. It is worth

38

noted that most evaluation studies on adaptive cruise control systems have been
performed by simulations, mainly due to the fact that very few vehicles in existing traffic

are adaptive cruise control enabled.

The string stability property was first studied by Caudill and Garrard in 1977 (Caudill and
Garrard, 1977). The physical meaning of string stability is that perturbations introduced
by a sudden breaking of the leading vehicle or a slow moving vehicle entering the traffic
stream attenuate as they propagate upstream. Ioannou and Chien (1993) defined in
mathematical terms the string stability by a transfer function of the form expressed by

equations 2.12 — 2.15.

 

 

 

xn— t _xnt _Axdes
03(5): ( 10 0) ((1:00) (2.12)
(XII-2(1) — XII-1(1) )— Ax((n_1)_,1)(,)
v ”('1th “ ”(n-1X!)
G,,(s)= 2. 3
V(n—l)(t)_v(n-2)(t) ( 1 )
a “(fixtl—“(n-IXI)
G" (s): .
“(n-1X0 win-2x0 (2 14’
satisfy 056131, G,‘,’(s)ls1, 03(si st (2.15)

 

 

 

 

 

 

 

 

 

where 5 denotes Laplace transformation.
The above set of equations represents the input-output behavior of a following vehicle
(Hall, 2003). Equation 2.12 defines the transfer function Gdn(s) as the ratio of the
I spacing error of vehicle n and that of the preceding vehicle n-I. As partial requirement
for string stability, the first item of equation 2.15 requires the magnitude of Gdn(s) to be

less than or equal to 1. This criterion implies that when vehicles n and n-1 react to a

39

given downstream traffic change, the difference between the actual spacing in front of
vehicle n and its desired spacing never exceeds the same spacing difference measured on
vehicle n-I, which is the preceding vehicle. Equations 2.13 and calculate the ratio of the
relative speed of vehicle It with respect to vehicle n-I to the relative speed of vehicle n-l
with respect to vehicle n-Z. Similarly, equation 2.14 measures the ratio of acceleration
rate differences. When equations 2.12 — 2.14 all satisfy the inequalities of equation 2.15,
the corresponding vehicle following policy described by these equations is said to be

string stable.

It has been proven that constant spacing control cannot keep vehicle string stable without
data communication between each following vehicle and platoon leader, while constant
headway control and variable headway control have the ability to guarantee string
stability (Swaroop and Rajagopal, 1998; Yanakiev and Kanellakopoulos, 1995). In
particular, string-stable adaptive cruise control systems have been suggested beneficial in
protecting vehicles from rear—end collisions. As an example, Touran et a1. (1999) have
used Monte Carlo simulations to evaluate the probability of collision for a string of
vehicles. Their conclusion is that adaptive cruise control systems significantly reduce the
probability of collision between vehicles under their control. However, they have also
assessed that such systems slightly increases the chance of collision when non-equipped

vehicles follow equipped cars.

Requiring string stability in adaptive cruise control systems is not only critical to ensure
that the automated speed control is comfortable and safe, but also to bring potential
benefits on traffic flow behavior. However, both optimistic and pessimistic opinions
have been found in the literature. On the positive side, Shladover et al. (2001) indicates

40

that the major benefits of adaptive cruise control systems are associated with their ability
to smooth traffic flow. In another instance, both Treiber and Helbing (2002) and Davis
(2004) reported that a proportion of at least 20% of vehicles equipped with adaptive
cruise control functions can be sufficient to avoid the creation of congestion in traffic
streams flowing at near capacity following the abrupt deceleration of a vehicle within the
stream. Based on the smoothing effect of adaptive cruise control, Bose and Ioannou
(2001; 2003) further proposed that the accurate speed tracking and smooth response of
the vehicles with adaptive cruise control systems designed for passenger comfort may
reduce fuel consumption and the levels of pollutants emitted by the following vehicles.
In their simulation results, a proportion of 10% of equipped vehicles in traffic flow
translated into a 28.5% reduction in fuel consumptions and a reduction in pollutant
emissions of up to 60.6% during rapid accelerations. On the conservative side, Kerner
(2003) identified that adaptive cruise control systems may adversely cause extra
congestion near certain bottlenecks. This phenomenon was explained by fact that
adaptive cruise control systems typically require larger gap in merging sections than what

typical drivers usually allow.

Besides the diverging arguments on traffic flow impacts, decreasing traffic flow rates
have been observed in certain simulations in which adaptive cruise control systems were
instructed to use headways longer than what human drivers would typically adopt
(Shladover et al., 2001). When compared to an average time headway of 1.1 s
maintained by human drivers, it was reported that the headways maintained by adaptive
cruise control systems had to be shorter than 1.0 s to increase the traffic flow rate above a

situation in which the systems are not being used, regardless of the percentage of vehicles

41

assumed to be equipped with adaptive cruise control in the traffic flow. Considering that
current adaptive cruise control systems are highly restricted by the technical limitation of
on-board radar sensors, headway settings around 1.40 — 1.55 s are typically suggested for
use with these systems. The simulation results reported by Shladover et al. showed that
adaptive cruise control systems have minor impact on highway capacity. Comparing
with a reference highway capacity of 2050 vph, the maximum capacity increase was
observed when between 40% and 80% of the simulated vehicles were controlled by an
adaptive cruise control system. The maximum capacity achieved within this interval was
2250 vph, which is 9.8% higher than the reference capacity. When more than 80%
vehicles were assisted with adaptive cruise control, the capacity started to decrease,
eventually dropping to 2200 vph when all vehicles were assumed to be controlled by an
adaptive cruise control system. Again, it was suspected that a possible reason behind the
capacity drop with market penetrations levels above 80% was the assumed 1.4 s time
headways imposed by the cruise control systems, which are longer than was drivers

currently maintain when behind the wheel.

Adaptive cruise control systems are also facing practical operational problems. First, they
can onlybe operated above some defined speed. Specifically, current radar technology
typically requires vehicle speeds over 25 mph and a minimum time spacing of 1.4 3
(Patterson, 1998). Hedrick et al. (1999) pointed out other difficulties at low speeds,
including delayed wheel speed measurements, complex torque convert dynamics, and
various un—modeled vehicle dynamics such as the responses of the vehicle’s brake
actuator, throttle actuator and gear shift. As a result, vehicles have to be controlled by

human drivers in slow moving traffic and stop-and-go situations. This adds complexity

42

to the driving task as there could then be frequent switches between automatic and human
controls as the vehicle being driven successively accelerates in and decelerates out of the
range of operational speeds of adaptive cruise control systems. Second, maintaining
large headway on multi-lane highways may create a situation in which vehicles equipped
with adaptive cruise control systems will have to constantly adjust their speed to vehicles
moving into their lane immediately ahead of them. Third, adaptive cruise control
systems are not reliable on roadways with certain geometry characteristics. Radar
sensors used in vehicles have for instance difficulties to locate vehicles on sharp curves
(Kim and Lovell, 2005). Vehicles in neighboring lanes can also easily be misinterpreted
as the lead vehicle. Finally, the ability of adaptive cruise control systems to detect and
monitor vehicles traveling ahead is limited by the detection range of the sensor. As
important is also the fact that no information can typically be obtained from the vehicles
that are traveling ahead of the vehicle immediately in front of the subject equipped. In
contrast, human drivers can often anticipate changes in traffic by observing the actions

taken by drivers ahead of the vehicle immediately in front of them.

2.4.2. Collision Warning and Collision Avoidance Systems

Collision warning and collision avoidance systems are aimed at the avoidance of rear-end
accidents, or at least a reduction in the severity of accidents when a crash is inevitable.
These systems use on-board sensors to continuously monitor the environment
surrounding a vehicle and alert drivers of situations presenting crash jeopardy. The main
distinction between collision warning and collision avoidance systems is here in the type
of actions taken by each system. Collision warning systems are typically designed to
only issue alerts. The driver is then left with the responsibility to take the appropriate

43

actions. Further than merely issuing warnings, collision avoidance systems are also
designed to automatically initiate braking actions when it is determined that a driver has

failed to take appropriate actions to reduce the crash risks.

Collision warning systems have been available for heavy commercial vehicle in North
America since mid 19903 (NTSB, 2001). In particular, they have been reported to reduce
rear-end collisions by 73% (Eaton Vorad, 2004), thus explaining the current interest in
their use to improve transportation safety. Forward-looking sensors associated with
collision warning or collision avoidance systems are usually set to monitor up to around
500 ft ahead of an equipped vehicle (Eaton Vorad, 2004), while rear-looking sensors
typically provide shorter monitoring coverage. At the present time, the on-board radars
are essentially used to determine the relative position and speed of a preceding vehicle, or
relative position of objects located behind a vehicle. As an example, the Eaton Vorad
EVT-300 monopulse radar can provide the position, velocity and azimuth of up to 7
surrounding vehicles, provided that these vehicles are located within a 350-ft long and
l2-degree wide field of vision in front of the sensor (Girard et al., 2001). The main
advantage associated with the use of on-board sensors to detect surrounding vehicles and
objects is that they often offer quicker detection capabilities of roadway hazards than
typical human driver. This leads to well-defined safety benefits when driving in non-
ideal weather conditions, such as in rain, snow, fog and at night, when the reaction time

and awareness of even the most careful drivers can be impeded by limited visibility.

Although collision warning systems have been reported to be useful in cutting down rear-
end collisions, the determination of the exact moment to issue a collision alarm has been

found difficult. An over-sensitive system produces unacceptably large amount of false

44

alarms and appears to be disturbing. However, if a system delays alarms too much, the
safety benefit from using the system would then be affected (Zheng and McDonald,
2004). To realize a wide protection range by issuing earlier warnings while keeping
unwarranted alarms at a low level, many researchers recommended the use of multi—level
warning systems (Horowitz and Dingus, 1992; Lerner et al., 1993; Wilson et al., 1996;

NHTSA, 2002; Morsink et al., 2003).

A typical example of multi—level collision warning systems is the NHTSA Rear-End
Collision Alert Algorithm (NHTSA, 2002). This algorithm provides three sensitivity
levels: near, mid and far. Once the collision warning system is activated within a selected
sensitivity level, it stays in one of four statuses: no alert, early warning, intermediate
warning and imminent collision warning. To determine the appropriate warning

decision, the algorithm makes the following assumptions:

The average delay (i.e., driver’s perception-reaction time and machine lag)
between issuance of a warning and start of vehicle deceleration is 1.5 s.

The vehicle deceleration rate is either zero or one of the constant values
prescribed in Table 2.4. The choice of a value from Table 2.4 is determined by
the selection of a system sensitivity level and a displayed warning level.

n Empirical kinematics equations are used to calculate the minimum gap between
the host vehicle and the vehicle ahead that would occur throughout a braking
process.

The deceleration rates remain constant during braking unless the level of

displayed warning is altered.

45

Close following and other mode of operation such as adaptive cruise control are
handled as exceptions.

The lead vehicle is assumed to maintain its current level of deceleration during
the entire deceleration process.

Table 2.4 Assumed Host Vehicle Maximum Braking Capability.
Source: (NHTSA, 2002).

 

 

 

 

 

. . . Alert Level
SensItIvrty
Early (g) Intermediate (g) Imminent (g)
Near 0.38 0.45 0.55
Mid 0.32 0.40 0.55
Far 0.27 0.35 0.55

 

The NHTSA collision warning system checks for collision hazard every 100 ms. Once a
warning sensitivity level is selected, three minimum gaps that are later used to trigger
alarms are calculated for each alert interval. The calculated minimum gaps during the
projected deceleration are then compared with a vehicle distance gap expressed by
equation 2.16. The anticipated safety gap, dNHTSA , consists of a constant spacing, set at
2 m, and a variable spacing, representing the distance traveled by the host vehicle at the
current speed during a 100 ms interval. The variable portion of the equation ensures that a

minimum 2-m gap will always be maintained throughout an evaluation interval.

dNHTSA = 2 + (Vn )(0-1) (2.16)
If the safety gap dNHTSA is determined to be smaller than the minimum gap
corresponding to a specific warning level, an alert is then considered but not yet issued.
A warning is only issued after three continuous intervals have been checked and when at
least two of the three intervals resulted in the positive determination of a hazard. In such

a case, the highest warning level that has been confirmed in the three continuous intervals

46

will be displayed. Once a warning is issued, it is kept on for a minimum of 1 3 unless a
higher alert level is required. After 1 s, a collision alert may go to a lower level or be

cleared.

The NHT SA algorithm has been evaluated through computer simulation and field tests.
The simulations considered a vehicle approaching a stopped vehicle at a speed of 60 mph
with an initial distance set to 250 m. The simulation was executed 10,000 times for each
combination of deceleration rate from the host vehicle (0.35g, 0.55g and 0.75g) and
distribution assumption on driver reaction time. Four driver reaction time distributions
are considered, all based on a lognormal distribution but with each defined with different
parameters. Results based on Chang’s driver reaction time distribution model Chang et
al., 1985) was found to be the most pessimistic (i.e., the highest collision rates), while
Gazis’ model generated the most optimistic results (i.e., the lowest collision rates).
Collision rates conditional to each deceleration rate were compared across the four
reaction models. When the deceleration rate was fixed at 0.75 g, the probability of
collision was reported in the range of 0.254 — 0.353. When the deceleration rate was
lowered to 0.55 g, the range of collision rate decreased to 0.017 — 0.129. For softer
deceleration (i.e., 0.35 g), the probability of collision was 0.00002 — 0.016. In these
results, collision probability seems to increase with higher vehicle deceleration rate,
which tends to delay the NHTSA collision warning algorithm issuing an alarm, and

subsequently leaves shorter duration for drivers to initiate braking.

The NHTSA algorithm was also tested on both test tracks and public roads. Scenarios
staged on test tracks covered major rear-end collision reasons (e.g., stopped leading
vehicle, low-speed leading vehicle, and braking lead vehicle) while the public road tests

47

were conducted during the development phase of the NHTSA algorithm to verify its

performance and sensitivity. Table 2.5 summarizes the major problems identified in the

field tests, as well as the possible solutions that have been put forward.

Table 2.5 Selected Field Test Results

Identified Problem
(Cases)

Reasoning
(Solution)

 

Late alert than theoretical result
(approaching static object)

Capability limitation of radar system

 

Early alert than theoretical result;
could be later in other cases
(approaching slow vehicle)

Inaccuracy in acceleration resolution

 

Large amount of false alarm at low speed
(Local Driving)

Disturbance from roadside objects
(disable CWA at low speed)

 

(Disable early warning;

Alert when the driver was already braking shrink assumed reaction time to 0.5 s)

 

(Disable warning when acceleration

Alert when passrng is larger than a threshold)

 

No alert when extremely close to preceding
vehicle at highway speeds

Algorithm property
(add “tailgating” mode to the CWA System)

 

In close-following cases, it was discovered that the system saved approximately 1 s in
reaction time, when vehicles were traveling at 60 mph, and the lead vehicle applied brake
to provide a 0.3 g deceleration. However, it was noticed that in some scenarios the
performance of the algorithm was dependent on the capability of the radar system to
report valid targets on curves and at longer ranges. It was also reported that the algorithm
performance was most affected when the host vehicle was traveling at higher speeds. For
example, in one trial of the “approaching stopped lead vehicle at 60 mph” scenario, the

target was not reported to the alert algorithm until after the theoretical imminent alert

48

range was passed. In this case a command to decelerate immediately at a rate of 0.55 g

could not avoid a collision, thus constituting a system failure.

The higher deceleration rate value used by the NHT SA algorithm is close to the edge of
braking capability of modern passenger cars, while the lower rate occurs more often in
normal driving conditions. These rates can be compared to a study conducted by Zheng
and McDonald (2004), who analyzed more than 8000 normal braking events on freeways
and in downtown areas. These observations covered major causes of braking, such as
signals, merging and diverging, ramp metering and roadside disturbances. Based on
statistical results, Zheng and McDonald found that drivers tend to use early and gentle
braking in normal driving and braking events. As shown in Figure 2.5, deceleration rates
over 0.2 g were very rare, even for stop-brake situations on freeways where drivers are
most likely to adOpt harsh braking. In this case, the average deceleration rate at the 99
percentile is only 0.343 g. They therefore recommended a range of 0.2 g — 0.25 g for
imminent collision warning threshold. Although the high deceleration rate of 0.55 g is

achievable, it occurs very rarely in normal driving conditions.

In collision warning systems, it would be reasonable to use high deceleration rate to delay
collision warnings and reduce unwarranted alarms. However, delaying all possible
alarms is not a panacea to treat false alarm problem, as this may result in less stable
traffic, less safe driving, and less comfort as driver would get to know later of the
existence of dangers and would tend to brake later and harder. To provide customers
with an effective and easily acceptable product, a possible solution to enforce collision

warning sensitivity without making the whole system annoying is to integrate

49

     
 

1.00

 

P980. 0%9’8' 0'39

0.98 -
0.96 -
0.94 -
0.92 -
0.90 -
0.88 i
0.86 A

0.84

Percentile

P850. 0139

 

 

 

j

0 0.1 0.2 0.3 0.4 0.5 0.6
Ave. Deceleration [9]

Figure 2.5 Cumulative Distribution of Deceleration Rates under Normal Driving
Conditions

Source: (Zheng and McDonald, 2004).

information from vehicle beyond the immediate leading vehicle. For instance,
consideration of downstream traffic information may help reduce false alarm rates in
situations in which the vehicle immediately ahead is slightly decelerating but the vehicles
further ahead are not. At the opposite end, reception of a message reporting of a sudden
traffic breakdown downstream of vehicle’s current position may help alert a driver of the

danger ahead and allow him to take appropriate preemptive actions.
2.5. Envisioned Wireless-Based Driver Assistance Applications

Accessibility to downstream vehicle and traffic information in the above situations is
made possible by emerging vehicle-to-vehicle communication capabilities. In addition to
improving collision warning systems, these emerging communication capabilities can
also be used to improve a variety of vehicle-based control systems. Feeding vehicles

with downstream traffic information from multiple preceding vehicles creates for

50

instance an opportunity to improve existing driver assistance systems. Driver assistance
systems using wireless communications to expand their functionalities are commonly
known as cooperative systems. This section reviews more specifically documented
studies regarding the development of cooperative adaptive cruise control systems and

c00perative collision warning and avoidance systems.

2.5.1. Cooperative Adaptive Cruise Control

As previously reviewed, primary limitations of existing adaptive cruise control systems
include excessively large vehicle gap, limited monitoring range, and trivial improvement
in traffic flow capacity. To maintain traffic safety while allowing vehicles to travel
closely at high speeds, mechanisms to provide more prompt response to changes of
downstream traffic are necessary. In such a case, the development of more accurate and
long-range radar sensors may not provide an adequate solution, as these sensors would
still be limited to only monitor vehicles that are immediately adjacent to them and within
a direct line of sight. By allowing vehicles within a given range to exchange information
between themselves, cooperative adaptive cruise control systems offer a solution to the
problem associated with existing radar-based systems. First, with the ability to broadcast
information from one vehicle to the next, the effective range of traffic monitoring is
greatly expanded. It further becomes possible to collect information not only from the
vehicle that is immediately adjacent, but also from vehicles that are two, three, four or

more vehicles away, or from vehicles that may currently be out of sight.

Although cooperative adaptive cruise control systems are still in a development phase,

some preliminary experiments have already been conducted (Kato et al., 2002). To date,

51

the only documented cooperative adaptive cruise control algorithm is the one that was

developed by Shladover et al. (2001). This model is formularized in equations 2.17-2.23.

accn ([+Tr) = kOan—l(t) + k1 (”n—1 (I) ' ”1(0)

 

d (2.17)
+ k2[(xn—i (t) _ xn (Ol‘ Axni‘: (0]
des _ h b s .
Axn—i (t) — MAX(Axn—i (D’Axn—i (D’Axn-l l (2.18)
2 2
v v
AIL. (t) = " (i) — "—1 it) +0.02x v3;_1 ( I) (2.19)
2a},nln 20:3?
’1 _
Axn—i (t) — 0'5 X ”n (t) (2.20)
S _
Axn-i(t) — 2 (2.21)
subject to all" in < ace" (1) < a?!” (2.22)
0 < v,, a) < v3“ (2.23)

Similar to existing adaptive cruise control models, the model described above will
automatically accelerates a host vehicle to a desired speed in the absence of any lead
vehicle. As indicated by equation 2.17, the acceleration command for the host vehicle,

accn , is determined by considering the current acceleration level of the vehicle

immediately ahead, the differential between the current speed of the host vehicle and the
speed of the vehicle ahead, and the current distance between the two vehicles with
respect to a given desired distance. As indicated by equation 2.18, the desired distance
between two vehicles can be taken as the maximum of three distance criteria. The first
criterion, expressed by equation 2.19, sets the desired distance so that the two vehicles

would not collide should they both have to break at their maximum deceleration rates

52

while considering the number of vehicles between them. In this case, the second term of
equation 2.19 introduces a 0.02 s perception time based on an assumed 0.02 s time
needed for wireless communications. The second criterion, defined by equation 2.20,
calculates the distance to maintain a given minimum time headway behind the lead
vehicle. The third criterion, defined by equation 2.21, finally states that a minimum

spacing of 2 m must be maintained at all times between each pair of succeeding vehicles.

Unlike typically envisioned cooperative adaptive cruise control systems, the model
described by Shladover et al. only allow wireless communication transmissions between
immediately adjacent vehicles. Nevertheless, the ability for vehicles to exchange

information with their neighbors is seen as a valuable addition. As a result of this ability,

it was determined that the constant time headway governing the spacing of vehicle, hdes ,
could be reduced from 1.4 s in a typical adaptive cruise control setting to 0.5 s in a
cooperative setting as vehicle control systems would then have greater awareness of what

is happening downstream of their position.

The above cooperative adaptive cruise control model was tested in simulation to control
all vehicles along a 10-mile section of a single-lane highway. The objective of the
simulation was to measure the traffic flow capacity that can be attained with 100%
market penetration. To allow the measurement of the maximum stable flow, the
simulated test section was divided into 10 equal-length segments connected by nodes
representing on and off ramps. At each node, the entering and exiting flows were fixed at
200 vph and 100 vph, respectively, resulting in a gradually increasing flow as we move
downstream along the freeway segment. The traffic flow entering the highway at the

most upstream node was varied to simulate various levels of congestion. The simulated

53

flow started from an extremely high value explicitly set above capacity and was reduced
in decrements of 50 vph for each simulation run. When a scenario in which queuing did
not appear within 90 minutes of simulation time was identified, the previous vehicles
flow reaching the end of road was recorded and assumed to represent the traffic flow
capacity corresponding to the specific driver assistance system being tested. Using this
method, it was estimated that a traffic stream with 100% of vehicles equipped with
cooperative adaptive cruise control could produce a maximum flow rate of 4,550 vph.

However, these results remain theoretical as no corresponding field tests have been made.
2.5.2. Cooperative Collision Warning

Parallel to the improvements that cooperative adaptive cruise control systems provide
over basic adaptive systems, elementary collision warning systems can be extended to
cooperative collision warning systems by the introduction of vehicle-to-vehicle
communications. An example of development of such a system is provided by the
European CarTALK 2000 research project (Morsink et al., 2003). Similar to the
previously reviewed NHTSA collision warning system, the CarTALK2000 system is
based on a multi-level alert design. Using vehicle—to-vehicle wireless communication,
the monitoring range was extended to as far as 1000 m or 5 vehicles downstream,
whichever is exceeded first (Asher and Galler, 1996). The decision-making process of
CarTALK 2000 cooperative collision warning system is expressed by equations 2.24-

2.26.

54

( :I: an: )
accn(t+T) =MIN accn_,-(,+T),accn_,-(,+T) .
(2.24)

where i = I..k and k = # of downstream veh.

1 es
5 iVn(t) — VII—i(t)i2 + an—l(t)l:Ax,(f_l-(,) _ Axn—i(t):l

l .
[Airing —Axn__,-(,)]+T[Vn(z) - vn—rm -§an—r(r)T] (2.25)

when Vn—i(t+T) at 0

*
awn—Ion“) =

 

area: V30 )an—i(t)

“CCn—i(r+T) = des '
an—i(t )[Axn—rit) — mil-l(1)+ V110 )7] + VII—III) (2.26)

 

when Vn—i(t+T) = 0

In the above equations, the safety margin AxdeS was set to 15 m for each pair of
neighboring vehicles, and driver reaction and machine lag was assumed to be 1.8 s in
total. The remaining parameters are obtained by the host vehicle itself. Considering the
case of a three-vehicle platoon, information from the leading vehicle is transmitted to the
vehicle in the middle and used for making warning decisions. The last vehicle in the
platoon then receives information generated from both preceding vehicles. Following the

calculation of accn(1+T) , which represents the minimum necessary deceleration to avoid

colliding with the vehicle immediately ahead, an appropriate warning level would then be

determined based on the parameters of Table 2.6.

The CarTALK 2000 system has been tested on three identical experimental vehicles. The
collision warning algorithm design and performance tests were restricted to the only case
of three identical vehicles traveling in a single lane. The leading vehicle was used to

simulate hazardous conditions through the sudden application of brakes. Collision

55

Table 2.6 Warning Levels of the CarTALK2000 System.
Source: (Morsink et al. 2003).

 

 

Necessary Deceleration (tn/32) Warning Level
-0.8 <= ace 0
-1.2 <=acc<-0.8 1
-1.7 <=acc<-1.2 2
-2.1<=acc<-1.7 3
-2.5 <= acc < -2.1 4
acc < -2.5 5

 

warnings were issued to drivers of the two following vehicles when braking was needed
to achieve certain level of deceleration. Experiment results showed that the proposed
cooperative collision warning systems shortened the reaction time of the driver of the last
vehicle by 0.8 s when active braking aid was not provided (i.e., only warning was issued).
With active braking, this reduction was enlarged to 1.7 3, mame because the reaction

time of the drivers from the middle and last vehicles were both eliminated.

In addition to the cooperative adaptive cruise control and collision warning systems
reviewed above, there have also been other efforts towards the development of
cooperative systems, such as an European project named CHAUFFEUR, which
experimentally demonstrated automatic heavy truck following (CORDIS, 2005), and a
study conducted in Japan on cooperative vehicle navigation (Yamashita et al., 2005).
These later applications not only focus on passenger comfort and safety, but attempt to
generate a range of benefits extending from automated vehicle control to informational
services. A general view of these automotive applications is provided in the following

section.

56

2.6. Other Automotive Applications Enabled by Wireless Communications

While advanced cruise control and collision warning systems have been the focus of the
above review, it is worth noting that other ITS applications taking advantage of modern
communication infrastructures may as well be developed. As an example, Figure 2.6
illustrates a vision developed by ITS-Michigan for data flows within an integrated
vehicle-infrastructure system (Luckscheiter, 2004). Within the vehicle—infrastructure
integration (VII), high-speed internet connections are seen as the backbone of the
communication system. Such connections would allow information exchanges between
database hosts, traffic management centers, and telematics service providers. Vehicle-
roadside communication nodes would be used to send or retrieve information from
vehicles passing near fixed communication nodes, while vehicle-to-vehicle
communications would allow information to be propagated among vehicles without the

need for extensive roadside communication infrastructure.

Roadside-to-vehicle communications first enable the provision of in-vehicle navigation
services to motorists. Comparing with traditional means of receiving information
through highway advisory radios, variable message signs or cell phones, emerging in-
vehicle communication systems adds the capability to receive wireless broadcasts from
either roadside equipment or surrounding vehicles. This makes it possible to consider
applications using data uploads at strategic points to provide travelers with local maps
and business directories. The uploaded information can also be used to inform travelers
of construction zones and congested traffic conditions ahead and to propose alternative

routes. At the opposite, roadside-vehicle communication links can also be used to

57

retrieve information from passing vehicles, such as travel time between known points,

thereby converting each passing vehicle into a probe vehicle.

    

 

 
   
 

 

Internet — Wide Area Network
.. air
/ State Wide
y ' Vehicle Probe Data 3“?“
_/ Incident Into 3 °" ":3
Maintenance Info Cmom
SatelIte—to- we'm" 1““ Speeds
Vehicle Vehicle Locallzed Incidents
Integrated Malntenanc
Cellular Roadway Weather
Segment
M . Conditions , ,
DSRC Lo

 

 

 

 

 

30"" , State Wlde MDOT
/ \ // \ .............
awn" . Road Dianna”:

n 1133111: rnmc I1 :1

 

 

 

 

datasy
‘ RoadsycMap data
Vehicle-to—Vehlcle ' ' i ‘ ' gal-time traffic info
(DSRC Floating Car -‘ m Vehicle probe data
Network)

Roadside & Intersections
(Loops, Cameras)

Figure 2.6 Data Flows in Integrated Transportation Systems

Source: (Luckscheiter, 2004).

Roadside-to-vehicle communications further enable the development of applications
using in—vehicle devices to enhance the awareness of drivers of possible roadway hazards
or alert them about hazards that are beyond their field of vision or the vehicle’s own
communication range. Applications can for instance be developed to alert drivers of the
presence of water, ice or snow on the pavement ahead. Applications can also be
developed to warn drivers of a potential for collision at an intersection following the
detection of a vehicle fast approaching the intersection while facing a stop sign or when
the green signal is about to or has already terminated. Similar warning systems can also

be developed at railroad crossings to warn drivers about approaching trains.

58

Another potential application is emergency vehicle preemption warning. Currently,
emergency vehicles indicate their presence through the use of sirens. This sometimes
confuses drivers when they cannot determine the direction from which the siren is
coming. Unnecessary traffic disruptions may also result from vehicles slowing down for
an emergency vehicle that is not on the same street. With vehicle-to-vehicle
communications it becomes possible to consider warning systems alerting only drivers
that are in the path or conflicting paths with an approaching emergency vehicle. An
extension of the warning range is also theoretically possible, provided that there are

vehicles capable of propagating the information along the way.

Besides the applications introduced in earlier sections and described above, many other
uses of wireless communication have been considered. An extensive list of the currently
envisioned primary applications is provided in Table 2.7, which enumerates individual

applications for each of the seven categories of Table 2.2.

59

 

5,3. 9 wEEuB 60.5 .250 22.880

H 8:205 588.25 Susana H 8.029 2 sages “5205 H Baggage 9 sec 5.5
pine .58 £335

H OCH. 2 maoHHHHEoU 083m 38% H wag? owEem EH H 835m 23.5 8 Sea 580
H wuaoHEsonH 933D an: H o>umav< - MEEH< £363: H 30803an $59th gum—“HE
H mEEaB 8:58 8:830ch > E553 EB c3

H BEE“? 2029 > minnow "Hanoi H “8:935. - 0583c 0:3

-ch wEEaB nous—o; 3:me own—EH.

 

85.5935 E893 :ouan.=<

> wEFaB eds—Hum .8320on > >~3£MHHH-0HoEu> o>ua5noou

> mEFaB omen—Ho 0:3 > concave. oaHm o>ufioaoou

. mafia?

eH bofincm 5:365 > mswammofi ESE: > 56:80 H.558 032380
> cannon uaom oHoEonwwmnwfifl > 2533 0908 aascwHHH > mEEaB 53:8 032380
> ouswaawé EoEo>oH>H cmHm aoam > wEFaB ozuom u< oHoEo> aocombEmH > 35:00 9430 o>Hamo< 058380

36E? mEEaB

> .850 9 MEFEB :oHHHHEoU 30% > minuaoamxe. oHoEu> HogwbEm > minus; 8% 9:5
> wEFaB seabemom > 3&5 oxflm oEoboonH HoEmbEmH .> wEEmB 0908 2:5

 

 

€258 Susi}. 5.88 52.233 25 .555 =88=ee<

 

 

.38 £85.65 32:8
223:..2 m: 32a§2§ 8.35m 2 use.

 

Gage
SHEER Sun oHoEu> HHmHaHrH.

 

H mHHHEaB 53.6 $3-»:ng

H mean? 23 ~53
H 880%.: EH8 203.5
H Egg :8: Beam
HonoQoyH HHHo>mH 593m

 

 

HSEomeHaSH Homm< 3309
0:82:85 2%.; seem 25$

HHHoEowaHaE HooHnH

8250 uEaHH. oHoEoyH
2263.25 cofiamHmonH

Hm HHoHHmEHoHHHH wHHHmmoHU HHaEmouonH
305.820

2305 buEEoHH 03.55 32:90
HonHaHH. San boeam 289:0
uwéma :23: 22625

MEEMB :onHHHOo HHSEMBHHMHHH

 

       

on $8 $25

EHHSEHHm 52h.
£95 :935 ch 98 3:52:38

HHoHHHHEooHnH Rama 2029 Hofiwbfim
San boHam 2qu> HaHoHoEEoU

85.520

28.8on 2028/ 3655800
bofl>n< oHoEo> HaHoHoEEoU
HHHoEowEaSH mHHHmmoHU HovHom

2023, 5 c3 35 E 880m

 

H madam? oHBuHEm wHHUHHmnH 33 H conawfiaz

. HE.w 85630 88% Bofiacm

H mHHHEHes 03:5 33 H baud 82> oHoEo> xofiwbfim

Hana - ago E23

H case a DEE 202; is: H 9.28 Bees? ”was 33 H 0&5. 333 22.3» 35905

H mEEmB cons—05 :me HHon H Hobcou 30E 05.5. HcomHHHoHHHH H coquHoHHHH uEaHH. £82an

H 8338 mOw H wHHEBoSH acaméo HHHowHHHoHHHH H REFS?

. Ho>o=ob MES; 30% 9.50

25—. .EEoD €3.39 nouauHHHHHH< 2H3. .EEoU 93:5an .558?an KPH. .EEoU APHnoOV :e53=HHHH<

 

 

 

.3258 38 58:33: 825%? w: H.83a8_8§ BeeHmHEHH Hun uses

61

.SBSHEHEES Augosbmabaw oEmHHmoMHéHéHoEoH/ H

.HHoHHauHEHEEoo uHoEo>.oH-oHoEoN H

 

HI—(D—{P—{F—‘i—‘H

9:;on 30>»

38338 252335

o>u - £285 38 new H25
moHoEo> 2 Sum 08F Heath
mEHoéoyH oHoEo> HHwHaHrH.
mafia: 225$ Exam
8325 03.58

53>on
835m 9H. 35 2053/ 955%
Baez HH§~H bag

@5885 EU :35:

Hcoéam omaHmOHHo‘H wEv—Hmm

MEHHOH. 32ngon

8.288 =88HHHHHH<

 

I—lfi—(I—‘V-{Hl—(i—‘D—‘H

H

I—t

HHuEAH mEHEom \wHHEoooyH
8303 Sam was—am

sun 88 55238 a 535
2:282 HHmUéO

:oaawraz Eneméo
wEHHomom HHoEmHSH
mEHHmoHHHBoQ 8:32
muaoHEsoQ £32

wctoHEoE Hoe 038583

HonHHEH. Sun 030an3

wEmEo>HH<HmEQQonm 6039:0383

cosmoEHoZ annex oEF-HHH-HmHHH

Hcofixmm a. @9585 coon
HHHoEHHanH 25.8on

6.258 829:9;

 

l—dt—tl—{HI—‘D—(HHI—q

I—l

H

:02 85:8 H85 22%?5
82> a. £23. 6525
wHHoHHgSmoyH HoHoHH oHHHoméH
O>U - EoEumaHHaE HooHnH

38H. mHHonmHEmH 2:3ch
53% boHooHE

ago—H509 232930
”538on 83.5% HoHSnH .80

baomxaozcom HHHoEHEoHHHZ
HHoHHHHEHoHcH Ho>E.H. 25:4.

@5833 .m8 SEEowHHHo>c<

80:95on
a. San HoEonHHU Ho 5:63.93“

:38me 859.9th
m:=2c.85 \ Egg 5an

5.88 59332

 

 

.HH..H§8 H38 £9233: EoHHSHHE< m: EQEEBHB BSHWHEHH 2 «He;

62

2.7. Summary of Literature Review

This chapter reviewed the current state-of-the-art in wireless communications standards
and usage of wireless communications in the field of transportation and the technologies
currently available for vehicle-based driver assistance and automated vehicle control ITS
applications. Specifically, the standardization and main features of 5.9-GHz DSRC were
presented in detail, and both elementary and advanced versions of vehicle-based driver
assistance systems were successively described. The main observation that can be made
from the review is that traditional radar-based adaptive cruise control and collision
warning systems are very limited in practical use due to the nature of radar sensors. With
vehicle-to-vehicle DSRC becoming available, the accessibility to traffic data in addition
to the vehicle immediately ahead prompted the development of cooperative adaptive
cruise control and cooperative collision warning systems. Both cooperative systems
amend their predecessors and, according to simulation study, offer significant
improvements. In this context, however, it is worth noting that many problems still need
to be solved before the envisioned vehicle-based driver assistance systems could be
successfully deployed. The next chapter will identify these problems and specify three of

them that this research will specifically focus on.

63

CHAPTER 3

PROBLEM DEFINITION

To date, most of the research on automated vehicle control technologies has focused on
specific type driver assistance functions. This means that the various studies that have
been conducted have typically specialized in single applications. Within this context,
each study appears to have correctly reflected the individual influences of each
application on human driver and traffic flow behavior. However, the effects of
combining several wireless-based applications that are expected to operate
simultaneously still remain unclear. Without due consideration, this uncertainty has the
potential to grow over time as new wireless vehicle-based ITS applications are developed
and will start competing with existing applications for the available communication
spectrum, vehicle control resources and driver attention. Within this context, this chapter
aims to identify the gap between existing technologies and application setups that would
allow the simultaneous existence of multiple driver assistance systems. These definitions
delineate the problems that will be addressed in the remaining parts of the dissertation.
They will also help realize the importance of solving these limitations and identify the

methodologies leading to the solution.

3.1. Envisioned Problems with Systems Implementing Multiple Applications

While various applications are currently being researched and deveIOped, the prospect of
having multiple applications installed on a single vehicle create implementation scenarios
much more complex than the deployment of single applications. In many cases, co-

existing applications will need to interact with each other to share common information,

64

coordinate their decision-making process and access the driver’s attention. For example,
both a cooperative collision warning system and a cooperative adaptive cruise control
system can share information on speeds and acceleration rates that are transmitted over a
wireless network from the vehicles ahead. However, the two systems would have
different goals in using the data. On one hand, the collision warning system would aim to
maintain a safe following distance between the vehicle under its control and the vehicle
immediately ahead with the explicit objective to reduce the potential for rear-end
collision. On the other hand, the cooperative adaptive cruise control system would
operate under the objective to improve driving comfort and highway throughput by

allowing safe and smooth car-following.

The safe following distance sought by collision warning systems usually means
maintaining a gap between vehicles that allows for safe breaking distance considering the
reaction time of drivers and the deceleration capabilities of vehicles. At highway speeds,
this safe following distance can be much larger than the gaps that can be achieved under
cooperative adaptive cruise control since there is no driver delay when reacting to
downstream traffic changes. If the two applications are simultaneously implemented on a
vehicle without appropriate coordination, setting vehicle headway according to either
individual system would be problematic. Suppose the shorter headway suggested by
cooperative cruise control system is selected for both systems, the collision warning
module would keep issuing false alarm since its safety requirement is violated. On the
contrary, forcing cooperative adaptive cruise control system to abide by unnecessarily

large headway to keep collision warning system from issuing false alarm may create a

65

 

non-optimal situation in which large vehicle spacing would encourage other vehicles cut

in and interrupt the original group of vehicles under cooperative driving.

Another kind of problem arises from the need for each ITS applications to run
simultaneously to compete for limited data communication and processing equipment.
When available resource is insufficient or the overall system is overloaded, more data
packets are generated than the system devices can handle. The excessive data packets are
then piled up at their origins or at communication nodes with bottlenecks and temporary
put in communication queues. In such cases, the effectiveness of wireless-based ITS
applications are threatened by either data communication delay or data processing delay,
or both of them. However, common solutions such as improving the system resource
limitations are unlikely to alleviate the problem since vehicle-to-vehicle communication
rates are limited by available bandwidths defined by wireless communication standards.
An upgrade to the on-board data processing equipment may also significantly increase
the price of in-vehicle communication system and threaten the adoption of such systems
by vehicle owners. In other alternatives, it may be desired to transmit and process data
according to their relevance to critical applications instead of on a simple time-based rule.
In such a case, important applications such as safety-related applications can be allowed
to send and receive information with higher priority levels to reduce the likelihood of
these applications to be affected by data communication and processing delay when the

overall system is fully loaded.

The problems of coordinated decisions and data communication and processing delays
are merely two examples of problems faced by multiple ITS applications using vehicle-

to-vehicle communications. An analysis of the problem at hand has for instance led to

66

the identification of a need to address the following five research areas to enable the
simultaneous operation of various ITS applications relying on wireless communications

to obtain data critical to their operation:

a Coordinated decision-making algorithms

Impacts of wireless communications delays

:3 Coexistence of vehicles on different levels of automation
Data management strategy

Data fusion algorithms

The research problems associated with each of above-mentioned research areas are
reviewed in more depth in the following section. As it will be detailed later, specific
research problems associated with three of these research areas will be the object of the

current dissertation.

3.2. Problem Specifications and Potential Solutions

Specific explanations of the problems listed in the previous section are provided in this
section. For some of the problems, a few solutions that have been suggested in literature
are reported here. In particular, relatively few solutions are detailed as none of the topics

addressed have been extensively studied to date.

3. 2.1. Coordinated Decision-making Algorithms

Decision-making algorithms realize the core functionalities of vehicle-based driver
assistance applications. In particular, they offer the potential to harmonize potential

conflicting commands between different applications when appropriately integrated. For

67

instance, cooperative adaptive cruise control function and cooperative collision warning
function can be welded together to provide a smooth and safe ride. However, the
integration of decision-making algorithms from various applications must consider their
often differing operation contexts. For instance, although the conceptual cooperative
adaptive cruise control system and the latest collision warning system monitors are
designed to consider up to a certain number of downstream vehicles, current cooperative
adaptive cruise control system tested in computer simulation only take into consideration

the immediately preceding vehicle,.

For a successful integration, it would be necessary to change in a first step one or both of
the two systems and let them share uniform input information. Following that, a single
algorithm could then be developed to allow uniform vehicle control commands to be
made. A desired property of the combined vehicle control function should be to provide
smooth acceleration commands with reasonably close vehicle spacing. As a result,
instead of a collision alert to be issued by a collision warning system on the simple
consideration of a close spacing with a vehicle ahead, the decision to issue an alert, and
which alert level to consider, may be made function of the observed speed and position of
other vehicles ahead. In another example, the automated cruise control system may be
ordered to initiate small decelerations in response to low-level threats, leaving the option
to alert drivers only when a required braking may exceed the maximum deceleration rate

achievable by automated driving function.

68

3.2.2. Impacts of Wireless Communications Delay

Cooperative systems are based upon the ability of individual vehicles to collect
information about the speed, position and other relevant characteristics of surrounding
vehicles made possible by short-range wireless communications. Although wireless
communications allow the establishment of direct communication links with neighboring
vehicles, these links typically never provide instantaneous data exchanges. While
communication delays are generally not a concern when collecting informational-based
data, such as road maps, directories, or establishing in—vehicle Internet services, they can
critically affect the operation of vehicle control systems and impact the safety of vehicle

occupants, as will be detailed in an example below.

The amount of delay associated with wireless communications depends on a number of
parameters. One of the most important factors is the number of simultaneous broadcasts
being attempted in a given area. In this case, the amount of data traffic placed on a
communication network is a function of the density of vehicles equipped with wireless
emitters/receivers, the length of the individual messages being broadcast, the robustness
of routing protocols used to carry a message from an origin vehicle to the vehicles to
which it is intended, the rate of communication errors at transmission nodes requiring
rebroadcasts, and the level of background data traffic from other applications using the

same communication channels.

The potential impacts of communication delays on vehicle control decisions have been
studied by Biswas et al. (2005). Depending on the communication network configuration

and loading, it has been determined that communication delays between a given pair of

69

vehicles could range between a few milliseconds to a few seconds. While an information
delay of a few milliseconds may not significantly affect vehicle control systems, delays
of a few second may introduce significant errors between the reported and actual
positions or speeds of a vehicle. In the presence of such errors, collision risks could arise
from any delay in the application of brakes following the detection of a roadway hazard,
particularly at high-speeds, where such a delay may lead to a need for a deceleration rate
that is beyond the capability of a vehicle. Delays in brake application may also result in
harsh deceleration rates that are uncomfortable for vehicle passengers. Another potential
nuisance for passengers may arise from the implementation of constant speed
adjustments to correct previous speed decisions based on inaccurate information. If these
errors and constant corrections are in turn amplified as they propagate upstream a traffic
stream, a potential may then further exist for the creation of unstable traffic flow.
Ultimately, the importance of information delay is linked to the acceptance of automated
vehicle control systems by the public as it can be expected that only systems perceived as
reliable and safe will be accepted. Although there is no practical means to eliminate
communication delay, the impact of using delayed information in vehicle control systems
should therefore be studied. Methods to determine the validity of data prior to their use
in decision-making algorithms, such as to determine whether the received data are too old

as a result of communication delay, should also be developed.

3.2.3. Coexistence of Vehicles with Diflerent Level of Automation

Besides the fact that multiple ITS applications may be present on one vehicle, each
vehicle will also have to deal with other vehicles on the highway. While it is easier for

vehicle control functions to perform well on a fleet of identically equipped vehicles on

70

test tracks, significant difference in performance may occur when they are interacting
with neighboring vehicles that are highly or marginally automated, or even entirely under
manual control. In the foreseeable future, this will remain an inevitable issue. For
instance, considering that there are 220 millions cars, trucks and buses in the United
States and that 15 to 17 million new vehicles are typically sold every year, it would take

well over 15 years to replace all vehicles.

Understanding the relationship between market penetration and potential benefit is
particularly important to ensure the successful deployment of the proposed systems.
Some applications may yield near 100% benefits with low market penetration but other
may not reach significant benefits until a certain level is reached. It is for instance
desired that proposed cooperative driver assistance systems be initially effective in low
market penetration environment since most existing vehicles are only equipped with
elementary cruise control function. While an increasing number of vehicles are equipped
with adaptive cruise control system, these vehicles obtain their adaptive capabilities
through the use of on-board radar and not wireless communication capabilities.
Cooperative cruise control systems are on the other hand dependent on wireless
communications. With low market penetrations, adaptive cruise control systems are
therefore more likely to be beneficial than cooperative adaptive cruise control systems.
In such a situation it would therefore be crucial for cooperative adaptive cruise control
systems to be able to perform adaptive cruise control function when surrounding vehicles
that are not of the cooperative type. With the proportion of cooperative adaptive cruise

control enabled vehicles growing, increasing chances would then that clusters of similarly

71

equipped may travel together. In all cases, it would be interesting to know whether a 50%

market penetration of cooperative systems would yield 10%, 20% or 50% of benefits.

3.2.4. Data Management Strategy

Research topics in data management strategy include compatible data structures, filtration
of irrelevant and expired data, and prioritization function of safety sensitive data. First,
compatible data structures specify the method of organizing traffic-related data over
wireless networks. Since only one DSRC channel is available for vehicle—to-vehicle
communications, sharing information among applications is critical to reduce the waste
of communication resource by eliminating redundant transmissions of a same piece of
information. Second, filtration of irrelevant and expired data removes data that are either
unnecessary for an application, too old, or pertain to conditions that are or have become
too distant in space. Keeping information packages updated with only necessary
messages in a timely fashion may help saving bandwidth and reducing the accumulation
of delays. Last, prioritization of safety sensitive data ensures the delivery of safety
related information above other type of information. Such prioritization is required to

maintain the safety of the transportation network as a prime objective.

3.2.5. Data Fusion algorithms

Dailey et al. (1996) defined data fusion as the fusing of data from multiple sensors, such
as radar, infrared, sonar, and video camera. The original applications of data fusion were
mostly linked to military applications while civilian applications started from late 19805.
Since that time, the application areas of data fusion have stretched to a vast array of non-
military commercial and government applications in robotics, image processing, weather

72

surveillance and space exploration. In transportation area, a few documented studies
have been found related to data fusion. An example was for instance found with the
PRODYN real-time traffic signal control system, in which a Kalman filtering technique
was used to estimate traffic turning movement based on magnetic loop detector data
(Kessaci et al.., 1989). In a second application, video signals from two cameras were
used for automated vehicle guidance (Behringer et al., 1992). In a more recent vehicle
collision warning application, four types of data including radar tracking of a preceding
vehicle, yaw rate, GPS reading and digital map, and video input are used to predict the
future path of a subject vehicle (NHTSA, 2003). Other generic applications of data
fusions include algorithms to improve the accuracy and reliability of a GPS aircraft
navigation system, and algorithms to estimate traffic speeds and congestion levels from

loop detector data.

In this research, data fusion algorithms are used to identify traffic conditions, hazardous
situations or a need for a driving action based on consideration of various pieces of
information from surrounding vehicles that may have different latencies or degrees of
relevance. In particular, these algorithms must be able to Operate in a real-time
environment, where very little time is available to make appropriate evaluations and

decisions, and provide inputs to decision-making algorithms.

For the problem considered in this dissertation, data fusion can be viewed as the interface
between the data receiving module and the vehicle control logic, as illustrated in Figure
3.1. On one side of the data fusion block is information that has either been transmitted
from other vehicles or measured using on-board sensors. On the other side of the data

fusion interface, the fused data are fed to ITS functions providing decision-making and

73

automatic vehicle control assistances. In this case, the main role of the data fusion
interface is to combine various pieces of information into data that is meaningful for the
targeted ITS applications. The primary questions faced in this task are how to interpret
raw data, consider the various latencies due to transmission or sensor delays associated

with each piece of information, and infer meaningful and reliable output information

from the input data.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[ Wireless Network I
[Incoming Data] , , r ’ Communication Module] ‘ . ‘
, - — ' ' ’ I Data Fusion H Driver I—j “““““““
I’ ‘ S ‘
I
l I On-Board Sensors . Throttle, Brake j 3“
.‘ ,—-. Associated ,--~ I ‘
”‘--\“ o o o ’a‘--:‘ '
x ’3’, 11‘ Decmon-Makng {,0 ‘ {t l ,'
\ ” I'-‘\ \‘ I, I’-~\ \‘ \’
\\ l: ‘ ‘\‘ ,: I \ |\‘ ----------- I
iii- """ ‘ '. ,' ,' """"" r -------- -—--.‘ k ,' ,'
\ ‘ ~ — ’ I \ \ V O ’ I
\‘ ’I “ I,

 

 

Figure 3.1 Dissembled Vehicle Control Modules

Section 3.1 identified five areas of research that are deemed necessary for the integration
of various driver assistance and vehicle control systems based on vehicle-to-vehicle
communications. The knowledge required to fulfill these research needs spread across a
couple of fields. By appropriately choosing a subset of all the identified research needs,

this dissertation will focus more specifically on the following three problem areas:

a Coordinated decision-making algorithms
Impacts of wireless communications delays

Coexistence of vehicles with multiple levels of automation

74

Isolating parts of the problems serves in this case the primary purpose of limiting the
dissertation to a reasonable length while preserving a sufficient breath allowing
meaningful solutions to be developed for the problem at hand. In the following chapter, a

detail research plan will be presented to solve the three problem listed above.

75

CHAPTER 4

METHODOLOGY

Chapter 3 presented the main research problems associated with the development of
integrated wireless-based driver assistance systems. The problem of integration is one in
which various driver assistance applications are expected to operate simultaneously and
share with other neighboring vehicles or roadside equipment the limited computing
resources on-board a vehicle and limited bandwidth available for wireless data
transmissions. Specifically, it was determined that the current research should focus on

the development of a vehicle control decision making model that would:

Coordinate the control objectives typically associated with advanced cruise
control and collision avoidance systems,

Consider in its decision making process the problems potentially caused by
information delay due to sensor limitation or problems associated with wireless
transmission, and

u Be applicable to varying levels of vehicles equipped and non-equipped with the

advanced vehicle control functions.

This chapter presents the research methodology that will be employed for the
development and evaluation of a vehicle-based driver assistance control algorithm
considering the delays potentially associated with data obtained through wireless
communication. Specifically, the overall research is divided into four tasks. These tasks,
which are described in the first part of the chapter, include testbed development,

implementation of communication delay in simulation, decision-making model

76

development, and system evaluation. The second part of the chapter provides a more
detailed look at the evaluation criteria that will be employed in the investigation of

impact of communication delay and overall system performance.

4.1. Research Tasks

Four primary tasks were proposed for executing the research project:

a Task 1: Literature review and problem definition

a Task 2: Testbed development

Task 3: Implementation of communication delay in testbed simulation model

a Task 4: Development of decision-making model

Task 5: Evaluation of Impacts of Communication Delay and Mixed Vehicle Type

a Task 6: Conclusion and dissertation reporting

4.1.1. Literature Review and Problem Definition (Task 1)

The first task of the research project was to conduct an extensive literature review on the
state-of-the-art of wireless communication technology and the development of vehicle-
based driver assistance systems. The results of this task are outlined in the previous

chapters of this dissertation.

4.1.2. Testbed Development (Task 2)

The second task of the research project is to construct a simulation-based testbed
allowing the evaluation of vehicle control system performance. The need for the

development of such a testbed is based on the current limitations of existing microscopic

77

traffic simulation models. Current models do not provide adequate modeling capability
to simulate wireless communication networks, particularly the delays that may be
associated with data transmissions. To address this problem, a simulation model
integrating traffic simulation processes with wireless data communication functions will
be developed. This model will be based on the Intelligent Driver Model (IDM)
microscopic traffic simulation model that has been developed by Treiber et al. (2000,
2002). This model is selected based on the fact that it is an open-source model, i.e., that
its source code is freely available, and on the fact that using it as a reference structure for
the development of the testbed will thus eliminate a significant amount of programming

effort.

Starting from the original IDM model, revisions will be made to the model to tailor it to a
testbed core capable of performing microscopic traffic simulation routines along simple
single and multilane roadway elements. Expected changes to be made to the IDM

include:

Additional simulation capability of roadways consisting more than two lanes. In
the original IDM, the maximum number of lanes is two, which is insufficient to
evaluate vehicle control system performance on highways with more than two
lanes. Along with the capability to simulate additional traffic lanes, other relevant
changes would also be necessary, such as the expression of vehicle position, the
tracking of neighbors surrounding a subject vehicle, and lane—changing decision-

making.

Q

Introduction of new car-following models representing emerging driver assistance

systems.

78

er Introduction of accident detection capability to allow the evaluation of safety
aspects of emerging driver assistance systems, particularly the risks and
consequences associated with incorrect control decisions. This detection
capability will be based on evaluations of the respective changes in the position of
vehicles across simulation time intervals. An accident can be assumed to have
occurred if two vehicles are assessed to come too close together. Vehicles
involved in collision would remain indefinitely as obstacles to other upcoming
vehicles on the road, or be removed after a certain period of time.

o Introduction of ability to simulate extremely short time steps. Existing models
only allow simulations to be performed with time steps varying between 0.1 and
1.0 s (Gettman and Head, 2003). At such scale, a fixed simulation time step of
0.1 s would only allow the modeling of communication delays in increments of
0.1 s, or 100 ms. While such a short interval may be sufficient for many
applications, communication delays in transportation applications may frequently
vary in increments of less than 100 ms. To account for this, the simulation model
needs to be revised to allow the simulation to proceed at a rate specified by the

user and that can be defined in multiples of 0.01 s, or 10 ms.

5

Output capability of individual vehicle information over the entire course of
simulation. This feature is necessary to allow the information of all vehicles be
retrievable over the whole range of simulation. Resolution of data points over a

unit period of time will be decided by simulation users.

Besides the above modifications, the testbed will also be developed in such a way as to

allow future specific modeling needs to be easily implemented.

79

4.1.3. Implementation of Communication Delay (Task 3)

While the second task of the project is primarily concerned with the development of the
basic structure of the traffic simulation tool, this task focuses more specifically on the
implementation of appropriate mechanisms for simulating wireless communication
delays within the traffic simulator. The importance of wireless communication delay
stems from the fact that it causes vehicle control decision-making algorithm to use
inaccurate traffic information. By the moment a piece of data is received and ready to be
used, the vehicle that has originally disseminated the information may have already
changed its position, speed, or other parameters. The more delay there is in data
communication, the more likely it is that discrepancies may then exist between the actual
and reported status of a vehicle. While it is generally impossible to avoid wireless
communication delays, a certain level of delay can nevertheless be tolerated if it does not
cause significant impact the operation of vehicle control systems. The determination of
how wireless communication delay impact on vehicle control decisions will be obtained

through microscopic simulation experiments.

Similar to most existing microscopic simulation packages, the IDM model that is used as
a base for the simulation tool does not allow information be transmitted to vehicles
further than the immediate neighboring one, nor does it or account for communication
delays in the data transmission process. The capability to simulate both vehicle-to-
vehicle data transmission and communication delay will be added to the simulation tool
through a unique vehicle data storage structure. Figures 4.1 to 4.3 respectively illustrate

the initial data structure of the IDM model, a suggestion for an interim data structure, and

80

the three-dimensional data block that will be used to track communication delay in this

research.

Existing microscopic traffic simulation models typically update the status of simulated
vehicles sequentially, from the most downstream vehicles in the network to the most
upstream vehicles. At each time step, the vehicles that are reaching the extremities of the
network are the first to see their position, speed and acceleration level updated. Then,
car-following models are used to successively determine the speed and position of the
following vehicles. As shown in Figure 4.1, this results in an update process in which the
decisions made by a vehicle following another one are typically based on information
characterizing the status of the lead vehicle in the current time step. In this diagram, each
layer represents the set of data that is generated at each simulation step for all simulated
vehicles. Every time the simulation is moved forward by one time step, a new layer of
vehicle data is then constructed based uniquely on information contained from the

previous layer.

    

 

 

 

 

 

 

 

 

 

 

Time
1
t» 7":
i I ‘
1...:. _~‘{ " *' ____‘0_:___'_-_- ‘___I_'_7:lly_>____‘_AL‘ _____ LL...» "___‘l_,,"”’“""““t l
A --~----;~; , , / _/ pos . l
. . / , /
/ T x I.
w a/ «- “f— r'l- - f“
/ // .’/ I
l l f l 4 2
l ‘ ‘ l e
t-T so l E E z ;
l ,,,,,, , l ”M l s, !
tail Vehicle Platoon head

Figure 4.1 Data Structure of Traditional Simulation

81

An obvious problem is related to the usage of the latest information from the leading
vehicle. Since the information describes the vehicle’s status in the current time step, this
leads to the assumption that drivers or vehicle control systems react immediately to
changes in the status of a vehicle ahead. In reality, there is a certain interval, albeit
relatively short, between a status change and the moment a driver or a vehicle sensing
system becomes aware of the change. For a driver, this interval corresponds to the
perception time, while for a sensing system it may be associated to the sensor scan rate or
the time required to process the input data, whether or not communication delays exist.
This perception time is different from the reaction time, which is the time that elapses
between the moment a driver or a system has noted a change in traffic condition ahead
and the moment there is a change in the speed of the vehicle being driven. While the
majority of existing simulation software explicitly account for driver reaction time when
they update the status of individual vehicle, they typically neglect any perception delay
by assuming instantaneous perception of the slightest traffic changes. This results in
simulation models that may be intrinsically too responsive to changes in traffic
conditions. Such a problem can be fixed by changing the reference interval associated
with the lead vehicles to the previous simulation step. For most simulation models, this
means basing all control decisions on the vehicle status in the previous time step, as
illustrated in Figure 4.2. This data model makes it feasible to account for communication
delay, but apparently it is still limited to vehicles neighboring to each other in spatial

order.

82

   
    

 

----~* -‘--"- ‘ -' ‘ I: ~ “7 , _» ”1f" , A: .. '- j,““*'—7‘v~"llpos ,

,. t ' , a . sg; . g

, \ . , ,acc '5’
_' time g

’ l = ' ‘: " ' é % l

 

 

 

 

 

 

‘tan Vehicle Platoon head

Figure 4.2 Interim Improvement on the Data Structure of Traditional Simulation

To account for communication delays associated with cooperative systems, ITS
applications and individual vehicles in simulated transportation networks must be able to
retrieve information characterizing the status of multiple vehicles in the past. This
typically means allowing the models to go back in time and to retrieve the requested
information. To enable this feature, the data structure of Figure 4.2 is extended to include
not only information pertaining to the current and previous time steps, but also
information from time steps that are further back in time, as shown in Figure 4.3. In the
illustrated three-dimensional data structure, information about different vehicles is sorted
along the width of the data block, with the most downstream vehicle located at the right
most position. Vehicle data generated at different simulation steps are stacked vertically,
while the third dimension represents the various data that are being stored. The diagram
illustrates how communication delay would affect control decisions for the nth vehicle at
the current time step. First, the acceleration command for this vehicle would be based on
its status in the previous time step. If it is assumed that information describing the status
of the two vehicles immediately ahead of the vehicle under consideration is received

83

without significant delay, the control decision of vehicle n may then also be based on
information characterizing the status of vehicles n—1 and n-2 in the previous time step as
well. If it is then assumed that the information received from the 3rd and 4"h lead vehicles
carry a certain amount of delay, information from a previous time step will then be
passed to the function determining the acceleration command of vehicle n. Further
communication delays will in turn result in the consideration of data from time steps that

are further back in time.

 

 

 

 

manna-m

 

 

 

tail Vehicle Platoon head
Figure 4.3 Three-dimensional Data Structure for the Implementation of

Communication Delay

4.1.4. Development of Decision-making Model (Task 4)

Task 4 will pertain to the design of a coordinated vehicle control decision-making model
providing cooperative collision warning and cooperative adaptive cruise control
functionalities. Vehicles under control of the new decision-making algorithm should be
able to perform relatively close car following, compared to radar-based adaptive cruise
control, without sacrificing traffic safety. Current cooperative adaptive cruise control
models only consider the immediately preceding vehicle in their decision-making
functions, in addition to assuming constant wireless communication delays. To allow for
the modeling of a variety of control scenarios, the decision—making models that will be
developed with the ability to consider traffic information from multiple downstream
vehicles. The assumption of constant communication delay per individual transmission
link will not be removed, as this should the scope of a separate project. However,
varying communication delays will be indirectly accounted for in the design of the
vehicle control algorithms by considering scenarios in which information from lead
vehicles may hop though a number of vehicles, thus incurring transmission delays in

multiple instances.

In dealing with wireless communication delay, data fusion algorithms will be used to
interpret traffic information from wireless communication signals coming from different
vehicles. Comparing with the amount of information that drivers have nowadays access
to, wireless communication would be bringing in much more traffic data, so that it would
be almost impossible for human drivers to interpret every piece of the information, or
feed the incoming data directly to decision-making algorithm. A data fusion module will

be used in this case to remove duplicative and obsolete traffic data, and prioritize safety-

85

critical information. Another usefulness of data fusion is that it can serve as a short-term

memory to temporarily store traffic data for decision-making.

Another consideration of the decision-making algorithm is that it should be able to work
cooperatively with similar systems. The algorithm should be able to downgrade to
provide elementary functions, such as adaptive cruise control, to work independently
when no other cooperative systems are presented in vicinity. Transitions between these
different states should be seamless and smooth. For example, when automated driving
mode is switched from adaptive cruise control to cooperative active adaptive cruise
control, the longitudinal movement of a vehicle should be monitored without any

apparent gap between the two states.

4. I .5. Evaluation of Impacts of Communication Delay and Mixed Vehicle Type (Task 5)

This task will investigate the performance of the proposed vehicle control system in
multiple scenarios with varying wireless communication delays and multiple vehicle
types. Scenarios considering a 100% level of market penetration will be first used to
examine the operating characteristics of the vehicle control decision-making algorithms.
Scenarios with lower levels of market penetration, mixed vehicle types, and variable
communication delays will then be used to test the robustness of the vehicle control
algorithm in various situations. Depending on the results of the evaluations, a need may
arise to modify the control algorithms to fix identified problems, thus creating a potential

for Tasks 4 and 5 to be performed interactively.

It should be noted however that wireless communication delay is treated deterministically

in this study. Although wireless communication delays are typically stochastic in nature

86

and their probability distributions should be determined by considering at least the
influence factors described in Section 3.2.2., this research is focused more specifically on
traffic and vehicle performance analysis and treats communication delay as fixed input.
Future studies with specific modeling of wireless communication shall be integrated with
the simulation tool presented in this research and work interactively to more accurately

depict the picture of vehicle control with varying levels of information delay.

4.1.6. Conclusions and Dissertation Preparation (Task 6)

The final task will lead to the formulation of the research results and recommendations
for future study topics, as well as the final write-up of the dissertation. Expected

outcomes from the research include:

A microscopic traffic simulation tool for automated vehicle control applications;

In A prototype decisiommaking model coordinating multiple driver assistance
functions;

Understandings of potential impacts from wireless communication delay and
mixed vehicle types; and

a Research papers based on the contents of different research tasks.

4.2. System Evaluation

The evaluation of the proposed vehicle control algorithm will be based on the following

eight indicators:

to Acceleration profile,

Number of vehicles collided and collision velocity,

87

Time-to-collision (TIC),

Time exposed tirne-to-collision (TET),
Time integrated time-to-collision (TIT),
Cumulative threatening frequency (CT F),
a Cumulative jerk frequency (CJF), and

in Acceleration noise.

These indicators were selected based on their ability to provide quantitative measures of

the safety and stability of the vehicle control commands issued the proposed. algorithms.

4.2.1. Acceleration Profile

A vehicle acceleration profile records the acceleration and deceleration rates of a subject
vehicle over a given range of time. While the acceleration plot of a single vehicle may
only provide the control effort of the subject vehicle, a stack of acceleration profiles from
multiple vehicles arranged according to their spatial order provide an interesting way of
illustrating the interaction between a group of vehicles. Examples of such profiles are
shown in Figures 4.4 (a) and (b). The two illustrated diagrams plot the acceleration
commands outputted over time by a series of successive vehicles traveling in a single
lane and reacting to a sudden deceleration of the first vehicle. Both diagrams illustrate a
scenario in which the first vehicle collide with an object on the road, causing it to come to
an immediate stop and subsequently causing the five following vehicles to collide with
each other. The only differences between the two diagrams are with the vehicle control
system logic used. In this case, the staking of acceleration profiles clearly illustrates the

differences between the two control logics considered. On the left, the logic is able to

88

provide smooth deceleration commands, while the resulting control commands on the

right are marked by significant fluctuations.

   

Acceleration &:s\ , Acceleration

ll

l:
t., lllll

     

(a) (b)
Figure 4.4 Sample Acceleration Profiles

Plots similar to those of Figure 4.4 are particularly useful for analyzing how individual
vehicles react to the behavior of other vehicles ahead. In this research, a desired
acceleration profile would be one exhibiting no sudden accelerations or decelerations
other than those justified for safety reasons. Another sought characteristic would be a
profile with minimum oscillations in acceleration commands, as such oscillations would
reflect instability in the ability of the vehicle control algorithm to correctly adjust to
changes in traffic condition ahead or to compensate for information delay. Referring
back to Figure 4.4, this means acceleration profiles similar to the one shown on the right
should be avoided, while a profile similar to the one on the left should be sought to the

extent possible.
4. 2.2. Number of Vehicles Collided, Collision Velocity

The number of accidents that occurred along a given stretch of road provides a direct
assessment of the ability of a vehicle control system to prevent accidents, while the

collision velocity indicates the severity of an accident. Soundly designed systems should

89

be intrinsically collision-free. However, since it not possible to control all factors that
can potentially lead to crashes, an important feature of the vehicle control system being
developed should be an ability to both minimize the number of accidents and reduce the
severity of unavoidable accidents. In Figure 4.4, the number of collision can be
measured by the number of broken lines, i.e., lines that do not extend to the end of the

simulation.

The measurement of collision velocity is a bit more complicated. Based on the
identification of vehicles involved in a collision, the last speed record of the two vehicles
that have collided will have to be retrieved from the simulation output file. Collision
velocity can then be defined as the difference of vehicle speeds immediately before the
collision. Since the length of a simulation step will typically be very short, usually less
than 0.5 s, such estimation would be a reasonable approximation of the real collision

velocity.

4.2.3. Time-to-Collision

While the number of vehicles collided and collision velocity provides information about
accidents that have occurred, they do not fully assess collision risks. To help assess such
risks, Hydén (1996) defined the time-to-collision (TTC) as the time that remains until a
potential collision between two vehicles would occur if the vehicles’ current courses and
speed differences were maintained. Equation 4.1 indicates how this parameter is
calculated. In this case, a small TTC would be representative of an imminent collision,
either due to small vehicle spacing or a large speed difference. A desirable situation

would thus be one in which the TTC is as high as possible.

90

x _ - x — Ax
TTCW) = n m "m 0 . V Vn(t) - VII-IO) > 0 (4- 1)
Vn(t) ‘ Vn-1(t)

 

4.2.4. Time Exposed Time-to-Collision

To differentiate between various degrees of collision risks, the critical time-to collision
(TTC*) is defined as a threshold above which a vehicle may not be considered as facing

an imminent collision threat. It is worth noted that TTC* is not a border separating
accidental and non-accidental cases. As an artificial likelihood indicator of potential
collision, it does not affect the maneuver of vehicle control system, unless explicitly
involved in decision-making logic. The effectiveness of using TTC lies in that small TTC

values can always be observed for a period of time before accidents happen.

Typical choices for ITC* in studies on collision warning systems vary between 2 s and 4
s (Hirst and Graham, 1997; Minderhoud and Bovy, 2001). Such values are selected with
the objective to reduce the number of false warnings while maintaining a certain level of
efficiency. In this study, since there is no concern of false warnings, the TTC* is set at 5

s to ensure adequate consideration of all safety-critical events.

To provide an estimate of the cumulative risks faced by motorists over a certain period of
time, Minderhoud and Bovy (2001) define the time exposed time-to-collision (TET) as

the summation of all time periods in which the time-to-collision (TTC) parameter of a

given vehicle is below the critical time-to-collision (TTC* ). This parameter is defined

mathematically by Equations 4.2 and 4.3.

91

Tend

TET” = 26”“) -T (4.2)
z=o

5 0 else

"(0‘ 1 vost'rcn(,)sr'rc* (4'3)

4.2.5. Time Integrated Time-to-Collision

Although TET measures the duration a vehicle is under collision threat, it does not reflect
the severity of the potential hazards. From the parameter it is indeed impossible to
determine whether a vehicle avoided several mild traffic incidents with relative ease or
escaped with significant effort. To incorporate the severity of the potential threat,

Minderhoud and Bovy derived another indicator considering the summation of the
difference between TTC and TTC*. This indicator, labeled time integrated time-to-
collision (TIT), is defined by Equation 4.4, with an“) still subject to Equation 4.3. Since

potential hazards become more severe with lower TTC values, a large TIT would thus be

indicative of a collision threat for which significant efforts are required to avoid.

Tend {IT *
TIT" = 2 5”“) C — "Cm ] T (4.4)
0
It can be noted that both TET and TIT are measures of collision threats that are
aggregated over a period of time. As such, these two parameters allow the evaluation of
safety risks faced by individual vehicles in terms of both the severity and duration of the
risk. This may allow differentiating between high safety threats that only exist for a short

period and low safety threats that exist for a long period of time.

92

4.2.6. Cumulative Threatening F requeney

While both TET and TIT consider the number of intervals in which the TTC exceeds a
critical value, they do not provide any indication of how frequently collision threats
occur. This information can be obtained by determining the cumulative threatening

frequency ( C TF ), which counts the number of times that the TTC becomes lower than the

critical timeito-collision TTC*. This parameter is expressed mathematically by the

Equations 4.5 and 4.6.

Tend

CTFn = Zena) (4-5)
i=0
0 else

gnu) = * * (4.6)
l V OSTTCNU) STTC A TTC,1(I_T)>ITC

Well-designed control systems should result in low CTF values. In situations in which
vehicles face a single hazard, vehicles should experience the threatening of at most one
potential collision. CT F greater than 1 would thus indicate that vehicles experience
multiple dangerous situations, possible as a result of the inability of their control system

to appropriately respond to changes in traffic conditions ahead.

4. Z. 7. Cumulative Jerk Frequency

Another useful parameter is the cumulative jerk frequency (CJF). As defined by
Equations 4.7 and 4.8, this parameter counts the number of times a vehicle switches
between acceleration and deceleration. Well-designed vehicle control systems should

produce CJF values close to zero or reflective only of necessary speed changes, such as

93

when a vehicle that is approaching a queue that is finishing dissipating is able to start

accelerating before having to stop.

Tend

CJF” = 27,10) (4.7)
i=0
0 else

In“) _ l Vaccnm -accn(,_T) <0 (4'8)

4. 2. 8. Acceleration Noise

While CJF measures the number of times a vehicle switches between acceleration and
deceleration, the vehicle’s overall stability is reflected by acceleration noise (Jones and

Potts, 1962). The mathematical expression of this parameter is given by Equations 4.9

 

 

 

 

and 4.10.
T
1 e 1
an(avg) : an(t) ' d‘ : lvn(Te )- Vn(T’Sm 4 9

(Tend " Tstart ) Tstart (Tend "‘ Tstart) ( - )

T d

1 en

An = [an(t) _ an(avg )]2 'dt (4-10)

(Tend - Tstart) T
start

Acceleration noise is non-negative and the lowest possible acceleration noise is zero,
representing constant speed, constant acceleration, or deceleration rate over the whole
range of simulation time. When multiple vehicle control system configurations are
tested, systems with low acceleration noise generate smoother riding than systems with

high values.

94

CHAPTER 5

TESTBED DEVELOPMENT

To enable the evaluation of impacts that data transmission delays may have on decisions
taken by vehicle control systems, Tasks 2 and 3 of the research plan defined in Chapter 4
called for the development of a simulation testbed capable of simulating wireless
communication delays within a microscopic traffic simulation environment. In this
context, this chapter reports on the research activities that lead to the development the
Communication and Traffic Simulation Model (CATSIM), which will be used towards
the remainder of this project to help design vehicle control systems less sensitive to
vehicle communication delays and help evaluate the impacts of communication delays on

the operation of such systems.

This chapter first evaluates the wireless communication capabilities associated with
existing microscopic traffic simulation models and identifies the need for the
development of a customized simulation testbed to satisfy the needs of the present
research. The second part of this chapter introduces the open-source Intelligent Driver
Model (IDM) upon which the CATSIM model is based. The IDM is described within
this section in terms of its functionality, control flow, program structure, and required
functional changes. The following two sections present in more details the modeling
behind the CATSIM model and the results of some tests that were conducted to validate
the resulting model. The last section of the chapter finally presents a brief summary or

the research activities that were conducted in this portion of the research.

95

5.1. Need for New Simulation Testbed

Microscopic traffic simulation is widely applied in various areas of transportation
engineering. Examples of applications include evaluations of the effectiveness of
alternative traffic control strategies (Jeannotte et al., 2004), assessements of the ability of
simulation network to adequately service all existing and anticipated traffic, evaluations
of safety features of transportation networks, etc. . In all cases, an accurate representation
of driver behavior plays an important role in obtaining accurate results that are

representative of reality.

Recently, microscopic simulation models have also been applied to the development and
evaluation of vehicle-based driver assistance systems (VDAS). An example of
application can for example be found in Shaldover et al. (2001). Testing VDAS using
simulation saves time, labor and expenses. Comparing to test track experiments,
simulation is also superior from a safety standpoint as no material of personal injuries
may result of failed experiments. To date, the simulation of VDAS has primarily focused
on the implementation of new car-following logics and accurate tracking of vehicle data.
However, as will be detailed later, these experimentations have all been affected by
various limitations associated with existing simulation models in the modeling of

wireless communication systems.

Since the first research on microscopic traffic simulation models in the 19508, significant
attention has been drawn to the development of car-following models that would be
representative of reality, as evidenced by the numerous theoretical models described in

the literature (Pipes, 1953; Newell, 1961; Gipps, 1981; Leutzbach and Wiedemann, 1986;

96

Van Aerde and Rahka, 1995; Halati et al., 1997; Aycin and Benekohal, 1999; Treiber et
al., 2002). Some of these models have in turn been implemented as the kernels of
simulation packages that are now commercially available, such as AIMSUNZ, VISSIM,
CORSIM and INTEGRATION (Jeannotte et al., 2004). Rather than focusing on the
soundness of individual car—following models, this chapter focuses primarily on the
compatibility of the car-following models found in existing simulation software with
respect to the modeling of wireless communications in a transportation environment.
Specifically, information is presented to reflect modeling choices made by these models
regarding simulation time steps, vehicle data update schemes, and handling of driver

perception-reaction delay.

The most popular models for the evaluation of emerging transportation applications are
AIMSUN, VISSIM and PARAMICS. These models all feature an application
programming interface that allows the addition of functionalities enabling data retrieval
from the simulation and interactions with various driver behavior processes. Despite
these features, the modeling of vehicle-to-vehicle wireless communications remains a

difficult task within each model.

97

 

 

 

 

 

 

 

 

 

 

ELK H 1 W
lit-vehicle data storage LH ITS appllcatlons
and modeling

Vehicle-to-vehlcle wireless
Network communication

 

 

 

 

 

 

 

 

simulator
t Driver response and
Data extraction vehicle cogtrol
commands comman s

 

 

 

 

I I

Trafl'lc simulator

 

 

 

 

Figure 5.1 Wireless-based Application Modeling with Existing Simulation Models

First, since each model restricts access to its simulation code, wireless communication
functionalities must generally be developed in parallel to the simulation. As illustrated in
Figure 5.1, this means first developing functions to extract the information that need to be
wirelessly transmitted from the simulation. Functions are then needed to emulate the
queuing and routing processes of communication networks. Since the information
received by individual vehicles cannot typically be stored within the traffic simulation
environment, a database tracking the information received and stored by each vehicle is
then needed. Following the development of appropriate interfaces between the database
and target vehicle control applications, functions would finally be needed to return
vehicle control commands and traffic signal control actions to the simulation

environment.

A second problem is linked to the ability of simulating the magnitude of expected

communication delays. Since simulating communication delays typically creates a need

98

to retrieve data from previous time frames, simulation models must have the capability to
advance their clock at increments corresponding to the level of expected delays. As was
demonstrated in Biswas et al. (2005), such delays can be as short as a few milliseconds.
The shortest time step that can be modeled in existing models is 0.1 s, or 100 ms. This
may already be too long for adequately simulating information delays associated with
many applications. Furthermore, since existing simulation models do not typically allow
data retrievals at intervals of less than 1 5, this reduces even more their ability to

appropriately simulate the impacts of information delays on vehicle control applications.

The use of car-following models that are intrinsically collision—free constitutes a third
limitation. In existing models, accidents can typically only be staged by putting invisible
static vehicles on the roadway. With automated vehicle control systems, however,
accidents may result from the selection of inadequate vehicle control parameters,
particularly when considering that vehicles have finite acceleration and deceleration
capabilities. In this case, the ability to simulate accidents would be particularly useful in
helping identify appropriate vehicle control parameters and developing more robust

control algorithms.

5.2. Testbed Development Strategy

The various limitations descried in the previous sections have led to the need to develop a
new simulation model specifically designed to facilitate the evaluation of ITS
applications relying on sensor data and wireless communications. To facilitate the
development of the proposed testbed, the existing Intelligent Driver Model (IDM) was

selected as a starting point for its development based on the fact that the IDM is an open-

99

source model, i.e., a model for which source code is freely available. Using the IDM as a
reference structure for the development of the CATSIM model eliminates in this case a
significant amount of programming effort as development can start from an existing
functional simulation model. Starting from the original IDM model, revisions could for
instance simply be made to it to tailor it to the specific modeling needs being considered.
Details of the original IDM relevant to the current project are described in the following
section, with subsequent sections presenting the main modifications that have resulted

into the CATSIM model.
5.3. Open-Source Intelligent Driver Model

The IDM car-following model and its accompanying traffic simulation package were
developed by researchers at the Dresden University of Technology in Germany. In this
dissertation, the simulation model is referred to as the “IDM model”, and the
corresponding car-following model is referred to as the “IDM car-following model.” The
IDM model is available to the general public at the following website: www.traffic-
simulation.de. The software used in this research was retrieved from the above site in
August 2005. As of August 2007, it was believed that the coded used in this research still

represented the latest available version of IDM.
5.3.1. General Description

The IDM model allows the simulation of two-lane traffic with two types of vehicles,
passenger cars and trucks. A screenshot of the IDM model is shown in Figure 5.2. The
U-shaped freeway segment shown in figure does not represent an actual curved freeway

section but is simply a display adopted by the model’s developers to allow more vehicles

100

 

 

 

 

 

 

 

 

 

 

 

 

Time 108 ' i I I ' I I I I I - I
7 ’ I I I - - - I
1‘ .
O
I
_ O
l
I Mainlnflow fl._-_-_--._._.M._-,_,_- J::.___ D] 3300Vehicleslh
- Ramplnflow if““TIMI_“WTW-"TmWM_m2"; bl 400 Vehiclesth
Ramp p_Factor all _DJ 0.0
I DeltabRamp _ch _r] -2.0 rnts‘2
I TimeWarp Factor 1] _f _b] 6.9 times
I
0
1a
Q
' 4 Car .-
‘ Tl’UCk -
~~ I - - - I I I I I I I I ' I
h I I I I I I I - -
I I

 

Figure 5.2 IDM Model Screenshot

Source: (Treiber, 2005).

to be shown on a single computer screen without excessive zooming out and shrinking of
the size of individual vehicles. As can be seen by the menu in the middle of the figure,
configurable simulation parameters include mainline flow rate, on—ramp flow rate,
vehicle merging courtesy factor, ramp vehicle merging factor, and the ratio of simulation
time to real time. Although only five scalable factors are shown, other parameters
involved in the simulation can also be changed by directly adjusting parameters within
the computer code and recompiling the simulation program. The major hidden

parameters that can be adjusted through this process include:

Simulation case
Time step

Truck percentage

101

a Road length
a Vehicle length

Minimum spacing

The default running environment for the IDM model is a Java Applet. This environment
means that the program is always contained by a webpage or an Applet Viewer. While
this format is more advantageous than a standalone application for website publication,
the inherent security properties of Applets prevent the model from generating output files
that can be saved on a computer hard drive. This means that the default IDM model
cannot directly output simulation results in data files. To save data from a simulation
output on computer disk for further evaluation and analysis requires a customized

development to convert the IDM model into a standalone application.

5. 3.2. IDM Control Flow

Control flow within the IDM model is demonstrated by the flowchart shown in Figure
5.3. The figure illustrates the sequence of modules called within each simulation cycle.
The simulation starts with default control parameters as soon as the webpage shown in

Figure 5.2 is loaded.

A simulation run always starts with an empty roadway. Entering vehicles are created by
the Vehicle Generation module. Within this module, the IDM compares the number of
vehicles passing the entrance point since the start of simulation against the product of the
flow rate and simulation time. New vehicles are generated at a uniform rate
corresponding to flow rate. A newly generated vehicle is only shown on the screen when

it enters the simulated roadway section.

102

 

—D-I Vehicle Generation I

 

 

 

Vehicle Update
a Vehicle Entering
a Arc/Deceleration
0 Transition
n Lane-Changing
a Exiting

No@

Yes
I Exit Simulation I

 

 

 

 

 

 

Figure 5.3 IDM Model Control Flow

The next module called upon in a simulation cycle is Vehicle Update. This module
controls vehicle entry into the network, acceleration/deceleration decision making,
vehicle speed and position updates, lane changing, and network exiting. Updating of
vehicles in IDM follows a traditional downstream-to-upstream order, in which the most
downstream vehicle is first updated, followed by the next upstream vehicle until the most
upstream vehicle is updated. Although a typical IDM simulation run allows vehicles to
travel in either of two drive lanes, the data of vehicles in both lanes are stored in

computer memory as a single sequence.

After application of the car-following and lane-changing rules on all vehicles, the Vehicle
Update module checks for vehicles that may have reached the end of the simulated
roadway network. Vehicles that are determined to have reached the end of the network
are removed from the simulation, allowing for a release of the corresponding space used

by the vehicle in computer memory.

103

5.3.3. IDM Program Structure

While the flow chart of Figure 5.3 demonstrates the sequence of operations used to
simulate traffic flow behavior, the computer code implementing the IDM is much more
complicated than what the simulation control flow chart suggests. Functions such as data
storage, vehicle visualization, and simulation control interface are among the major
elements that add to the complexity of the program. Since a description of all the
program elements would be too exhaustive for the purpose of this dissertation, only the
elements that are critical to the traffic modeling and the manipulation of vehicle data are
introduced to help the reader understand the complexity of the task of modifying the IDM

program to create the CATSIM model.

The IDM was developed as a Java object-oriented computer program. Object—oriented
programs are built around the notion that data structures and data processes can be
organized in a structure of objects of specific characteristics. Each object within the
structure then contains specific variables and functions allowing other objects to access or
modify these variables. Objects with similar characteristics are then organized in classes.
For instance, car and trucks could be viewed as two distinct objects. However, both can
easily be seen belonging to a more generic class of object regrouping all types of

vehicles.

The development of object-oriented programs can be generally interpreted as a process of
defining suitable classes of objects and determining how objects of each classes can
communicate with other objects. One of the principal advantages of object-oriented

programming techniques over procedural programming techniques is that they enable

104

programmers to create modules that do not need to be changed when a new type of object
is added. A programmer can simply create a new object that inherits many of its features

from existing objects. This makes object-oriented programs easier to modify

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. _ E i
M1croSrm2_0 E MicroStreet
i i +
I I Car
SimC anvas I I
: V
l E <<abstract>>
l ............ l IDM
IDMCar IDMTruck
Legend
A "-D' B Class B is a member of Class A
C ‘—> D Class C realizes abstract Class D

Figure 5.4 IDM Model Program Structure

Figure 5.4 illustrates the general structure of the IDM computer program. Each box in
the figure represents a major class of objects. Objects in each class are assigned specific

roles in the execution of a simulation, as detailed below:

MicroSim2__0. This is the highest class of objects in the model. Objects defined
within this class implement simulation control elements such as the traffic flow
scrollbars shown earlier in Figure 5.2.

o SimCanvas. Class containing objects required to visualize roadway sections and

simulated vehicles on the computer screen.

105

MicroStreet. Class of objects implementing the data structure required to
simulate vehicle movements along a roadway segment. For instance, objects in
this class are used to stores vehicle data in computer memory, access relevant
information from each vehicle, or sort vehicles according to their spatial order
along the simulated roadway.

Car. This class of objects represents the vehicles being simulated. Each object in
this class defines generic vehicle variables. The variables defined with each
object include both static data, such as vehicle length, width and color, and
variable data, such as position, speed, acceleration and lane position. Each object
also contains information defining the car-following and lane—changing models to
be used by each vehicle.

IDM. This is an abstract class of vehicle type. The usefulness of an abstract class
is to provide a template for the creation of other classes of objects. In this case,
the primary functions of the abstract class are to define car-following parameters
and rules for all the car-following models that may be employed by the IDM.
Since this is an abstract class, no values are given to the parameters being defined.
Values for each parameter are left to be defined within the specific classes created
using the template.

IDMCar/ IDMTruck. These are extended classes of the IDM virtual class.
Objects within these classes are identical to object of IDM class, except that
parameters that were left without values in the IDM are now given specific values
related to the type of vehicles that each object represents. In this case, two

specific objects are created, one implementing a car following model for

106

passenger car (IDMCar) and one implementing a carp-following model for trucks

(IDMTruck).

5.4. CATSIM Model

The main functional changes that have been implemented. to transform the IDM model

into the CATSIM model include:

Introduction of the ability to simulate roadway segments with more than two lanes
per direction. While the original IDM only simulates roadway segments with two
traffic lanes, this was judged insufficient to evaluate vehicle control system
performance in a range of typical situations. To reflect the fact that many urban
freeways have nowadays more than two lanes per direction, the computer coded
was modified to allow the simulation of a user-defined number of lanes. Besides
changing graphical display, this required modifications in the expression of
vehicle position, the tracking of neighboring vehicles surrounding a subject
vehicle, and the decision-making related to lane-changing behavior.

Addition of a new type of car-following models that can be used to model
situations in which vehicle control decisions are made by an automated systems
instead of a human driver.

Relaxation of the no accident constraint that is generally associated with car-
following models in microscopic traffic simulation models. As explained earlier,
maintaining this constraint would have reduced the ability of the model being
developed to evaluate the correct operation of proposed car-following control

algorithms, particularly in their ability to safely control vehicles.

107

Introduction of accident detection capability. Functions were added to evaluate
the positions of two vehicles following each other in the same lane and determine
whether these vehicles may have collided. Collisions are assumed to have
occurred if the spacing between the two vehicles is smaller than a certain
threshold. Vehicles involved in a collision would then remain indefinitely as
obstacles to other upcoming vehicles on the road, or be removed after a certain
period of time.

Introduction of capability to consider extremely short and variable simulation
time steps. As indicated earlier, existing models only allow simulations to be
performed with time steps varying between 0.1 and 1.0 s (Gettman and Head,
2003). At such scale, a fixed simulation time step of 0.1 5 would only allow the
modeling of communication delays in increments of 0.1 s, or 100 ms. While such
a short interval may be sufficient for many applications, communication delays in
transportation applications may frequently vary in increments of less than 100 ms.
To account for this, the simulation model was revised to allow the simulation to
proceed at a constant rate that can be specified by the user in multiples of 0.01 s,
or 10 ms.

Output capability of individual vehicle information over the entire course of
simulation. This feature allows the information of all vehicles be retrievable over
an entire simulation. In this case, the resolution of data points, or interval at
which information is being stored in memory, is left to be decided by users of the

simulation model.

108

A typical screenshot of the CATSIM model is shown in Figure 5.5. Major modifications
visible from the screenshot with respect to the IDM snapshot of Figure 5.2 include the
addition of buttons for simulation progress control, the addition of vehicle taillight
showing the braking magnitude of each vehicle, and the visualization of immobilized
vehicles. The following subsection presents in more details the model’s control flow,
program structure, and the method by which wireless communication delay is

implemented within the program.

 

_ ,7 (I 1,, 3, _ __, e,t.;:.g ‘;, 3.,1g...,- ,,j _, . ,p .
tawha ,.,d . J fi,fit. , ,w_wdrwgfi_uw,, . . .JQIX
' ‘ ‘ '_ . ->4~ whpn-‘kd‘J 1,5, . « g . . i . .

 

 

Time 0:00:51.200

a" “l - .. I ‘7' i U ' H ‘
Status: 40 vehicles are being suul eted. Aba't :I Resume
Wile -I EXI

”WV.-. . .r_ n- -.---._' L- ...____— ~.— -- . -w .__—__.-— _ J

 

Figure 5.5 CATSIM Screenshot

5.4-1. CATSIM Control Flow

The CATSIM control flow is illustrated in Figure 5.6. In the figure, all bold and
underlined elements represent modules or functions that were added to the original IDM
control flow model illustrated in Figure 5.3. Within the traffic simulator, the three main
additions are the Real-Time Data Collection and Storage module, the Randomly
Occurring Events Generation module, and a series of simulation progress controls. These
modules, along with the added incident detection functions in the Vehicle Update

module, have transformed the original IDM model into the traffic simulator core of the

109

CATSIM testbed. Other modules at the periphery of the traffic simulation model include
modules allowing the inputs of car-following data, vehicle type and traffic events, as well
as post-simulation processing modules dealing with output configuration and result
analyses. While the objective of Figure 5.6 is to present the functional modules of the
CATSIM model, these functions were implemented by developing computer program

with reference to the specific elements of the original IDM program structure.

5.4.2. CATSIM Program Structure

As illustrated in Figure 5.7, the CATSIM object-oriented program structure is similar to
the basic IDM program structure, except that modifications have been made to rename
certain modules to reflect class function changes and that additional classes have also
been implemented in the program structure. The following elements provide a definition
of the CATSIM main object classes, as well as a brief explanation of the relationships

between these classes:

a SimulationBatch. An object of this class corresponds to an object of the
MicroSim2_0 class in the original IDM model. The main difference with an
MicroSim2_0 object is that objects belonging to the SimulationBatch class have
the capability to execute a batch of simulation runs. Not only batch runs can be
made for various random numbers, changes can also be made in other parameters,

such adjusting flow rate between succeeding runs.

110

 

 

 

 

 

 

 

Car-Following Vehicle Type
Model Editor Editor

 

 

 

 

Traffic &
Incidents_ Info

 

 

I

——-n—-...-l---.-----v--o----r-.----.-—---------.-u--. ---——------——-—--—

 

 

 

 

 

 

Randomly Occurring
Events Generation

(i.e., Incidents!

I

 

 

 

 

 

fil Vehicle Generation I

 

I

 

 

 

Vehicle Update

a Incident Detection
a Vehicle Entering

a Acheceleration

in Transition

0 IzanesChanging:

a Exiting

 

 

 

I

 

 

Real-Time Data

Collection & Storage

 

 

 

l-------_---_---__---_-------_ -----__-.-—---.----.---_-..-—-.---_..-

 

 

 

I Output Configuration I

 

 

I Result Analyses I

.u-.....--.-

-----—-——

l
l
I

Test Bed

 

Figure 5.6 CATSIM Control Flow

111

 

 

 

 

 

 

 

 

 

.1
I
l
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
l
I
I
I
I
I
I
I
l
I
I

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Jr ------------ . .
Simulation I II b i 0
Batch I I : VehicleFleet fl- Vehicch ataBuffer ‘— VehicleDataOutput
a is: i
t I I : Vehicle
Dynamic l : I ;
Simulation I I I i
Ob'ects l i: v
J, l : I <<abstract>> __________ VehicleInformation
It:::::::::::'.l I DrivingLogic Buffer
<<abstract>> <<abstract>>
ModelIDM ModelGipps
CarlDM TruckIDM CarGipps ruckGipps
Legend
A --->B Class B is a member of Class A
C —DD Class C realizes abstract Class D
E —>F Class E has access to member of Class F

Figure 5.7 CATSIM Program Structure

o DynamicSimulationObjects.

Objects of this class serve as containers for all

traffic-related objects. The major functions defined within this class include the

management of simulation progress, the control of how vehicles enter and exit a

simulated roadway segment, and coordination with objects from lower level

classes. An object of DynamicSimulationObjects class is created each time a new

simulation is started. When batch runs are used, an object is then created for each

run within the batch.

VehicleFleet. Object of this class are time stamped and hold all vehicles objects

created during a simulation. Compared with the similar MicroStreet class in the

original IDM, objects of the VehicleFleet class are provided with functions

112

 

allowing them to detect collisions between succeeding vehicles and to sort
vehicles on roadway sections with more than two lanes.

Vehicle. This is a sub-level class of the VehicleFleet class. Although this class is
at a similar level as the Car class of the original IDM, the most significant
difference between the two classes is that each vehicle in CATSIM has a unique
identification number, while vehicles in IDM are not numbered and can only be
accessed once during a vehicle status update.

VehicleDataBuffer. This class of objects is used to temporarily archive
VehicleFleet data. It corresponds to the 3-D data structure proposed in Chapter 4.
At the end of each simulation cycle, a copy of the latest instance of VehicleFleet
is recorded in VehicleDataBuffer. Data collected in VehicleDataBuffer are used
for two purposes. First, they are used for keeping track of wireless
communication delay in the simulation model. Second, data in the buffer are
processed and ported to VehicleDataOutput for permanent storage. The size' of
VehicleDataBuffer and the frequency of data processing in the buffer and of
buffer cleaning for new data are user-adjusted depending on computer hardware
capability.

VehicleDataOutput. Objects of this class are used to store data of individual
vehicles when written data outputs are requested. As the simulation time step in
CATSIM car be set extremely short to accurately model wireless communication
delay, large amounts of vehicular data can be expected. However, traffic data
with extremely fine resolution is unnecessary for the interpretation of traffic

characteristics. Therefore, traffic data are usually sealed with a user-adjustable

113

scaling factor to reduce its resolution when transferred from VehicleDataBuffer to
VehicleDataOutput.

DrivingLogic. This class of objects is the counterpart of the IDM class in the
original IDM model. Both the DrivingLogic and IDM classes define abstract
objected. By extension, the extended classes based on the DrivingLogic class are
also abstract. As can be seen in the bottom two layers of Figure 5.7, the first
extension define different families of car-following models, such as IDM car-
following model or Gipps model. The second extension then generates vehicle
types within each family of car-following models.

VehicleInformationBuffer. This class of objects has been designed to model the
transmission of traffic data via wireless communication. Objects within this class
implement data search capabilities in VehicleDataBuffer. For instance, objects in
this class can be used to retrieve past data from specific vehicles corresponding to
a certain level of communication delay. The detailed mechanism of collecting

traffic information for past time steps is provided in the following section.

5.4.3. Implementation of Wireless Communication Delay

The modeling of wireless communication delay in computer programs mainly considers

the following factors:

D

Number of neighboring lanes, where vehicle data need to be collected,
Number of downstream vehicles that are considered by the subject vehicle,
Presence of preceding vehicle, and

Maximum communication range of the subject vehicle.

114

The modeling of wireless communication delay in the CATSIM model is illustrated using
the example of Figure 5.8. The numbering scheme shown in the figure is only for
information collection and is independent of the vehicle identification numbering
scheme. In this case, the subject vehicle is the one identified with the [0, O] coordinates.
The first coordinate represents lane position. The lane on which the subject vehicle is
located is always numbered as 0, while lanes to the left are numbered as -l, -2, -3, ...,
and lanes to the right are correspondingly numbered as 1, 2, 3, and so on. The second
coordinate denotes the relative position of a neighboring vehicle to the subject vehicle.

In this case, vehicles downstream the subject vehicle are positively numbered as 1, 2, 3,

while vehicles upstream are numbered as -l, -2, -3,

 

Figure 5.8 CATSIM Vehicle Numbering Scheme

When wireless communication delay is considered, for example as shown in Figure 5.8,
the subject vehicle [0, 0] would receive delayed traffic data from preceding vehicles.
When the wireless communication between the subject vehicle and vehicle [0, 3] is
modeled, the simulation program starts from vehicle [0, O] and traces the communication
back to vehicle [0, 3] by iterating the second coordinate from O to 3. Once vehicle [0, 3]
is identified in the simulation program, its vehicle identification number can be obtained
and used to retrieve past traffic information within a data structure similar to the one

shown in Figure 4.3.

115

5.5. Testbed Validation

The CATSIM model was validated by examining the speed—flow diagrams produced by
the model when using various car-following models. Attention was more particularly put
on the diagrams produced when using traditional car-following models reflecting human
driver behavior as these have been the objects of many validation efforts. An example of
a speed-flow diagram resulting from the use of the Gipps car-following model (Gipps,
1981) is shown in Figure 5.9. This diagram was produced for a scenario considering
traffic traveling along a one-mile circular highway. In each case, traffic was simulated
until traffic equilibrium was achieved, i.e., until all vehicles had speed falling within a
narrow range. Since a circular roadway was used, this created a more difficult situation
to achieve stability as shockwaves are continuously propagating along the roadway.
Multiple runs were executed with different flow levels to achieve different levels of
traffic density. Reported results do not include data from the first 80 s of each run to
ensure that traffic instability in the initial portion of each simulation did not affect the
validation. As can be observed, the resulting speed-flow diagram resembles the well-
known shape of the relationship, consisting here of a parabolic bottom and an almost flat

top portion.

116

 

 

 

 

80

70_ o o o o o o o o.
A 60~

o

50..
V o
'8 401 .
8.
CD 30-4 0

O
20« , ’
O
10 I I l I 1
8(1) 9(1) 10(1) 11(1) 1200 1300 14(1)

Flow Rate (veh/h)

Figure 5.9 Simulated Speed-Flow Diagram based on GIPPS Model

5.6. Summary

The objective of this chapter was to present the development of CATSIM testbed, which
has been constructed to research automated vehicle control systems. Prior to the
presentation of the CATSIM model, existing microscopic simulation packages were
reviewed for their capabilities in realistically simulating wireless communication delay
and how these delays may affect the operation of automated vehicle control systems. It
was determined that limitations exist in various aspects, including simulation time step,
collision detection, data storage, and source code accessibility, leading to a need to

develop a specialized traffic simulation testbed that was subsequently named CATSIM.

The development of the CATSIM model is based on the open-source IDM model, which
source code was freely available. The IDM model was first reviewed for its
functionality, control flow and program structure. While the control flow and program
structure were both referenced by the CATSIM model, additional functionalities required

for the objectives of this research were identified in five areas:

117

Additional simulation capability of roadways consisting more than two lanes,

New car-following models that will represent driver assistance systems,

Accident detection capability,

Extremely short and variable simulation time step, and

o Output capability of individual vehicle information over the entire course of

simulation.

Following the identification of functional needs, the CATSIM model was presented,
using a similar layout of sections of for the description of the IDM, to highlight the
modifications that have been made in simulation control flow and computer program
structure. The other significant addition to the CATSIM model is the implementation of

functions allowing the simulation of wireless communication delay.

The developed CATSIM model was lastly validated by examining the generated speed-
flow diagram when applied with Gipps car-following model. Multiple simulation runs
were executed and the CATSIM model was found to produce outputs generally

resembling the well-known shape of the theoretical speed-flow curve.

118

CHAPTER 6

LONGITUDINAL VEHICLE CONTROL ALGORITHM

Chapter 5 described the CATSIM testbed that has been developed to allow the evaluation
of automated vehicle control involving wireless communication delay. Following the
description of the CATSIM simulation model, this chapter presents a longitudinal vehicle
control algorithm that has been developed for operation in an environment in which the
information used for acceleration/deceleration decisions may be delayed as a results of
delays in wireless communications. The vehicle control algorithm presented in this
chapter is based on sliding-mode control theory. The new algorithm is similar to the
existing vehicle control systems that have been reviewed in Chapter 2 in that it is
restricted to only provide longitudinal vehicle control. The exclusion of latitudinal
vehicle control is mainly due to the fact that the automation of both longitudinal and
latitudinal vehicle control would significantly increase the complexity of vehicle control
systems. However, it should be noted that the CATSIM model already provides the
capability of simulating both car-following and lane choice decisions, which may allow

its use in subsequent research on more complex vehicle control systems.

The primary advantage of the new longitudinal vehicle control algorithm is that it takes
into consideration traffic information transmitted from multiple downstream vehicles for
acceleration and speed control decisions. Based on the review of conventional vehicle
control systems outlined in Chapter 2, the first part of this chapter identifies the
functional needs for wireless-based driver assistance systems. The second part of this
chapter describes in detail the development of a sliding mode-based vehicle control

model that previews only the immediate preceding vehicle. The following section then

119

develops an expanded vehicle control algorithm that has the capability of monitoring the
status of multiple lead vehicles. The last part of the chapter finally presents a series of
filters that provide complimentary refinements for the output of multiple-vehicle

following algorithm.

6.1. Longitudinal Vehicle Control Functional Needs

Adaptive cruise control systems are currently the most advanced type of vehicle speed
control systems available for passenger vehicles in the automotive market. As indicated
in Chapter 2, range-sensing equipment are now mounted on the front of vehicles to
monitor the distance with vehicles located within a certain range ahead and to adjust the
speed of the host vehicle in such a way as to maintain a safe following distance. In most
cases, the vehicle control decisions are typically based solely on the detected speed
differential and following distance with respect to the vehicle immediately ahead. A
review of various studies that have been conducted to evaluate the performance of
adaptive cruise control systems has lead to the conclusion that this type of system, while
objected to improve motorist safety and driving comfort, is also not problem-free.
Particular problems include excessive large headway, limitation on roadway sections
with sharp curves, and limited monitoring range in terms of both spatial distance and

number of lead vehicles.

Wireless communications are adopted in this research to remedy the problems associated
with radar-based vehicle control systems. Specifically, wireless—based vehicle control
algorithms are expected to utilize traffic information from multiple downstream vehicles

in a way that automated vehicle control would become more proactive than adaptive

120

cruise control. For instance, it may become possible for a vehicle control system to issue
deceleration commands in response to the observed braking maneuvers of the second or

third vehicle ahead before the vehicle immediately ahead may start to brake.

To achieve the desired control improvements over existing adaptive cruise control
systems, the following guidelines have been identified for the development of the

proposed wireless—based longitudinal vehicle control model:

to Integration of traffic information from multiple lead vehicles. Adaptive cruise
control systems typically cannot collect information from the vehicles that are
traveling ahead of the vehicle immediately in front. In contrast, human drivers
can often anticipate changes in traffic by observing the actions taken by drivers
ahead of the vehicle immediately in front of them. Traffic information from
multiple lead vehicles provides a potential to mimic the anticipative behavior of
human driver and to further enhance traffic safety by reducing human perception-
reaction time.

o Integration of vehicle acceleration data. Currently, vehicle acceleration can
only be measured by installing accelerometers onboard vehicles as radar sensors
cannot reliably detect the acceleration behavior of other vehicles. By using
wireless communications, the acceleration data of individual vehicles could now
be broadcasted to all surrounding vehicles that are capable of receiving wireless
messages. In this case, the proposed vehicle control algorithm should be able to
utilize acceleration information received from downstream lead vehicles to enable

the host vehicle to be driver in a safer and smoother fashion.

121

Development of collision avoidance capability. While adaptive cruise control
systems are able to slow down a host vehicle, its speed regulation is only realized
by controlling engine throttle instead of active braking. Therefore, even if users of
adaptive cruise control systems are partially released from vehicle speed control,
they are still responsible to identify safety threatening events, such as a sudden
braking of the vehicle in front, and to apply brakes to avoid collision. The vehicle
control algorithm considered in this research is objected to be able to initiate
active braking and avoid potential collision.

in Scalability between single and multiple vehicle following. Market penetration
rate has been a concern for the envisioned deployment of most automated vehicle
control systems. As the developed vehicle control system is likely to operate
initially in traffic streams in which most surrounding vehicles remain under
manual control, traffic information from multiple lead vehicles may therefore not
always be available. In the case in which no communication link could be
established, the vehicle control logic should be able to scale back to a simpler

logic that follows only the vehicle in immediate front.

6.2. Sliding-Mode Controller for Single-Lead Scenario

The vehicle control algorithm developed in this research is based on the principles of
sliding mode control. Sliding mode control is a type of variable control structure with
feedback in which high-speed switching control commands are used to adjust the
dynamics of a nonlinear system to approach a desired status and then stay close to this
status. The advantage of sliding mode control over traditional closed-loop control is the

independence of system performance from system modeling accuracy. This allows

122

mitigating the effects of potential mismatches between a mathematical model used to
describe a particular system and the system’s reality. In this case, the development of a
vehicle control logic based on sliding mode control avoids more specifically the difficulty
introduced by the potential need to consider variable wireless communication delays.
After exploring a series of other model types, such as linear car-following model, Gipps
model, and the IDM model, the sliding-mode approach is recognized as a promising one

to realize stable vehicle control in the presence of information delay.

To illustrate how sliding mode control principles were used to develop the vehicle control
logic, first consider a situation in which a vehicle is following another one on a given
roadway. For this situation, the spacing between the rear bumper of the lead vehicle and

the front bumper of the following vehicle at time t is defined by:

Ann—mo) = xn-io) - xn(t) "ln—i’ (6.1)
where n is the index of the vehicle for which vehicle control is being determined, and i

denotes the ith vehicle ahead of vehicle n. When i is equal to 1, Equation 6.1 simply

calculates the distance between a pair of succeeding vehicles.

When constant time headway is adopted as a following rule, vehicles try to maintain a
separation distance with the preceding vehicle that generally increases with speed. In its

simplest form, the desired separation distance under such a rule is calculated as:

Milne) = hill“ Vnm’ (6.2)

If information about the current acceleration level of the following vehicle is known,

Equation 6.2 can be expanded into Equation 6.3 to account for the tendency of this

123

vehicle to either increase or decrease its following distance from one time interval to the

I16 XI.

d d
muffin“) = hnes Vn(t) +faan(t)’ (6.3)

In Equation 6.3, the second term on the right increases the desired following distance
when the following vehicle is accelerating and decreases it when the vehicle is braking.
The factor fa is used to convert the observed acceleration at time t into a following
distance adjustment. This factor can take various forms, depending on the importance

put on the effect of vehicle acceleration and on the model being considered.

At any given time, the error in the following distance between two vehicles can be
defined as the difference between the actual and desired distances between the two
vehicles. Using the definition of desired following distance provided by Equation 6.3,

the error in following distance at time t can then be mathematically expressed as:

d d
€(n-1xr) = Ann—mt) —Ax(,ff])(,) = Ax(n-l)(t) -hn“ vn -fa an(t) (6.4)

The ultimate goal of well-designed vehicle control logic is to reduce the value of 8 and '
maintain it as close as possible to zero. In the context of sliding mode control, the
minimization of vehicle spacing error is realized by choosing a sliding surface along
which a would tend to zero or some negligible value. The sliding surface is defined here

as:

(fin—nu) = ‘1 «em—um, (6.5)

124

where A is the sliding coefficient and its value is always positive to allow the gradient of 8

to always points to zero.

While the above equations represent the fundamental principles of sliding mode vehicle
control, vehicle actuator lag should also be considered. A commonly adopted form is the

first-order delay model shown in Equation 6.6, where r is the vehicle delay coefficient.

Tan“) +an(t) = 0,3165) (6-6)

Differentiating Equation 6.4 with respect to time yields:

é(n—l)(t) = A-’.C(n—l)(t) “hill“ IIn “fa dn(t) (6.7)

Rearranging Equation 6.6 and substituting into Equation 6.7 further yields:

des
an(t) ’“nm (6.8)
T

 

é(n—1)(t) = Aim—1x2) — hi2!“ 9n - fa

Substituting Equation 6.5 into Equation 6.8 then yields:

ad?) _an(t)
. . I I
MOI—DU) _hges Vn — fa I T = —/I 8014)“) (6-9)

 

Finally, rearranging Equation 6.9 for odes" produces the following relationship:

an(t) _—

d
des _ 1_ Thnes
f

r . r/l
Jana) +—AX(n—1)(t) +—€(n-1)(t) (6-10)
fa a

This is the vehicle control logic used by Zhou and Peng (2004) in a previous study on

adaptive cruise control logic. It represents a basic vehicle control logic considering

125

information from a single lead vehicle. The relationship of desired acceleration and actual
acceleration achieved on vehicles is specified by Equation 6.6. The actual acceleration

can be expressed by solving Equation 6.6, which leads to the following relationship:

des I 1—e—G]

an(t+T) = an(t) + (amt) _ an(t) (6.11)

Using the definition of Equation 6.10, the final vehicle control algorithm used to update

vehicle acceleration is given by Equation 6.12.

an(t+T) : an(t) +
—T
(II—‘3 A] (6.12)

+ ’1 [(xn—Kt) _ xn(t) —ln—i)" hgisl v"(1) — fa 0,,(01 fa

- I153 anm + (VII-1(1) - mm)

 

In a last step, calculated vehicle accelerations are checked with vehicle acceleration and

braking limits. If a calculated acceleration exceeds the physical limit of a vehicle, the

actual vehicle acceleration is replaced with either its maximum acceleration amaxn or

maximum deceleration aminn, depending on whether the vehicle is accelerating or

braking. In addition to maximum acceleration and deceleration rates, the rate of change
in acceleration is also restricted by a deceleration jerk limit. This parameter represents

the harshest braking action possible by a vehicle attempting to avoid a collision.

6.3. Sliding-Mode Controller for Multiple-Lead Scenario

When data from multiple downstream vehicles become available, there is hypothetically
better chance to improve vehicle control performance if the additional information is

126

appropriately used. However, few studies have been found on multiple vehicle
following. In this research, the approach selected to incorporate traffic information fi'om

multiple downstream vehicles is to use an adjustment factor applied to the desired
headway, 116165. The headway adjustment factor fn,i(t ) used for this purpose is defined as:
an—i(t)

fn,i(t) = (-1) am”) (6.13)

"-1

 

where vehicle i is again the it}? lead vehicle ahead of vehicle n. For a given subject
vehicle, there may exist one or more adjustment factors, depending on the number of lead
vehicles whose traffic data are accessible. In such a case, the factor to apply is obtained
by taking the average of all the individual factors. The resulting factor is then used to

adjust the desired headway of the subject vehicle according to Equation 6.14.

hfles’m’j = hi?“ (1— mean(fn,;(t) )) (6.14)

When compared to its nominal value, the desired time headway is generally shortened
when the lead vehicles are exhibiting accelerating patterns, to counteract the fact that
accelerations tend to increase following distances. On the other hand, the time headway
is generally elongated when lead vehicles are braking, to counteract the tendency to push
vehicles closer. Within this context, it is anticipated that using lead vehicle information
may allow a control logic to be more proactive than if it were to consider only one lead

vehicle.

127

Introducing the adjusted desired headway defined in Equation 6.12, Equation 6.15 is
finally obtained to delineate vehicle acceleration with the use of information from

multiple lead vehicles.

an(t+T) = an(t) +

r 61 ,ad' . —T
_hnisl J “n(t) +("n—1(t)_vn(t)) T[1—e A]

< (xn—l(t) "xn(t) "ln-i) >

+2 d . fa
-—h es,adj v _f a

"_1 n(t) a n(t) I

(6.15)

 

 

 

At equilibrium, the values of all acceleration terms are zero, all vehicles share the same
speed, and all headway adjustment factors are one. In such a context, it can easily be
seen that Equation 6.15 then reduces to the basic expression of constant time headway

logic.

6.4. Multiple-Lead Vehicle Control Algorithm Refinement

While sliding-mode control keeps the vehicle following error close to zero, the actual
vehicle control command typically flips between acceleration and deceleration at a high
frequency. This is commonly known as a chattering effect around system equilibrium.
Figure 6.1 shows the high-frequency flips of acceleration commands observed on the lead
vehicle of a platoon of 40 vehicles accelerating from standstill to equilibrium on a closed
ring road. In this research, the method proposed to reduce chattering effect is to use a
moving average filter, as shown by Equations 6.16 and 6.17, to stabilize the output of

vehicle control algorithm.

128

1
Zan<r+m

 

__ =_k (6.16)

an(t+T) = J k+1

2i ( T) = 0 If 21.n(t+T) < aabs,min 6 17
n H En(t+T) if En(t+T) 2 aabs,min ( O )

As shown in Equation 6.16, k previous acceleration commands are considered when
filtering a present acceleration command. The output of Equation 6.16 is further filtered

by Equation 6.17 to remove trivial acceleration commands that may increase vehicle

instability.

 

 

 

Acceleration (In/3‘2)
:1;

 

 

 

'6 I T I I I I
0 50 100 150 200 250 300 350 400 450

Time (s)

 

Figure 6.1 Non-filtered Vehicle Acceleration Commands

Figure 6.2 shows the vehicle acceleration commands for the scenario of Figure 1 when
the refinement filter is added to the car-following logic. In this case, a moving average
filter with window size 3 (i.e., considering 3 acceleration commands issued in the past)

was used, with a requirement that all acceleration commands have an absolute value

larger than 0.01 m/s2. It can be observed in the figure that most flips in acceleration

129

commands ended up being filtered out, thus creating an acceleration profile much

smoother than in the non-filtered case.

 

I
.—
41

 

.21

-3 J,

Acceleration (m/s"2)

 

-5.

 

 

 

‘6 I I 7 I I I

0 50 100 150 200 250 300 350 400 450
Time (s)

Figure 6.2 Filtered Vehicle Acceleration Commands

130

CHAPT ER 7
IMPACT OF COMMUNICATION DELAY, MULTI-ANTICIPATIV E VEHICLE

CONTROL AND MANUALLY CONTROLLED VEHICLES

Chapter 6 presented an automated vehicle control algorithm that allowing a host vehicle
to fuse traffic information transmitted from downstream vehicles and to account for
potential information delay in its decision making. This chapter presents an evaluation of
the vehicle control logic that has been developed on the basis of the following influence

factors:

a Wireless communication delay,
Multi-anticipative vehicle control algorithm, and

1:) Presence of manually controlled vehicles.

A set of simulation runs were executed to estimate the impact of each of the above three
factors by changing the value of one particular factor at a time while holding the other
two constant. In these runs, two levels of wireless communication delay are examined,
zero delay and 300 ms delay. The evaluation on multi-anticipative vehicle control
algorithm also considers two levels, by comparing single-leader versus multi-leader car-
following. Finally, the effect of human controlled vehicles is evaluated by varying the
market penetration rate of autonomous vehicles within a traffic stream in a range

extending from 0% to 100%.

The first section presents three evaluation scenarios that were implemented to test each
combination of impact factors considered, and documents the values that were assigned

to the vehicle control and simulation parameters in each scenario. The next three

131

sections, grouped by impact factor, present the simulation results associated with each of
the three scenarios considered for each of the impact factor being studied. The last

section finally provides a general summary of the findings of the simulation studies.
7.1. Evaluation Scenarios and Assumptions

Three car-following scenarios were developed to investigate the impacts of influence
factors on vehicle and traffic flow behavior. All scenarios consider a platoon of 40
vehicles traveling in a single file. Each simulates a specific traffic event to test specific
types of vehicle responses. These three scenarios are illustrated by the speed profiles of
Figure 7.1, which each illustrates the speed trace of the lead and first 15 following
vehicles in the platoon. These scenarios are explained in greater details in the following
paragraphs, with Table 7.1 providing numerical values associated with the main scenario

parameters.

The first scenario considers vehicles traveling in a single file on a one-lane circular road.
Its purpose is to evaluate how quickly equilibrium can be achieved by a given control
logic under various communication delay and information sharing assumptions. All
vehicles are assumed to be initially immobilized. At a given time, the first vehicle in the
platoon starts to accelerate and continue to do so with a parabolically diminishing rate
until it reaches its desired speed. In turn, all the vehicles behind the lead vehicle start to
accelerate, considering both their desired to attain a certain speed and the constraints

imposed by the vehicles ahead.

132

 

35 I I r I I I fi I I

Speed (m/s)

 

 

 

 

0 l 1 l 1 I l 1 J l
O 100 200 300 400 500 600 700 800 900 1000
Time (s)

(a) Acceleration from Standstill to Equilibrium

 

35 T r fi T I

Speed (m/s)

 

 

 

 

 

 

   
         

 

 

 

 

 

 

 

 

 

 

 

 

 

10 l l l l I
40 60 80 100 120 140
Time (s)
(b) Emergency Braking
35 I I I I I I T I
30 \\ a
’6? 20 - ..
E .-
‘o 15 — . l , ‘
Q) I I
d) i I I
(to). 10 ’- i I I I iI -‘
5 - ‘ iii". -
0 _ i I 1 ii I
-5 I I I I I L m g
0 10 20 30 4O 50 60 70 80 90

Time (s)

(c) “Brick Wall” Collision

Figure 7.1 Vehicle Speed Profiles of Evaluation Scenarios

133

Table 7.1 Vehicle Control and Simulation Parameters.
Standstill to Emergency “Brick Wall”

 

 

 

variable Equilibrium Braking Collision
Initial vehicle speed 0 m/s 30 m/s 30 m/s
Initial headway -- 1'0 S 1'0 5
Road length 1000 m " "
Platoon size 40 veh

Desired speed 30 m/s

Desired headway 1.0 s

Spacing between stopped vehicles 4.0 m

Vehicle length 6.0 m

Maximum Acceleration 0.8 m/s2

Maximum Deceleration -5 m/s2

Maximum Braking Jerk - -30 m/s3

Information delay (wireless communication 03 s

delay + on-board dev1ce processm g tlme)

Sliding mode model parameter (T8) 0.1

Sliding mode model parameter (r) 0.8

Sliding mode model parameter 0») 0.3

Critical time-to-collision (770*) 5 s

 

The use of a circular road implies that at some point the lead vehicle in the platoon will
start to interact with the last vehicle in the platoon. In most cases, this results in a need
for the lead vehicle to decelerate as it generally travels across the circular path before the
last platoon vehicle has fully accelerated. This creates a complex situation in which
shockwaves created by accelerations and decelerations along the road keep propagating
around the circle, thus creating the oscillation speed pattern observed in Figure 7.1(a).
To achieve equilibrium in such a situation, a control algorithm must have the ability to

dampen shockwaves. Equilibrium is assumed to be reached when the speeds of all

134

vehicles in the platoon are maintained within 10.5 m/s of the average space mean speed

of all vehicles.

The second scenario simulates emergency braking along a straight, open highway
section. Its purpose is to evaluate the ability of vehicles to respond to typical collision
threats and return traffic to equilibrium following a sudden disturbance. As illustrated in
Figure 7.1(b), all vehicles are initially assumed to travel at a constant speed. At a given
time, the lead platoon vehicle initiates a harsh deceleration, decelerating at a constant rate
until it reaches a speed of half its initial speed. Following a few seconds of travel at a
constant speed, the vehicle starts accelerating again until it reaches back its desired speed.
All following vehicles then successively respond to the deceleration and acceleration of

the vehicles ahead of them according to their imbedded control logic.

The third scenario pushes the threat of collision to the highest level by staging a so-called
“brick wall” collision scenario. This scenario is similar to the second one, except for the
fact that the lead platoon vehicle is assumed to come to an instantaneous stop following a
hypothetical collision with an object on the road. After this event, all following vehicles
will then try to decelerate to avoid colliding with the lead vehicle. However, given the
suddenness of the stop, it is expected that a certain number of vehicles will be unable to
avoid colliding. In Figure 7.1(c), colliding vehicles are those for which the
corresponding speed trace line suddenly ends and is completed by a dotted line. By
comparing the number of vehicles colliding and the speed at which they collide, the
effectiveness of various vehicle control logic in anticipating deceleration needs can then

be assessed.

135

The evaluations presented in this chapter are subject to certain assumptions to clearly
demonstrate the impacts of influence factors. These assumptions are primarily that all
vehicles under control of driver assistance systems have the same and constant desired
headway, desired speed, initial speed, physical dimensions, and acceleration and braking

capabilities.

7.2. Impacts of Information Delay

This section presents the results of the evaluations focusing on the impacts of information
delay on vehicle control when decisions are based uniquely on information about the
position, speed and acceleration level of the vehicle immediately ahead. A specific
assumption for this scenario is that all vehicles implement the sliding mode control logic
presented in Chapter 6 as experiments with traffic mixing human drivers are presented
later. Simulation results are presented below successively for the three scenarios

described earlier.

Scenario 1: Standstill to equilibrium

For the standstill to equilibrium scenario, Table 7.2 shows that the time for the simulated
platoon to reach equilibrium is more than doubled when an information delay of 0.3 s is
introduced into the control loop. In the absence of delay, equilibrium, as defined by a
speed differential of less than 0.5 m/s, is reached in just over 5.5 minutes within the
circulatory test track. With a delay of 0.3 s, it takes more than 11 minutes to reach
equilibrium from an identical initial condition. An analysis of simulation results clearly
indicates that this extended time is primary the result of the inability of individual

vehicles to obtain precise information regarding the position, speed and acceleration of its

136

lead vehicle in the presence of information delay. This results in situations in which
control decisions more frequently overshoot or undershoot their true target, thus creating

a need for extra time to reach equilibrium.

Table 7.2 Time to Reach Equilibrium from Standstill, Single Leader.

 

 

Number of Information Duration to Reach Equilibrium
Lead Vehicles Delay (ms) Equilibrium (8) Speed (m/s)

1 0 331 15.0

1 100 438 13.6

1 200 555 12.5

1 300 684 11.5

1 400 880 10.7

1 500 1040 10.0

 

 

Another interesting result is a reduction of the average vehicle speed at equilibrium. In
the absence of information delay, the equilibrium speed is 15.0 m/s (54.0 km/h). With a
delay of 0.3 s, the speed drops to 11.5 m/s (41.4 km/h). This reduction is again explained
by the use of erroneous information in the control loop. With a 0.3 3 communication
delay, vehicle control decisions are typically based on the position that the each
preceding vehicle occupied 0.3 3 ago. At a speed of 11.5 m/s, this translates into
assuming that lead vehicles are 3.5 m closer than they are in reality. To maintain a
desired headway, a speed reduction is then imposed to increase spacing with the lead

vehicle, thus leading to a lower equilibrium speed.

An interesting coincidence in the data of Table 7.2 is that the difference in distances
traveled in l s at equilibrium with and without information delay equals the distance that

a vehicle experiencing information delay travels during the assumed information delay.

137

For example in the scenario of 0.3 s delay, the difference in distance, (15 m/s — 11.5 ms) -
ls = 35 m, corresponds to the distance that a vehicle experiencing information delay
travels during the assumed information delay, 11.5 m/s - 0.3 s = 3.5 m. Further tests

considering varying levels of information delay also support this apparent relationship.

Scenario 2: Emergency braking

The focus of the emergency braking scenario is on the safety impacts of information
delay. Table 7.3 presents for this scenario the results of the evaluation for the surrogate
safety measures described in Chapter 4. Only TTC and acceleration noise values are
shown as the TTC values produced for each vehicle were all above 5 8. Since the critical
time-to-collision (TTC*) is set at 5 3, this results in all TTC-based indicators having a

value of zero.

For each parameter, Table 7.3 shows the values associated with the vehicles returning the
lowest TTC or highest acceleration noise, as well as cumulative sums reflecting the
values assessed for all vehicles in the platoon. Vehicles are identified according to their
position in the platoon relative to the lead vehicle. From these results, it can be observed
that the introduction of communication delay unexpectedly results in higher TIC values
and in an assessment that information delay may potentially improve traffic safety.
Although such an assessment is contradictive to the common assumption of negative
impacts, it should be noted that safety-critical situations did not occurred in both
simulated alternatives. The true cause of the increases in TTC is related to the weights
assigned to the errors regarding vehicle spacing error and speed in the control algorithm.

In the absence of information delay, vehicles have few difficulties maintaining a given

138

spacing to satisfy the constant l-s headway separation rule. However, as was explained
before, vehicles tend to perceive shorter vehicle spacing with information delay, thus
pushing them to maintain longer separation distances. This increase in separation

distance then results in lower collision threats and increased assessed TTC values.

For this scenario, a better performance indicator seems to be the acceleration noise, as
this parameter focuses on risks created by speed variations and is not dependent on the
presence of particular safety threats. In Table 7.3, it can indeed be observed that the
acceleration noise expectedly increases with the introduction of information delay,
indicating an increase in speed variations and higher collision risks that contradict the

TTC evaluations.

Table 7.3 Surrogate Safety Measures, Emergency Braking, Single Leader.

 

 

 

 

 

Information TTC Acceleration Noise
Delay Lowest Sum Highest Sum
(s) Veh (s) (s) Veh (m/sz) (m/sz)
0 2 6. 1 330 0 0.59 9.01
0.3 3 6.8 420 0 0.59 15.22

 

Figure 7.2 presents another look at the impact of information delay on acceleration noise.
This figure shows the impact on each vehicle in the platoon. The figure clearly indicates
that acceleration noise is almost the same with or without information delay for the first
few vehicles. However, as we move further back into the platoon, vehicles are observed
to experience marked increases in acceleration/deceleration activity with communication
delay. For the most part, these results can be explained by differences in shockwave

propagation speed that are introduced by communication delay.

139

 

To illustrate the above effect, Figure 7.3 compares the trajectories of all simulated
vehicles in the alternatives with and without information delay. As can be observed in
Figure 7.3(a), the shockwaves created by the deceleration/acceleration cycle of the lead
platoon vehicle meet in the absence of delay before reaching the end of the platoon.
Consequently, no vehicle beyond the 25‘h platoon vehicle is affected by the
deceleration/acceleration event of the lead vehicle. In Figure 7.3(b), which presents the
alternative with information delay, the shockwaves are clearly seen extending throughout
the entire platoon as a result of slower propagating speeds. Similar to the first scenario,
this reduction in shockwave propagation can be explained by the increased difficulty to
achieve equilibrium or quickly respond to traffic events caused the by the use of slightly

erroneous data in the vehicle control process.

 

 

I No Information Delay

0.6 CI Information delay is 300 ms per hop

 

.0 .0 53
w A u.

Acceleration Noise (m/sz)

_O
N

 

0.1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0 lllllllllllll
111111 V11 ITIIII'IIIIIIIII lllllllll

0 5 10 15 20 25 30 35

Vehicle in Platoon

Figure 7.2 Vehicle Acceleration Noise, Single Leader, with VS. without Delay

140

 

 

Distance (m)
Distance (m)

 

 

 

 

 

 

Time (s) Time (s)
(a) without Information Delay (b) with Information Delay

Figure 7.3 Vehicle Trajectories in the Emergency Braking Scenario

Scenario 3: “Brick wall” collision

The third scenario focuses on the safety impacts of information delay in extreme
circumstances. For this scenario, the simulation results indicate that 12 vehicles collide
with each other following the sudden immobilization of the lead platoon vehicle in the
presence and absence of information delay. As indicated by Table 7.4, collision speeds
do not significantly change with the presence of information delay. This leads to an
apparent lack of impacts on safety and this result is again explained by errors in vehicle
information introduced by information delay. In the two previous scenarios, information
errors caused vehicle controllers to believe that the lead vehicles were closer than they
were in reality. This perception pushed for an increased in following distance. While the
same perception exists here, information errors also lead control systems to believe that
the vehicles ahead are traveling faster than in reality and that higher vehicle speeds may
be allowed. “’hile the perceived closer spacing pushes on one hand for a reduction in

vehicle speed, the perceived higher speed of vehicles ahead pushes on the other hand for

141

an increase in vehicle speed, thus muting the impacts of increased or reduced information

delays.

Table 7.4 Colliding Speed, “Brick Wall” Collision, Single Leader.

 

 

 

Vehicle Position Colliding Speed (m/S)
in Platoon N0 Delay 300 ms Delay
1 25.0 25.0
2 23.0 23.0
3 21.5 21.5
4 20.1 20.1
5 18.7 18.8
6 17.4 17.5
7 16.2 16.2
8 14.4 14,5
9 12.7 12.8
10 1 1.0 1 1.0
11 8.8 8.9
12 5.6 5.7

 

Table 7.5 summarizes in a format similar to Table 7.3 the values estimated for the
surrogate safety measures described earlier. In this case, the introduction of vehicle
control delay clearly results in reduced safety. For instance, the introduction of

information delay lowers the TTC value for the vehicle having the highest collision risk

(l3th platoon vehicle) from 2.1 to 1.9 8. Higher individual TET and TIT values are also
produced, while no apparent changes are observed in individual CTF and CJF values.
While the platoon TTC and CJF show improvement with information delay, all other
parameters indicate an increase in safety risks consistent with the individual values. The
improvements in cumulative TTC and CJF values with information delay are again

explained by effects of erroneous information on vehicle control decisions.

142

Table 7.5 Surrogate Safety Measures, “Brick Wall” Collision, Single Leader.

TTC TET TIT CTF CJF
Lowest Sum Highest Sum Highest Sum Highest Sum Highest Sum
n Veh (s) (s) Veh (s) (s) Veh (52) (s2) Veh (40 (40 Veh no (it)
1“” 13 2.1 152 13-19 2 16 13 3.7 14.1 13-21 1 9 14-16 2 6
1“” 13 1.9 184 13-21 3 33 13 6.5 30.9 13-25 1 13 14,15 2 5

 

 

Setup

 

 

 

 

(a) without delay; (b) 300 ms information delay

 

A comparison of the TTC evaluation results of Table 7.5 to those of Table 7.3 clearly
indicates that the brick wall scenario presented higher safety risks than the emergency
breaking scenario. This is an expected result that validates the use of selected
performance measures. When going from the emergency to the brick wall scenarios, the
lowest individual TTC drops from roughly 6 s to roughly 2 s. The cumulative platoon
value also shows a similar drop, with a reduction from an initial range of 300-400 s to a
range of 150 — 200 s. Acceleration noise showed no difference in this comparison and

therefore is not listed in the table.

7.3. Impacts of Multiple Lead Vehicles

This section presents the results of the evaluations focusing on the impacts of vehicle
control logic when control decisions are based on information about a number of lead

vehicles.

Scenario I .' Standstill to equilibrium

For the standstill to equilibrium scenario, Table 7.6 shows that traffic equilibrium can be
reached much faster in the presence of information delay by incorporating in the control

logic information from a number of lead vehicles. These gains are primarily obtained by

143

extending system vision from the vehicle immediately ahead to the first two lead
vehicles. Extending vision beyond the first two vehicles did not provide additional
benefits. Each increase in the number of lead vehicles considered resulted here in an
identical equilibrium speed and a slight increase in time to reach equilibrium. The longer
time to reach equilibrium is attlibuted to the increased complexity of achieving stability
when having to deal with an increasing number of variables each marked with a certain

degree of error.

Table 7.6 Time to Reach Equilibrium from Standstill, Multiple Leaders.

 

 

 

 

 

 

Number of Information Duration to Reach Equilibrium
Lead Vehicles Delay (ms) Ejuilibrium (5) Speed (m/s)

1 0 331 15.0

300 684 11.5

2 0 85 15.0

300 136 11.5

3 0 80 15.0

300 142 11.5

4 0 83 15.0

300 146 11.5

0 86 15.0

5 300 149 11.5

 

 

Scenario 2: Emergency braking

Similar to the emergency braking scenario with a single—lead vehicle, all individual
vehicle TTC evaluations produced values above 5 8, thus resulting in zero values for all
other TTC-based surrogate measures. In this case, the results of Table 7.7 clearly show
that including information about additional lead vehicles can lead to safety
improvements. This is demonstrated by the higher TTC and lower acceleration noise

values produced for the alternatives with two or more lead vehicles. However, similar to

144

the results of Table 7.6, the maximum benefits appear to be when system vision is
extended from one to two lead vehicles only, as TTC values start to decrease while

acceleration noise start to increase again when considering three or more lead vehicles.

Table 7.7 Surrogate Safety MeasuresLEmergency Braking, Multiple Leaders.

 

 

 

TTC Acceleration Noise
Setup Lowest Sum Highest Sum
N Veh (s) (s) Veh (m/sz) (m/52)
1 3 6.84 420 0 0.59 15.22
2 1 7.65 757 0 0.59 10.03
3 1 7.66 709 0 0.59 10.33
4 1 7.66 679 0 0.59 10.54
5 1 7.66 619 0 0.59 11.01

 

The estimated acceleration noise values for each vehicle in alternatives considering one
and two lead vehicle are compared in Figure 7.4. This figure clearly shows that all
vehicles experience a reduction in acceleration/deceleration activity, and thus less
instability, when two lead vehicles are considered instead of one. Compared to Figure
7.2, the impact appears to substantially reduce the acceleration noise of the first few
following vehicles, with vehicles further behind experiencing a less pronounced
reduction. It can further be observed that the acceleration noise with a control system
considering two lead vehicles in the presence of 0.3 s delay is even lower than for a
system considering a single lead vehicle with no delay for the first 17 following vehicles.
This is yet another strong indication of the potential safety benefits of designing vehicle

control system considering information from more than one vehicle ahead.

145

0.7

 

 

El lnfonnation delay is 300 ms per hop

0.6 I Following two downstream vehicles

 

.9 .9 9
1...) A Lil

Acceleration Noise (m/sz)

.0
N

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0 11.11111Trrriillvlrriv’titiiit’i

0 5 10 15 20 25 30 35

Vehicle in Platoon

Figure 7.4 Vehicle Acceleration Noise, Single VS. Multiple Leaders, with

Information Delay

Scenario 3: “Brick wall” collision

For the “brick wall” scenario, considering more than one lead vehicle results in a
reduction of the number of vehicles colliding from 12 to 2. The change primarily occurs
when expanding the control logic to consider two lead vehicles. This indicates a marked
improvement in safety. The surrogate safety measures of Table 7.8 emphasize once more
that the consideration of multiple lead vehicles has positive impacts on safety. For TTC,
the best results in terms of individual vehicle performance are obtained when considering
four or five lead vehicles, while the best platoon performance is with only two lead
vehicles. For TET, the best individual performance is for three or four lead vehicles

while the best platoon performance is for two lead vehicles. For TIT, a four lead vehicle

146

setup is best, while a two-vehicle setup is best for CTF. The deterioration of CJF value
with multiple lead vehicles is largely attributed to the chattering effect of sliding mode
control around its sliding surface. While these differing results may be indicative of
random and simulation effects, they nevertheless again clearly demonstrate the potential
safety benefits of including more than one lead vehicles in the vehicle control loop in

scenarios considering information delay.

Table 7.8 Surrogate Safety Measures, “Brick Wall” Collision, Multiple Leaders.

TTC TET TIT CTF CJF
Lowest Sum Highest Sum Highest Sum Highest Sum Highest Sum,

 

 

Setup

 

n Veh (s) (s) Veh (s) (s) Veh ($2) (52) Veh (it) (It) Veh (it) (It)

 

 

1 13 1.9 184 13-21 3 33 13 6.5 30.5 l3-25 l 13 14,15 2 5

2 3 2.4 415 3 3 6 3 6.8 9.0 3,5,6 1 3 (a) 36 144
3 3 2.9 365 6-9 2 9 6 3.2 8.0 (b) 1 5 35,39 27 496
4 7 3.3 357 7-9 2 7 7 2.8 6.4 (C) l 5 (d) 19 361
5 8 3.3 305 9 3 12 8 3.3 10.6 (e) l 6 38 23 326

 

(a) 26,34,35,37; (b) 3,6,7,8,9; (c) 3,7,9,9,10; (d) 33,34,36,37,38; (e) 3,8,9,10,l 1,12.

 

7.4. Impacts of Manually Controlled Vehicles

This section presents the results of simulations that were conducted to assess the impact
of having manually controlled vehicles present in a traffic stream at the same time as
automated vehicles. For this assessment, the same three test scenarios as before were
considered, with the exception of the use of the Gipps (1981) car-following model as the
primary decision making logic to emulate human driver behavior in manually controlled
vehicles. An anticipated result from this experiment was that the relative large headway

requirement of Gipps model would undermine the effectiveness of emergency braking

147

test, and subsequently call for additional runs for the emergency braking scenario with a
modified single-leader vehicle control logic emulating more closely human driver

behavior.

For this evaluation, simulations were conducted for each evaluation scenario with
automated vehicle market penetration levels of 0%, 10%, 30%, 50%, 70%, 90% and
100%. A stochastic procedure was used during the vehicle generation process to
determine whether a specific vehicle would be controlled by an automated system or a
human driver, thus leading to different sequences of automated and manually controlled
vehicles in each simulation run. To account for this stochastic variation, each simulation
setup was tested with a minimum 5 runs, with the execution of additional runs when

deemed necessary.

Scenario 1: Standstill to equilibrium

The standstill to equilibrium tests were performed assuming two alternatives: 0.3 s
information delay and no delay. Tables 7.9 and 7.10 respectively show the time to reach
equilibrium and equilibrium speed in the scenario considering 0.3 s information delay.
The results for non-delay cases are shown in Tables 7.11 and 7.12. It can be observed
that the time to reach equilibrium from standstill uniformly decreases with increasing
proportions of automated vehicles in platoon, reflecting a better shockwave dampening
property for the automated vehicles than for the manually controlled vehicles.
Comparing the results by communication delay levels, it is further observed that the
reduction in the time to reach equilibrium is more prominent in cases with no

communication delay. This is attributed to the fact that with less delay each vehicle is

148

provided information about surrounding vehicles that more accurately reflects reality,
thus allowing for a better assessment of vehicle spacings leading to more compact

headways and increased travel speeds.

 

Table 7.9 Time to Reach Equilibrium from Standstill, Two Leaders, 0.3 8 Delay.

 

 

 

 

Market Penetration Rate of Automated Vehicles

Runs 0% 10% 30% 50% 70% 90% 100%

1 321 31 l 293 240 197 165 1. 36

2 -- 306 257 250 193 153 --

3 -- 312 299 210 196 162 --

4 -- 314 266 214 187 153 —-

S -- 326 253 220 187 173 --
Mean 321 313.80 273.60 226.80 192.00 161.20 136
Std. Dev. -- 7.43 21.09 17.36 4.80 8.50 --

 

 

 

Table 7.10 Vehicle Equilibrium Speed, Two Leaders Scenario, 0.3 s Delay.

 

 

 

 

Market Penetration Rate of Automated Vehicles

Runs 0% 10% 30% 50% 70% 90% 100%

l 8.40 8.60 8.82 9.37 10.17 10.83 11.56

2 -- 8.65 9.19 9.32 10.26 11.03 --

3 -- 8.60 8.75 9.90 10.16 10.83 --

4 -— 8.49 9.12 9.73 10.35 11.04 --

5 -- 8.59 9.24 9.64 10.23 10.63 --
Mean 8.40 8.59 9.02 9.59 10.23 10.87 1 1.56
Std. Dev. 0.0 0.06 0.22 0.24 0.08 0.17 0.0

 

 

149

 

Table 7.11 Time to Reach Equilibrium from Standstill, Two Leaders, No Delay.

 

 

 

 

Runs Market Penetration Rate of Automated Vehicles

0% 10% 30% 50% 70% 90% 100%

l 321 315 229 197 125 118 85

2 -- 301 236 174 151 128 --

3 -- 285 228 198 142 109 --

4 -- 308 227 128 130 133 --

5 -- 274 178 163 1 17 134 --
Mean 321 296.60 219.60 172.00 133.00 124.40 85
Std. Dev. 0.0 16.83 23.52 28.82 13.55 10.69 0.0

 

 

 

Table 7.12 Vehicle Equilibrium Speed, Two Leaders, No Delay.

 

 

 

 

Market Penetration Rate of Automated Vehicles

Runs 0% 10% 30% 50% 70% 90% 100%

1 8.40 8.48 9.50 9.99 11.82 12.04 15.06

2 -- 8.73 9.50 10.35 10.93 11.61 --

3 —- 8.81 9.63 10.00 1 1.28 12.39 --

4 -- 8.57 9.49 11.61 11.56 11.65 --

5 -- 9.01 10.38 10.60 11.99 11.42 --
Mean 8.40 8.72 9.70 10.51 11.51 11.82 15.06
Std. Dev. 0.0 0.21 0.39 0.67 0.42 0.39 0.0

 

 

The equilibrium speeds shown in Tables 7.10 and 7.12 indicate that higher equilibrium
speed can be achieved with more automated vehicles in platoon, confirming the
expectation that more vehicles capable of handling short headway shall lead to higher
equilibrium speed. When the experiment results are compared for a given market
penetration level for different levels of communication delays, the results verify previous

finding that communication delay reduces the equilibrium speed by introducing

150

inaccuracies in the reported data that eventually push the automated controllers to be

more cautious in their decisions.

The mean values of the time needed to reach equilibrium and of the equilibrium speed are
plotted in Figures 7.5 and 7.6 to better visualize how both measures change with
variations in the automated vehicle market penetration rate. In both figures, it can first be
observed that the curves for no delay and 300-md delay both converge to a single point at
0% market penetration level, correctly indicating that a traffic stream comprising only
manually controlled vehicles is not affected by the presence of communication delay, as
it should. At higher market penetration levels, it is observed from Figure 7.5 that the
impact of communication delay on the time to reach equilibrium is more prominent when
the market penetration rate is higher than 10%. The relatively less impact of
communication delay observed at 90% automated vehicle market penetration rate may be
attributed to the randomness of this test. It may further be observed that the points of “no
delay and 30%” and “with delay and 70%” both indicate approximately the same time to
reach equilibrium. Similarly, the points of “no delay and 50%” and “with delay and
90%” are also approximately on the same level. These findings indicate that the
performance degradation introduced by a 300 ms communication delay on the time to
reach equilibrium could in this case be mitigated by having a larger proportion of

automated vehicles (in this case, more than 40% automated vehicles)

Figure 7.6 further shows that vehicle equilibrium speeds gradually increase when the
proportion of automated vehicles changes from 0% to 90%. The increase in equilibrium
speed when market penetration improves from 90% t0100% is more significant than

lower market penetration levels when no communication is involved. Such phenomenon

151

indicates that even a small proportion of manually controlled vehicles may disturb flow
in the whole system and prevent all vehicles from reaching a higher equilibrium speed.
With the consideration of communication delay, however, the increase in vehicle
equilibrium speed is at a nearly constant slop with increasing market share. Similar to
Figure 7.5, Figure 7.6 may indicate that reductions in equilibrium speed caused by
information delays can be mitigated by introducing higher proportions of automated

vehicles in the traffic stream.

 

 

 

-O— No Commlmication Delay
+ 300ms Information Delay

   
  

 

 

 

150 ~

100—

Time to Reach Equilibrium (s)

50—

 

 

 

0 I I I I l I I I l

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
CACC Market Penetration Rate

 

 

 

Figure 7.5 Time to Reach Equilibrium from Standstill with Varying Market

Penetration Levels

152

 

16.00

 

 

—O— No Informal ion Delay

M I
1500 + 300ms Information Delay

 

 

 

14.00 ~

13.00 —

12.00 2

11.00—

10.00 ~

Equilibrium Speed (mls)

9.00 4

 

 

 

I
8.(X) I I I I I l I r I
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

CACC Market Penetration Rate

 

 

 

 

Figure 7.6 Equilibrium Speed with Varying Market Penetration Levels

Scenario 2: Emergency braking

Similar to the standstill to equilibrium scenario, the Gipps model was used to evaluate the
behavior of a mixed human driver/automated traffic stream for the scenario involving the
emergency breaking of the simulated platoon’s lead vehicle. In this case, automated
vehicles receive traffic information with 300 ms communication delay. All individual
vehicle TTC evaluations produced values above 5 8, thus resulting in zero values for all
other TTC-based surrogate measures. The results of Table 7.13 show that for this
scenario higher levels of automated vehicles unexpectedly lead to higher collision threats
and increased levels of instability. Given the long headway required by the Gipps model,
manually controlled vehicles generally maintain much larger safety margin for
maneuvering. Most of the observed disturbances are then associated with the
automatically controlled vehicles, which try to maintain headways of only 1 s and are

therefore more sensitive to fluctuations in traffic stream behavior.

153

Table 7.13 Safety Surrogates, Emergency Braking, Two Leaders, Manual: Gipps.

 

 

 

 

 

 

'I'I‘C Acceleration Noise
5“” Lowest Sum Highest Sum
(ms) (ml)
062 ______ V . 12.05 """"
o 0.62 .1178
o 0.62 11.82
0 0.59 1167
10% 1 13.04 9625.71 0 0.62 11.84
10% 1 13.04 7645.62 0 .

 

50% l 13.04 2221.74 0

50% 1 13.04 1657.60 0 0 62 10.29

50% 1 7.66 1782.09 0 0.62 10.73

50% l 13.04 2328.36 0 0.62 9.14
0

 

6.84 541.07 0.59 15.19

 

 

3 0
90% 3 6.84 498.75 0 0.59 15.12
90% 3 6.84 751.67 0 13.95
6 8.76 557.78 0 14.29
3 6.84 814.60 0 12.77
3. 68442021 o """ ' 1533....

   

 

154

As the vehicle headways that are maintained by the Gipps model are too large for the
intended evaluation of impacts of communication delays on mixed traffic streams, an
alternative human-based car-following model is needed to allow manually controlled
vehicles to drive with relatively shorter headways. To provide such a model, the
automated vehicle control logic was used by setting it to consider only one lead vehicle
with 0.8 3 “communication delay” to emulate are relatively short human perception-
reaction delay. This change results in the simulation of relatively similar headways for
both manually controlled and automated vehicles. Table 7.14 shows the simulation
results obtained with the modified human driver model. When comaring to the results
from the previous experiment shown in Table 7.13, it is observed that all TTC values for
individual vehicles remain above 5 s, while the sum of lowest TTC values show a
decreasing trend with increasing proportions of automatically controlled vehicles. This
trend is again attributed to the relative long perception-reaction delay associated with
manually controlled vehicles, which pushes for larger vehicle spacing before the lead

vehicle initiates emergency braking.

Measurements of acceleration noise for the emergency breaking scenarios show that the
platoon lead vehicle typically experience more instability than any of the following
vehicles. While the sum of acceleration noises for all vehicles in the simulated platoons
does not change substantially at low levels of automated vehicle market penetration,
noticeable decreases are observed when the proportion of automated vehicles within the
traffic stream reaches and exceeds 90%. This is explained by the ability of the automated

controller to quickly react to changes in traffic conditions ahead, which lead to an

155

Table 7.14 Safety Surrogates, Emergency Braking, Two Leaders, Manual =
Modified CACC.

 

 

 

 

TTC Acceleration Noise

Setup Lowest Sum Highest Sum
N Veh (s) (s) Veh (m/s3) (In/s2)
0%“ f M 1 H 8.36" ' 555.99 ' ‘0 I H 0.59 ”17.94 M '
10% 1 8.36 542.96 0 0.59 18.38
10% 1 8.36 562.57 0 0.59 18.05
10% 1 8.36 558.64 0 0.59 18.23
10% 1 8.36 561.22 0 0.59 18.20
10% l 8.36 561.62 0 0.59 18.29
30% 3 6.84 ' ’ 496.15 0 0.59 " 18.107
30% 6 7.63 549.56 0 0.59 17.91
30% 1 8.36 571.57 0 0.59 17.82
30% 1 7.66 564.64 0 0.59 17.72
30% 3 7.58 498.53 0 0.59 18.53
50% 1 8.36 533.91 0 0.59 18.00
50% 9 7.80 458.39 0 0.59 18.65
50% 6 7.56 493.16 0 0.59 18.35
50% 5 7.04 494.39 0 0.59 17.93
50% 2 7.1 1 422.23 0 0.59 18.53
70% 3 6.84 437.80 0 0.59 17.42
70% 2 7.1 1 464.61 0 0.59 18.34
70% 3 6.84 419.40 0 0.59 17.47
70% 3 6.84 437.71 0 0.59 17.41
70% 3 6.84 7 412.34 0 0.59 .. 16.87 .
90% 3 6.84 409.52 0 0.59 16.40
90% 3 6.84 414.59 0 0.59 16.51
90% 3 6.84 406.13 0 0.59 15.98
90% 2 7.1 1 410.76 0 0.59 17.90
90% 3 6.84 41 1.48 0 0.59 15.65
100% I 3 6.84 ” 420.21 770 0.59 i " 15.22 7' "

 

improved ability to dampen unnecessary speed fluctuations and create more stable traffic
flows. In particular, the fact that significant reductions in acceleration noise only occur
with market penetration levels is a tentative indication that high levels of automated
vehicle market penetration may be needed before envisioned automated vehicle

controllers could provide all anticipated safety benefits.

156

Scenario 3: “Brick wall” collision

For the “brick wall” scenario, an initial set of simulations with 5 replications for a
particular market penetration level were executed. Results from these simulations are
summarized in the grey highlighted area of Table 7.15. After executions of these runs, it
was observed that the primary variable of interest, the number of collisions, showed
significant variability within each market penetration level to prevent the identification of
statistically valid trends across market penetration levels. These variations are attributed
to the randomness when assigning vehicle types for the first few vehicles in platoon. In
the worst case, for instance, a machine controlled vehicle may initiate unnecessarily harsh
deceleration to avoid collision with the vehicle in front and be rear-ended by a manually
controlled vehicle trailing behind. To generate a clearer picture of safety trends with
increasing market penetration levels, additional simulations were then executed to
increase the number replications for each market penetration level to 10 runs. The results
of these additional runs are shown in the non-highlighted area of Table 7.15. In both
cases, it should be noted that 300 ms information delay were considered for each wireless
data transmission and data from vehicles involved in crashes are excluded from the

evaluation of the safety surrogate measures TTC, TET, TIT and acceleration noise.

In this scenario, it is observed that the average number of vehicles involved in crash
varies tend to vary with the level of automated vehicle market penetration rate. For
instance, while the first five following vehicles in the simulated platoon were found to be
involved in collisions when the traffic stream is composed exclusively of manually
controlled vehicles, only three vehicles are typically found to be involved in collisions

when the market penetration reaches 90% or higher. This clearly indicates the ability for

157

automated vehicle control systems to provide safety improvements under extreme

condition in the presence of moderate information delays.

Both the TET and TIT statistics show a clear benefit when large proportions of vehicles
are automatically controlled. At mid-market penetration levels, these two measures show
non-conclusive trends. For acceleration noise, larger acceleration noises are observed
with more vehicles automatically controlled. The deterioration of acceleration noise is
primarily attributed to the fact that manually controlled vehicles follow a preceding
vehicle with much larger spacing to than what automated vehicles typically maintain.
When impeding collision threats are present, larger vehicle spacings provide sufficient
deceleration space to effectively avoid collisions in a large set of traffic conditions,
provided that not too much time is lost by drivers in perception-reaction time. In
contrast, automated vehicles have to come to stop within a much shorter distance, leaving
much less room for errors. These systems also have to deal with the chattering effect of
sliding mode control around its sliding surface and potential traffic data inaccuracies due

to information delay.

158

 

 

 

159

 

 

9;: 2 3.0 R _ _ 85 mm : m 3mm 2 New 6 s8
$5 : 36 mm o 24.6 mm o m an: o m: w .88
8.2 o 23 2 s New 8 mg m 8.3.8. 8 new . o 4.8
£9 6 m2 2. a son 2 s m 8.98 n a: m seem
BE, 2 3.0 mm 2 New on 2 m wow: 2 m3 8 .888
8.2 m 6.0 S m em.» 8 m m .2.me m «2 m son
8.: 4 8.. «.8. o 2.2 on o o 2.68 o 03 w o sow
mm: m cod 2 m one 2 m m :6: m N3 m s2
3:; 2 Ed : o mam mm 8.3 4 onmom o Nom 6 s2
8.: 6 Ed 2 o :s 2 o m 8.9.: o w; o sou
m3; 2 48 t 6 3m 2 8.8 a. :88 o mom o .2:
3.: w $6 2 s new 8 so 4 3.5 m as m s2
3.: m 8.0 S m cos 2 m m new: m N; n .82
m5: m 8.0 2 m one 2 m m 8.3m w is m 882
8.2 o 3.0 E s New 2 3 m 5.68 8 mod 6 882
one a NS 3 m 8.2 mm 3. m 94.5: m 3% s s2
8.: m cod 2 m o2. 2 m m 8.2 m as m 882
8.: m cod 2 m 62 2 m m 888 m N2“ n so

8.: w m 8.0 S m o? 8 m m. 888 n . New. in ; ., . so

see as page es 4.5 me a as a 3 es s s .s

.55 no.3: saw “nose: 5.6 82am. new .833 :52 meow

.882 832282 a: ems use =86

 

.888.— E .5855 ...35 seems .8582 beam 388.6 m3 2%...

 

 

 

 

 

as: s 8.0 w w 8.4 s 2.3 8 :88 w 8.8 w 488
2.8 m z: 4_ m . 48.: o m 4 8.84m m a: m .88.
8.8 4 43 o m 88 s on m 4.42. m 8.8 4. 88
8.8 m 8.0 s m 8.8 2 mm m 8.84 4 mom m 888
2.8 m 8.0 s m as o m m 3.8m m 8.8 m 488
3.8 m 8.0 E m 8.6 S on.“ m 8.8 m 2.8 4 888
488 s 86 M: n 8.8 2 m m of _m n 44.8 m 848
_48 4 2: 8 o 8.2 t o o 8.88 o O48 4 488
8.2 2 86 8 : 8.8. 8 :6 m 828 z _ 04.8 o .88
$8 m 8.0 _ _ 4 8.4 o 4.8 m $64M 4 8.8 8 4.8
8.2 w of 2 m m: 2 on m 2.48 m 8.8 m 488
8.8 m 3.0 2 m 3.4 2 s , 4 888. m. :8 m 4.8
82 s :3 2 w 86 2 we m 24.8 m m; s .88
8.8 o :3 M: m 86 2 E n 5:8 8 8.8 m 88
8.2 m 8.0 o o 8.4 o E m 845; o 8.8 4 88
8.2 8 Rd : w 43 : w m :88 s 8.8 8 s8
8.: o 8.0 t s 43 2 s m 2848 m 8.8 m 888
8.2 S 8.0 8 m :8 8 :88 m 8.48 m w _ .m 8 s8
43: o 8.0 2 s New 2 w.» m 8.448 s 8.8 o 888
mass 85 £45 ms 8> me a 56> a a 8> a s s

Esm 82.3: 5% .885 saw 823: .55 8833 :52 meow

.882 5:48.82 ,5. BE. 6: :86

 

APEoUV who—33A 3:. ..SmmEoU 8:55 sum—m: .8339: 58am Samoan—6 mfib 033—.

160

 

.85.. ..o .8 6:88 I BEE—emu. .oz
2.2:: no .8 3.8.. I ©253wa
”Boz

 

 

 

 

 

 

3.8 m 8.0 o m as o m m 8.44 m 8.8 m 4882

. .. 6.8.. 4.. 85 s. m . as o m. m 3.4.24 . m 8.8 ..m . $84 -
t .48 m z: 2 m 43 _ a m 4 8.88 m a: m .88
8.48 m 8.0 o n 8.8 w m m 2.84 4 mom m .88
8.8 4 43 s m 8.4. s m m 84.84 m 8.8 m 888
8.8 m 8.0 s m as o m m 488 m 8.8 m 488
8.8 m z: 42 m 48.: o m 4 8.84 m a: m 4.48
2.8 . m moo s m :8 o m m 8.8 m 8.8 m .88
8.8 m 86 s m 5.6 m m m 888 m 8.8 m .88
8.8 m 8.0 s m go o 2 m 48.54 m 8.8 m 488
8.48 m 8.0 s m as o m m 8.8 m 8.8 m .88
8.8 m 8.0 : 6 $8 .6 o 4 8.8 m :18 m .88
8.2 m 5.0 w 4 88 m o4 N 268 4 48 4 488
2.8 s 8.0 s w 8.6 w w 4 2.88 w :8 o .88
8.8. 4 8.0 2 m 43 4_ m m 888 m m48 4 .88
8.8 m Rd 8 m 8.8 w m m 8.84 4 88 m .88
A485 =o> mes: me 56> me E no> 3 E no.» E s s
55m 4.8ng 55 none: .55 among: .55 .833 :52 meow

882 5:26.864. by HE. o: :85

 

.8258 88.84 ssh 563:8 ...35 seems asses: beam sousnSm as 2.3

161

7.5. Summary

This chapter evaluated the vehicle control algorithm that was presented in Chapter 6 by
assessing the impact of three influence factors on its operation. These factors were:
wireless communication delay, number of preceding vehicles considered, and the
proportion of manually controlled vehicles within the traffic stream. For each of the
impact factors, three scenarios were simulated to measure vehicle control performance in
responding to specific traffic situations: acceleration from standstill to equilibrium,
platoon lead vehicle emergency braking, and platoon lead vehicle involved in brick wall
collision. For these evaluations, all the simulation parameters and procedure were
detemlinistic with respect to wireless communication delay and number of preceding
vehicles considered in vehicle control logic. This meant that a single simulation run
could be considered parameter setting. When considering mixed vehicle types, a
stochastic procedure was used to determine whether an individual vehicle was manually
controlled or controlled using the proposed automated algorithm. In these cases, 5
simulation runs were executed first for each scenario evaluated, with an additional set of

5 runs conducted when necessary.

7. 5.1. Impacts of Wireless Communication Delay

The evaluations of automated vehicle control in the presence of wireless communication
delay showed relatively strong impacts in the standstill to equilibrium and emergency
braking scenarios. For the standstill to equilibrium scenario, the duration to reach
equilibrium with 300 ms information delay is more than twice the observed time to

equilibrium for non-delay situations. This is primary due to the fact that the 300 ms

162

delay introduces errors in the traffic data upon which vehicle control decisions are based.
In addition to the duration to reach equilibrium, equilibrium speed is also found to be
lower in scenarios with communication delays. This is again explained by errors
introduced in the data used for vehicle control, which lead vehicle controllers to maintain

larger time headways between vehicles by slowing down the vehicles.

For the emergency braking scenario, no safety critical situation occurred and only two
surrogate measures, TTC and acceleration noise, were employed to evaluate the impact of
information delay. While the analysis acceleration noise clearly indicates a negative
impact of information delay on the stability of vehicles, the TTC values unexpectedly
showed slightly better performance in the presence of information delay. The
contradictive part of this result is attributed to weights assigned to the errors regarding
vehicle spacing and the speed of the lead vehicle, which are oppositely impacted by
information delay during braking events. In this case, the additional safety margins
pushed by vehicle spacing errors are more prominent than the effects of sensing with a
certain delay a deceleration by the lead vehicle. As a result, increases in information

errors tend to lead to higher TTC values.

For the same reason of as those invoked for vehicle spacing and speed errors, simulation
results for the brick wall collision scenario showed no significant change in the number
of collisions and the speed at which the collisions occur with or without communication
delay. However, both the TET and TIT parameters showed that vehicles experienced
longer and more imminent safety threats when the information they receive is delayed by
300 ms, clearly indicating the deterioration of overall platoon performance caused by
information delay.

163

7.5.2. Impacts of Multiple Lead Vehicles

The automated vehicle control logic was evaluated by varying the number of downstream
lead vehicles considered in the decision control algorithm from 1 to 5. Results of the
standstill to equilibrium scenario clearly indicated that considering multiple leaders in
both the presence and absence of information delay situations is beneficial. Without
information delay, the duration to reach equilibrium reduced fiom 331 s in single-lead
vehicle following cases to about 80 s in cases in which multi-lead vehicle are considered.
When 300 ms information delay was considered, the duration to reach equilibrium
reduced from 684 s in single-lead vehicle following cases to about 142 s in multi-lead
vehicle cases. Equilibrium speed appears independent of the number of downstream
vehicles and only relates to the amount of communication delay, 15 m/s for non—delay

cases and 11.5 s for with-delay cases.

For the emergency braking scenario, considering more than one downstream vehicle led
to higher TTC values and lower acceleration noises, indicating both safety and stability
improvements. It was further observed by comparing the assessed acceleration noise for
individual vehicles shown in Figures 7.2 and 7.4 that considering two downstream
vehicles with the presence of 300ms information delay may allow the first 24 following
vehicles in platoon to experience less acceleration noise than situations in which control
decisions are based on information from a single lead vehicle without information delay.
This is another strong indication of potential safety benefits of designing vehicle control
system considering information from more than one vehicle ahead in an environment in

which information delay may exist.

164

For the brick wall scenario, considering more than one lead vehicle resulted in a
reduction of the number of vehicles colliding from 12 to 2, when the 300 ms information
delay was held constant. Surrogate safety measures, including TTC, TET, TIT and CT F
were also improved with multiple lead vehicles considered in vehicle control logic.
Except for TIT, which showed the best mark when four lead vehicles were considered,
other measures indicated the most significant improvements when having only two lead
vehicles were considered in the vehicle control. CJF considerably deteriorated when data
from multiple lead vehicles were incorporated in decision making, indicating that the
vehicle control algorithm still needs improvement to reduce the chattering effects of

sliding mode control around its sliding surface.

7.5.3. Impacts of Manually Controlled Vehicles

In the standstill to equilibrium scenario, the impact of manually controlled vehicles was
studied for full combinations of proportion of automated vehicles and whether or not
there is communication delay. The averaged results of 5 simulation runs showed clear
trends for both duration to reach equilibrium and equilibrium speed. With 300 ms
communication delay, both measures showed almost linear relationships with respect to
the level of market penetration of automated vehicles in the traffic stream. The duration
to reach equilibrium decreased from 321 s with all vehicles being manually controlled to
136 s with all vehicles being automatically controlled. On the other hand, the equilibrium
speed showed a growing trend, from 8.4 m/s with all vehicles manually controlled to
11.56 m/s with all vehicles automatically controlled. For the non-delay cases, the
durations to equilibrium are shorter than with—delay cases, and the equilibrium speed are

higher. While both measures still follow linear relationships with the market penetration

165

levels, more prominent improvement were observed in both cases when the market
penetration reached past 90%. This indicated that even a small proportion of manually
controlled vehicles may disturb the optimal flow of automated vehicles, delay the reach
of equilibrium, and prevent all vehicles from reaching the highest possible equilibrium

speed.

Two sets of simulations were executed for the emergency braking scenario, with the only
difference being the modeling of manually controlled vehicles. The Gipps car-following
model was used in the first set to simulate typical human driver behaviour. In the second
set, a modified automated vehicle control logic was used to simulate more aggressive
driving replicating more closely the behavior of the proposed automated control
algorithm. Results obtained using the Gipps model showed that both safety and vehicle
stability deteriorated with increasing number of automated vehicles. The major cause of
this unexpected result was that the manually controlled vehicles following the Gipps
model generally maintained headways greater than 2 s, which are more than twice the
headways typically maintained by the automated vehicles. When the emergency braking
event occurred, vehicle of different types react to the incident with substantially different
pre-event condition, thus masking the intrinsic property of vehicle control algorithm. In
the second set of simulations, the TTC values did not change substantially across different
market penetration levels and the acceleration noise associated with all vehicles only
reduced marginally for market penetration exceeding 90%. Results from this scenario
indicated that the subject vehicle control logic tends to maintain relative short headway to

vehicles ahead. Because of close following distances, the potential benefit of prompt

166

reactions to traffic events brought on by automated vehicle control are then likely to be

offset by the large headways maintained by certain type of human drivers.

For the brick wall scenario, levels of market penetration reaching and exceeding 90%
were found to reduce the number of crashes from 5 to 3. However, strong variations exist
in the number of crashes observed in the simulations at lower market penetration levels.
The major cause of increasing numbers of crashes when introducing small proportions of
automated vehicles in a traffic stream was linked to secondary accidents caused by the
difference in braking capabilities of manually controlled and automated vehicles. The
compilation of TTC values at the platoon level showed that increasing amounts of
automated vehicles typically led to reduced safety threats. This indicated that the most
imminent threats experienced by all platoon vehicles became less severe for all platoon
vehicles as a whole. For TET and TIT, the values of both parameters only showed
significant reductions for market penetration levels of 90% or higher. For the estimated
acceleration noise, cases with low market penetration levels were generally better than
fully automated situations. The relatively good performance at low market penetration
levels was again explained by the fact that manually controlled vehicles maintained much
larger headways than automated vehicles, thus allowing vehicles to react more smoothly

to changes in traffic conditions ahead.

167

 

CHAPTER 8

CONCLUSIONS

This study has attempted to research impacts of information delay on the performance of

wireless-based driver assistance applications with the underlying objective of providing

safer and smoother longitudinal control for passenger vehicles. Under the approach

specified in Chapter 4, the following research tasks have been performed:

E1

19

Review of professional literature to identify the latest developments of ITS
applications and wireless communication technologies.

Identification of potential problems with vehicle-based ITS applications and, in
particular, wireless-based driver assistance systems, in situations in which data
obtained through wireless transmission may be delayed due to heavy network
communication loads or other causal elements.

Development of a simulation-based testbed enabling the evaluation of wireless—
based vehicle control system performance.

Implementation of functions simulating communication delays within a
microscopic traffic simulation environment.

Development of a decision-making model for the automated control of vehicle
speeds in an environment in which communication delay may exist.

Evaluation of impacts of varying proportions of manually and automated
controlled vehicles in a traffic stream on the performance of the proposed vehicle

control algorithm.

168

8.1. Summary of Research

Existing vehicle—based ITS applications in commercial use include adaptive cruise
control systems and collision warning systems. Both applications rely on radar sensors to
collect information about downstream vehicles and are subject to limitations associated
with the technology being employed for their operation. Common technical restrictions
include unreliability on roadways with certain geometry characteristics, for instance on
sharp curves, and limited detection range. Besides these restrictions, each application is
also limited in its own way. For instance, adaptive cruise control systems are only
operable when a certain minimum vehicle speed and minimum time spacing is satisfied.
Typical headway settings also tend to be longer than the average time headway
maintained by human drivers, creating a situation in which vehicles equipped with
adaptive cruise control systems will have to constantly adjust their speed to vehicles
moving into their lane immediately ahead of them. Collision warning systems, on the
other hand, are experiencing difficulty in reducing false alarm rates when attempting to

reduce system response time when considering only the next vehicle in front.

A solution to improve both adaptive cruise control systems and collision warning systems
is to provide vehicles with information regarding vehicles that are located further
downstream through vehicle-to-vehicle wireless communications. This is the primary
aim behind the recent development of the 5.9-GHz DSRC standard. With vehicle-to-
vehicle communication becoming increasingly affordable, the accessibility to traffic data
characterizing vehicles beyond the one immediately ahead prompted the development of

cooperative adaptive cruise control and cooperative collision warning systems. Both

169

systems shall amend their predecessors and offer improvements if the potential problems

identified in Chapter 3 can be solved.

Potential problems addressed in this study include vehicle control decision-making
algorithm, impacts of wireless communications delay, and coexistence of vehicles with
multiple levels of automation. This study attempted to address these problems by
providing a simulation model, the Communication and Traffic Simulation Model
(CATSIM), developing a concept vehicle control decision-making logic, and finally
evaluating the vehicle control logic with the presence of wireless communication delay

and manually controlled vehicles.

The CATSIM model, which integrates traffic simulation processes with wireless data
communication functions and provides other functions for this study, was developed
based on the existing Intelligent Driver Model (IDM) microscopic traffic simulation
model. The capability to simulate both vehicle-to-vehicle data transmission and
communication delay was implemented in CATSIM through a unique vehicle data
storage structure, which stores vehicle data for multiple time steps and allows the
simulation engine to retrieve information characterizing the status of multiple vehicles in
the past. Additional simulation capabilities introduced in the model include roadways
consisting more than two lanes, new and replaceable car-following models that will
represent driver assistance systems, accident detection capability, extremely short and
variable simulation time step, and output capability of individual vehicle information

over the entire course of simulation.

170

The vehicle control algorithm that was developed in this research and introduced in the
CATSIM model was developed based on the principles of sliding-mode control. The
algorithm is capable of integrating information from multiple lead vehicles and can be
scaled down, if needed, to consider only information from a single vehicle to emulate the
operation of current radar-based vehicle control systems. The proposed algorithm was
evaluated by assessing the impacts of wireless communication delay, number of
preceding vehicles considered and surrounding presence of manually controlled vehicles

on vehicle control decisions.

8.2. Research Main Conclusions

The impacts of wireless communication delay were evaluated by simulating vehicles
behavior under specific control assumptions in situations with and without information
delay. The simulation results showed that wireless communication delay has relatively
strong impact on the performance of vehicle control systems, as detailed in the specific

result results detailed below:

:1 Longer duration to reach equilibrium. In the standstill to equilibrium scenario,
the duration to reach equilibrium with information delay was longer than
situations without delays. For the particular case showed in Chapter 7, the
duration to reach equilibrium with 300 ms information delay is more than twice
the observed time to equilibrium for non-delay situations.

Lower equilibrium speed. Information delay was also found to lead to lower
equilibrium speeds. In the particular case presented, vehicle equilibrium speed

was reduced from 15 m/s in the scenario without information delay to 11.5 s in

171

the scenario with information delay. This reduction is primary due to the
reduction in data accuracy introduced by communication delays.

6 Longer and more imminent safety threats. In the brick wall collision scenario,
both TET and TIT values clearly indicated that all vehicles in the simulated
platoon experienced longer and more imminent safety threats in the presence of
information delay than in situations without delay.

a Lower acceleration noise and reduced collision threats. In the emergency
braking scenario, trends regarding acceleration noise and TTC values were
contradictive.

E11 No impact on number of collisions and collision speed. Simulation results for
the brick wall collision scenario showed no significant change in the number of
collisions and the speed at which the collisions occur with or without

communication delay.

While the first three results listed above have been expected, the last two findings were
found to be contradictive to initial expectations. The expected results are explained by
errors introduced in the data used for vehicle control, which lead vehicle controllers to
maintain larger time headways between vehicles by slowing down the vehicles. The
contradictive results, however, were attributed to weights assigned to the errors regarding
vehicle spacing and lead vehicle speed, which are oppositely impacted by information
delay during braking events. Since the additional safety margins pushed by vehicle
spacing errors are more prominent than the effect of a delayed sensing in lead vehicle

deceleration, this explains why greater information delays lead to higher TTC values in

172

the emergency braking scenario and no impact on number of crashes and collision speed

in the brick wall collision scenario.

The automated vehicle control logic was then evaluated by varying the number of

downstream lead vehicles considered in the decision control algorithm from 1 to 5.

Considering multiple leading vehicles showed improvements over single vehicle

following in all simulation scenarios.

Ell

Shorter duration to reach equilibrium. Without information delay, the duration
to reach equilibrium reduced from 331 s in single-lead vehicle following cases to
about 80 s in cases in which multi-lead vehicle are considered. When 300 ms
information delay was considered, the duration to reach equilibrium reduced from
684 s in single-lead vehicle following cases to about 142 s in multi-lead vehicle
cases.

No impact on equilibrium speed. Equilibrium speed appears independent of the
number of downstream vehicles and only relates to the amount of communication
delay, 15 m/s for non-delay cases and 11.5 s for with-delay cases.

Lower acceleration noise and reduced collision threats. For the emergency
braking scenario, considering more than one downstream vehicle led to higher
TTC values and lower acceleration noises, indicating both safety and stability
improvements. It was further observed that, even with the presence of 300ms
information delay, considering two downstream vehicles may allow the first 24
following vehicles in platoon to experience less acceleration noise than situations
in which control decisions are based on information from a single lead vehicle

without information delay.

173

:1 Reduced number of collisions and reduced safety threats. For the brick wall
scenario, considering more than one lead vehicle resulted in a reduction of the
number of vehicles colliding from 12 to 2, when the 300 ms information delay
was held constant. Surrogate safety measures, including TTC, TET, TIT and CT F
were also improved with multiple lead vehicles considered in vehicle control

logic.

These results strongly indicate potential operational and safety benefits of designing
vehicle control system considering information from more than one vehicle ahead in an
environment in which information delay may exist. Despite positive results, negative
results were also observed in the brick wall collision scenario. A vehicle performance
measure, CJF, considerably deteriorated when data from multiple lead vehicles were
incorporated in decision making, indicating that the vehicle control algorithm still needs
improvement to reduce the chattering effects of sliding mode control around its sliding

surface.

The third influence factor evaluated in this study was the presence of manually controlled
vehicles. The impacts of this factor on the performance of vehicle control systems are

summarized as follows.

6 Longer duration to reach equilibrium with increased level of manually
controlled vehicles. In the standstill to equilibrium scenario, the impact of
manually controlled vehicles was studied for full combinations of proportion of
automated vehicles and whether or not there is communication delay. The

averaged results of 5 simulation runs showed that with 300 ms communication

174

delay, the duration to reach equilibrium showed almost a linear relationship with
respect to the level of market penetration of automated vehicles in the traffic
stream. The duration to reach equilibrium decreased from 321 s with all vehicles
being manually controlled to 136 s with all vehicles being automatically
controlled. For non—delay cases, the durations are shorter than with-delay cases
and still follow linear relationship with the market penetration levels.

Lower equilibrium speed with increased level of manually controlled
vehicles. The equilibrium speed also showed an almost linear growing trend,
from 8.4 m/s with all vehicles manually controlled to 11.56 m/s with all vehicles
automatically controlled. For the non-delay cases, the equilibrium speeds are
higher and also follow linear relationships with the market penetration levels.
Duration to reach equilibrium and equilibrium speed trends are more
prominent when market penetration is above 90%. This indicated that even a
small proportion of manually controlled vehicles may disturb the optimal flow of
automated vehicles, delay the reach of equilibrium, and prevent all vehicles from
reaching the highest possible equilibrium speed.

Deteriorated safety and vehicle stability with increasing number of
automated vehicles. In the emergency braking scenario, simulation results
indicated that the subject vehicle control logic tends to maintain relative short
headway to vehicles ahead and because of close vehicle following, the potential
benefit of prompt reaction to traffic event brought by automated vehicle control
are likely to be offset by large headways maintained by certain type of human

drivers.

175

Reduced number of crashes when market penetration reaching and
exceeding 90%. For the brick wall scenario, levels of market penetration
reaching and exceeding 90% were found to reduce the number of crashes from 5
to 3. However, strong variations exist in the number of crashes observed in the

simulations at lower market penetration levels.

Strong variations involved in the last result called for further attention. The major cause
of increasing numbers of crashes when introducing small proportions of automated
vehicles in a traffic stream was linked to secondary accidents caused by the difference in
braking capabilities of manually controlled and automated vehicles. The compilation of
TTC values at the platoon level showed that increasing amounts of automated vehicles
typically led to reduced safety threats. This indicated that the most imminent threats
experienced by all platoon vehicles became less severe for all platoon vehicles as a
whole. For TET and TIT, the values of both parameters only showed significant
reductions for market penetration levels of 90% or higher. For the estimated acceleration
noise, cases with low market penetration levels were generally better than fully
automated situations. The relatively good performance at low market penetration levels
was again explained by the fact that manually controlled vehicles maintained much larger
headways than automated vehicles, thus allowing vehicles to react more smoothly to

changes in traffic conditions ahead.

8.3. Future Research Directions

While the results of the study clearly show the benefits of designing vehicle control

systems considering information from at least two lead vehicles and developing

176

algorithms considering the potential impacts of information delay, it is suggested that
further research be conducted to obtain deeper understanding of the problem. In
particular, the study did not look at lane changing. The addition of a lane changing
module in the simulation model would allow the testing of more complex vehicle control
algorithms, including lane—changing functions. While it is anticipated that automated
vehicle controls with both longitudinal and latitude functions are likely to be more
promising than longitudinal control alone in promoting safety improvement on multi-lane
roadway segments, uncertainty remains regarding the impact of information delay on lane

changing safety.

A final element that should be explored is whether variations in the behavior of preceding
vehicles would affect the main results of the study. The current study essentially
assumed that all vehicles would behave similarly. In reality, drivers exhibit slightly
different behavior that may translate into random accelerations and decelerations. Any
robust vehicle control algorithm should then be able to identify these variations in

behavior and ignore them.

177

REFERENCES

ARIB (2004). Standards for Radio Systems in the Field of Telecommunications.
Association of Radio Industries and Businesses, Japan. http://www.arib.or.jp/english
/html/overview/st_e.html. Accessed November 1, 2004.

Asher, HI, and Galler, BA. (1996). “Collision Warning using Neighboring Vehicle
Information.” Proceedings of the I 996 Annual Meeting of ITS America, Washington,
DC, pp. 674-684.

Behringer, R., Holt, V., and Dickmanns, D. (1992). “Road and Relative Ego-state
Recognition.” Proceedings of IEEE Intelligent Vehicles '92 Symposium,Detroit, MI.,pp.
385-90.

Bengtsson, J. (2001). Adaptive Cruise Control and Driver Modeling. MS. Thesis,
Department of Automatic Control, Lund Institute of Technology, Lund, Sweden.

Biswas, S., Tatchikou, R., and Dion, F. (2005). Vehicle-to-Vehicle Packet Forwarding
Protocols for C00perative Collision Avoidance. Proceedings of IEEE Globecom 2005
Conference, St. Louis, MO.

Bose, A., and Ioannou, P. (2001). “Evaluation of the Environmental Effects of Intelligent
Cruise Control Vehicles.” Transportation Research Record 1774, Transportation
Research Board, Washington, DC, pp. 90—97.

Bose, A., and Ioannou, P. (2003). “Analysis of Traffic Flow with Mixed Manual and
Semiautomated Vehicles.” IEEE Transactions on Intelligent Transportation Systems,
Vol. 4, No. 4.

Caudill, R.J., and Garrard, W.L. (1977). “Vehicle-Follower Longitudinal Control for
Automated Transit Vehicles.” Journal of Dynamic Systems, Measurement, and Control,
Vol. 99, No. 4, pp. 241—248.

Chang, M.S., Messer, CI, and Santiago, A]. (1985). “Timing Traffic Signal Change
Intervals Based on Driver Behavior.” Transportation Research Record 1027,
Transportation Research Board, Washington, DC, pp. 20—30.

CORDIS (2005). http://www.cordis.lu/telematics/tap_transport/research/projects/chauff
eur.html. Accessed on September 17, 2005.

Dailey, D.J., Ham, P., and Lin, P-J. (1996). [T S Data Fusion. Final Research Report
prepared for the Washington State Department of Transportation, University of
Washington, Seattle, WA.

Davis, LC. (2004). “Effect of adaptive cruise control systems on traffic flow.” Physical
Review E, Vol. 69, No. 6.

178

Eaton Voard (2004). http://www.roadranger.com/csee/MungoBlobs/vole950-0600.pdf,
accessed on April 21, 2004.

Foster, ER. (1979). “A Constant Separation Control Law for Use in Automated
Guideway Transsit.” Proceedings of the 29th IEEE Vehicular Technology Conference,
Arlington Heights, IL, pp. 358-361.

Gettman, D., and Head, L. (2003). Surrogate Safety Measures From Traffic Simulation
Models. Report FHWA-RD-03-050, Federal Highway Administration, US. Department
of Transportation, McLean, VA.

Girard, A.R., Sousa, J.B., Misener, J.A., and Hedrick, J.K. (2001). “A Control
Architecture for Integrated Cooperative Cruise Control and Collision Warning Systems.”
Proceedings of IEEE Conference on Decision and Control, Orlando, FL.

Hall, R.W. (2003). Handbook of Transportation Science, 2nd Edition, Kluwer Academic
Publishers.

Hall, D.L., and Linn, RJ. (1991). “A Taxonomy of Algorithms for Multisensor Data
Fusion.” Proceedings of the 1990 Tri-Service Data Fusion Symposium, pp. 13-29.

Hedrick, J.K., Godbole, D., Rajamani, R., and Seiler, P. (1999). Stop and Go Cruise
Control. Final Report, California PATH Research Program, University of California,
Berkeley, CA.

Hirst, S., and Graham, R. (1997). “The Format and Presentation of Collision Warning.”
Ergonomics and safety of Intelligent Driver Interfaces, Mahwah, NJ. Editor, Lawrence
Erlbaum Associates, pp. 203-219.

Horowitz, A.D., and Dingus, T. A. (1992). “Warning Signal Design: A key Human
Factors Issue in an In-vehicle Front-to-rear-end Collision Warning System.” Proceedings
of the Human Factors Society 36'” Annual Meeting, Santa Monica, CA, pp. 1011-1013.

Hyde’n, C. (1996). Traffic Conflicts Technique: State of the Art. Transportation
Department, University of Kaiserslautem, Kaiserlautem, Germany.

Lear (2004). “Adaptive cruise control.” Advantage Online, I-Car E-newsletter.
http://www.i-
car.com/html_pages/about_icar/current_events_news/advantage/advantage_online_archiv
es/2004/02l604.html. Accessed on July 12, 2004.

IEEE (2002). IEEE 802.16 Backgrounder. http://ieee802.org/16/pub/backgrounder.html.
Accessed May 24, 2002.

IEEE (2004). 5.9 GHz Dedicated Short Range Communication Overview.
http://grouper.ieee.org/groups/scc32/dsrc/home_popup/brief_imgs/DSRC_Tutorial.zip.
Accessed April 21 , 2004.

179

IEEE (2007a). Keeping Cars from Crashing. http://www.spectrum.ieee.org/print/14l8.
Accessed March 20, 2007.

IEEE. (2007b). Status of Project IEEE 802.11p. http://grouper.ieee.org/groups/802/1l/
Reports/tgp_update.htm. Accessed March 20, 2007.

Ioannou, P.A., and Chen, CC. (1993). “Autonomous intelligent cruise control.” IEEE
Transactions on Vehicular Technology, Vol. 42, No. 4, pp. 657-672.

Jeannotte, K., Chandra, A., Alexiadis, P., and Skabardonis, A. (2004). Traffic Analysis
Toolbox, Volume 11: Decision Support Methodology for Selecting Traffic Analysis
Tools. Report FHWA-HRT-04-039. Federal Highway Administration, US. Department
of Transportation, Washington. DC.

Jones, T.R., and Potts, RB. (1962). “The Measurement of Acceleration Noise — A Traffic
Parameter.” Operations Research, Vol. 10, No. 6, pp. 745 — 763.

Kato, S., Tsugawa, S., Tokuda, K., Matsui, T., and Fujii, H. (2002). “Vehicle Control
Algorithms for Cooperative Driving With Automated Vehicles and Intervehicle

Communications.” IEEE Transactions on Intelligent Transportation Systems, Vol. 3, No.
3.

Kerner, BS. (2003). “Control of spatial-temporal congested traffic patterns at highway
bottlenecks.” Presented at the 84'“ Annual Meeting of the Transportation Research
Board, Washington, DC, USA.

Kessaci, A., Farges, J.L., and Henry, 1.]. (1989). “On Line Estimation of Turning
Movements and Saturation Flows in PRODYN.” Control, Computers, Communications
in Transportation. Paper from the IFA C/IFIP/IFORS Symposium, Paris, France.

Leader, S. (2004). DSRC: A Short-range Wireless Technology for Telematics
Applications. http://www.coe.montana.edu/ee/rwolff/shel%20‘leader%20dsrc.pdf.
Accessed July 12, 2004.

Lee, K., and Peng, H. (2002) A Longitudinal Human Driving Model for Adaptive Cruise
Control Performance Assessment. Proceedings of IMECE 2002: Symposium on
Advanced Automotive Technologies, New Orleans, LA.

Lerner, N.D., Kotwal, B.M., Lyons, RD, and Gardner-Bonneau, DJ. (1993).
Preliminary Human Factors Guidelines for Crash Avoidance Devices. Technical Report
DTNH22-91-C-07004, National Highway Transportation Safety Agency, US.
Department of Transportation, Washington, DC.

Liang, C., and Peng, H. (1999). “Optimal adaptive cruise control with guaranteed string
stability: Vehicle System Dynamics, Vol. 32, No. 4, pp. 313—330.

Luckscheiter, K. (2004). “Vehicle and Infrastructure Integration for Michigan.” Presented
at 9" ITS Michigan Conference, Detroit, Michigan.

180

MeshNetworks (2004). http://www.meshnetworks.com. Accessed May 18, 2004.

Minderhoud, M., and Bovy, P. (2001). “Extended Time-to-collision Measures for Road
Traffic Safety Assessment.” Accident Analysis and Prevention, Vol. 33, pp. 89-97.

Morsink, P.L.J, Cseh C., Gietelink, O.J., and Miglietta, M. (2003). Preliminary Design of
an Application for Communication Based Longitudinal Control in the CarTALK2000
Project.

Moslehi, F. (1998). “Wireless Connectivity.” World Wide Web — Beyond the Basics.
Marc Abrams, Editor, Prentice Hall.

NHTSA (2003). Automotive Collision Avoidance System Field Operational Test, ACAS
IFOT Third Annual Report, Report DOT HS 809 600, National Highway Traffic Safety
Administration, US. Department of Transportation, Washington, DC, USA.

NHTSA (2005). Traffic Safety Facts 2003. Report DOT HS 809 775, National Highway
Traffic Safety Administration, US. Department of Transportation, Washington, DC,
USA.

NT CIP Online Resource. Website developed by AASHTO, ITE and NEMA.
http://www.ntcip.org. Accessed November 1, 2004.

NTSB (2001). Special Investigation Report: Vehicle and Infrastructure Based
Technology for the Prevention of Rear-End Collisions. National Transportation Safety
Board, Washington, DC.

Pietrzyk, M.C., and Mierzejewski, EA. (1993). “Electronic T011 and Traffic

Management Systems.” Synthesis of Highway Practice 194, Transportation Research
Board, National Research Council, Washington, DC.

Pline, J .L. (1999). Traffic engineering handbook. 5th Edition, Institute of Transportation
Engineers, Washington, DC.

Patterson, K.A. (1998). Intelligent Cruise Control System Impact Analysis. MS. Thesis,
Virginia Polytechnic Institute and State University, Blacksburg, VA.

RoadRanger. (2000). http://www.truck.eaton.com/cgi-bin/eaton/news_releases/view_
article.pl? 80. Accessed on March 1, 2005.

Road Transport and Traffic Telematics (2004). Official Web Site. Technical Committee
278, European Committee for Standardization. http://www.nen.nl/cen278l. Accessed
November 1, 2004.

Robinson, R. (2004). “VH Use Case Set.” Presented at the Transportation Research
Board Workshop on Research Needed to Support Vehicle-Infrastructure Cooperation,
Troy, Michigan.

181

Schrank, D., and Lomax, T. (2004). The 2004 Annual Urban Mobility Report. Texas
Transportation Institute, Texas A&M University, College Station, TX.

Shladover, S., VanderWerf, J., Miller, M.A., Kourjanskaia, N., and Krishnan, H. (2001).
Development and Performance Evaluation of AVCSS Deployment Sequences to Advance
from Today's Driving Environment to Full Automation. Report UCB-ITS-PRR-2001-18,
California PATH Program, University of Califormia, Berkeley, CA.

Swaroop, D. and Rajagopal, KR. (1998). “Intelligent cruise control systems and traffic
flow stability.” Transportation Research C, Vol. 7, No. 6, 329-352.

Tokuda, K. (2004). “DSRC-Type Communication System for Realizing Telematics
Services.” Oki Technical Review, Vol. 71, No. 2, pp. 64-67.

Touran, A., Brackstone, M., and McDonald, M. (1999). “A Collision Model for Safety
Evaluation of Autonomous Intelligent Cruise Control.” Accident Analysis and
Prevention, Vol. 31, pp. 567—578.

Treiber M., Hennecke, D., and Helbing, D. (2000). “Congested traffic states in empirical
observation and numerical simulations.” Physical Review E, Vol. 62, pp. 1805-1824.

Treiber, M., and Helbing, D. (2002). Microsimulations of freeway traffic including
control measures. eprint arXivzcond-mat/0210096.

USDOT. (2003). National ITS Architecture, Version 5.0. CD-ROM., US. Department of
Transportation, Washington, DC.

USDOT. (2006). Private Sector Deployment of Intelligent Transportation Systems:
Current Status and Trends. US. Department of Transportation, Washington, DO

Wilson, T., Butler, W., McGehee, D.V., and Dingus, TA. (1996). “IVHS
Countermeasures for Rear-end Collisions, Driver Warning System Performance
Guidelines.” Proceding., 1996 Annual Meeting of ITS America, Washington, DC, 949-
957.

Yamashita, T., Izurrri, K., Kurumatani, K, and Nakashima, H. (2005). “Smooth Traffic
Flow with a Cooperative Car Navigation System.” Proceedings of the 4” International
Joint Conference on Autonomous Agents and Multiagent Systems, ACM Press, NY, pp.
478—485.

Yanakiev, D. and Kanellakopoulos, I. (1995). Longitudinal control of heavy-duty

vehicles for automated highway systems. Proceedings of the 1995 American Control
Conference, Seattle, WA, pp. 3096-3100.

Zheng, P., and McDonald, M. (2004). “Collision Warning Timing Based on Normal
Braking Behavior.” Proceedings of the 83'“ Annual Meeting of the Transportation
Research Board, Washington, DC.

182

Zhu, J., and, Roy, S. (2003). “MAC for dedicated short range communications in
intelligent transport system.” IEEE Communications Magazine, Vol. 41, No. 12, pp. 60-
67.

183