Content uploaded by Jiaqi Ma
Author content
All content in this area was uploaded by Jiaqi Ma on Apr 16, 2020
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=ttra21
Transportmetrica A: Transport Science
ISSN: 2324-9935 (Print) 2324-9943 (Online) Journal homepage: https://www.tandfonline.com/loi/ttra21
Leveraging existing high-occupancy vehicle
lanes for mixed-autonomy traffic management
with emerging connected automated vehicle
applications
Yi Guo & Jiaqi Ma
To cite this article: Yi Guo & Jiaqi Ma (2020) Leveraging existing high-occupancy vehicle lanes
for mixed-autonomy traffic management with emerging connected automated vehicle applications,
Transportmetrica A: Transport Science, 16:3, 1375-1399, DOI: 10.1080/23249935.2020.1720863
To link to this article: https://doi.org/10.1080/23249935.2020.1720863
Published online: 16 Apr 2020.
Submit your article to this journal
View related articles
View Crossmark data
TRANSPORTMETRICA A: TRANSPORT SCIENCE
2020, VOL. 16, NO. 3, 1375–1399
https://doi.org/10.1080/23249935.2020.1720863
Leveraging existing high-occupancy vehicle lanes for
mixed-autonomy traffic management with emerging
connected automated vehicle applications
Yi Guo and Jiaqi Ma
Civil and Architectural Engineering and Construction Management, University of Cincinnati, Cincinnati, USA
ABSTRACT
Transportation agencies have started including phased deployment
of connected and automated vehicle (CAV) technologies in their
transportation plans and programs. While theoretical analyses have
indicated significant benefits of CAVs for improving system perfor-
mance, deploying these technologies at existing or adapted highway
facilities concerns more than technological issues. This study intro-
duces the concepts and operational strategies of using managed
lanes, High-Occupancy Vehicle (HOV) lanes in particular, for mixed-
autonomy traffic management. The study enhances existing and
develops new algorithms for three selected freeway CAV applications
as a CAV technology integration, including speed harmonization,
cooperative adaptive cruise control (CACC), and cooperative merge.
Instead of evaluation on a synthetic segment, this study reports
a large-scale simulation-based real-world case study to investigate
CAV early deployment opportunities on Interstate 66 outside the
Beltway. The simulation results show that, for all scenarios, individ-
ual and bundled CAV applications can significantly improve traffic
performance in terms of delay and throughput. CACC platooning is
the most effective individual strategy to improve traffic performance.
The bundled CAV applications can benefit the system, even with low
CAV market penetration, and fully handle more than 130 percent of
the existing demand with high CAV market penetration rates. Addi-
tionally, left-side dedicated ramps and shared managed lane oper-
ational strategies are also beneficial, even during early deployment
stages.
ARTICLE HISTORY
Received 31 May 2019
Accepted 3 November 2019
KEYWORDS
Connected automated
vehicles (CAV); managed
lane; high-Occupancy vehicle
(HOV); cooperative adaptive
cruise control (CACC);
cooperative merge; speed
harmonization; dedicated
ramp
Introduction
Oversaturation occurs when more vehicles use a road than the capacity (i.e. demand is
greater than supply). Regardless of its original cause(s), excess congestion leads to unsta-
ble traffic flow, which is followed soon by breakdowns and bottlenecks. These congestion
events bring unwanted impacts, including wasted fuel, increased emissions, decreased
travel time reliability, delays to emergency vehicles, and higher accident risk.
Emerging connected and automated vehicle (CAV) technologies offer promising and
flexible solutions to traffic congestion. CAVs may be equipped with a suite of technologies
CONTACT Jiaqi Ma jiaqi.ma@uc.edu 765 Baldwin Hall, Cincinnati, OH 45221-0071
This material is published by permission of the Next Mobility Lab, operated by the University of Cincinnati, and produced for the US Depart-
ment of Transportation under Contract No. DTFH6116D00030. The US Government retains for itself,and others ac ting on its behalf,a paid-up,
non-exclusive, and irrevocable worldwide license in said article to reproduce, prepare derivative works, distribute copies to the public, and
perform publicly and display publicly, by or on behalf of the Government.
1376 Y. GUO AND J. MA
ranging from preliminary vehicle automation (i.e. automated decision making at the vehi-
cle level using onboard sensors) to a connected ecosystem in which vehicles and roadside
infrastructure communicate wirelessly among themselves. Since managed lane facilities are
often equipped with traffic sensors, communication networks, and tolling equipment (de
Palma and Lindsey 2011; Yin and Lou 2009; Zhang, Yan, and Wang 2009), which can be
leveraged to support CAV initiatives, managed lane operators have a unique opportunity
to leverage their facilities as early deployment sites for CAV to improve the traffic system
performance.
The motivation of this study is to investigate the effectiveness of CAV deployment in
enhancing existing traffic system performance with different managed lane operational
strategies. This study aims to evaluate the potential benefits of implementing CAV appli-
cations in existing or upgraded facilities and determine appropriate managed lane opera-
tional strategies correspondingly. Since more than three dozen CAV application concepts
have been developed (Guanetti, Kim, and Borrelli 2018; Elliott, Keen, and Miao 2019; ITSJPO,
2019), three of them are applied in this study due to their effectiveness (Guo et al. 2019;
Learn et al. 2017; Letter and Elefteriadou 2017;Maetal.2017; Malikopoulos et al. 2018;
Naus et al. 2010; Ntousakis, Nikolos, and Papageorgiou 2016; Shladover, Su, and Lu 2012;
Zhou, Li, and Ma 2017): cooperative adaptive cruise control (CACC), speed harmonization,
and cooperative merge.
CACC is able to reduce traffic congestion by improving highway capacity and through-
put and attenuating traffic flow disturbances. Since ACC has had little effect on lane capacity
(Milanés et al. 2013; Shladover, Su, and Lu 2012), the CACC systems could allow the mean
following time gap to decline from about 1.4 s when driving manually to approximately
0.6 s with utilizing vehicle-to-vehicle (V2 V) communication (Nowakowski Christopher et al.
2010), resulting in an increase in highway lane capacity.
Speed harmonization involves gradually lowering speeds upstream of a heavily con-
gested area in order to reduce the stop-and-go traffic, which may also be used to reduce
vehicle speeds, either delaying or preventing the onset of traffic congestion. The simula-
tion results in previous literature found significant travel time reductions (e.g. a 10-percent
reduction corridor-wide and a 35-percent reduction on localized bottleneck segments) at
CAV penetration rates of 10 percent or higher, which concurred with other simulation-
based studies (Talebpour, Mahmassani, and Hamdar 2013).
Cooperative merge leverages V2 V and V2I communications to enable CAVs to safely
merge into the traffic with less impact on mainline traffic (Chou, Shladover, and Bansal 2016;
Raboy et al. 2017). It helps CAV to identify upcoming acceptable gaps or signal mainline traf-
fictocreateanacceptablegapbycooperatingwitheachother.Chouetal.(2016) tested two
cooperative automated merging strategies with I2 V and V2V correspondingly. The results
show that the I2 V case reduced travel time in the merging section when the traffic flow
was high, and the V2 V case supports a significant increase in traffic flow without increasing
travel times. The results indicate the potential advantages of using cooperative automation
to relieve the bottleneck in the merging section. A recent Federal Highway Administra-
tion (FHWA) study (Raboy et al. 2017) also successfully validates a cooperative lane change
maneuver driven by a proof-of-concept field experiment.
Based on the previous simulation studies in a single managed lane, the bundled applica-
tion of these three aforementioned CAV applications can further improve the traffic perfor-
mance with existing highway facilities as compared with using individual CAV application
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1377
(Ma et al. 2018). These vehicle connectivity and automation technologies could allow con-
sistent speeds to be maintained throughout the facility, thereby increasing the traffic
demand levels that the roadway can support. The increased throughput and capacity of
the managed lanes could also potentially benefit parallel general-purpose lanes as traffic
shifts to the managed lane. Smoothed, optimized speeds would also create a reduction in
fuel consumption, harmful emissions, and highway crashes.
Litman (2017) indicates that the CAV may not be predominant in traffic until the 2040s
to 2050s; therefore the traffic flow will consist of conventional vehicles, connected vehi-
cles (CVs) and CAVs in a long period. In this transition period, CAVs and CVs can benefit
the traffic system performance with gradually increasing market penetrations before the
fully-autonomy (i.e. all presented vehicles are CAV) is achieved; therefore the traffic system
can be regarded as a ‘mixed-autonomy’ system since conventional vehicles coexist with
CVs and CAVs. To ensure a smooth transition from conventional human-driven traffic to
mixed-autotomy traffic with gradually increasing market penetration rate of CVs and CAVs,
the alternative operational strategies are also worthy of investigation to adapt the changes
to realize CAV benefits, especially at the early deployment stages. For example, based on
the results in the previous study (Ma et al. 2018), low market penetration of CAV may not
contribute to traffic performance enhancement. This is mainly because conventional vehi-
cles cannot cooperate with CAVs and involve stochastic driving behaviors that may disturb
CAVs’ operations. To resolve this issue, one possible alternative strategy is to create a ‘high-
penetration’ environment by offering CAVs eligibility to using existing managed lanes such
that CAVs can concentrate at a certain part of the traffic stream, interact and platoon with
each other, and therefore positively impact the traffic system performance.
The purpose of this paper is to investigate deployment opportunities of CAV tech-
nologies and alternative managed lane strategies on existing and adapted facilities in the
mixed-autonomy traffic system. This study conducts simulations on a real-world network
to examine the effectiveness of CAV deployment and corresponding managed lane oper-
ational strategies in enhancing traffic system performance. Both infrastructure operational
and CAV technological strategies are simulated, evaluated, and discussed. On the CAV tech-
nological side, this study evaluates the effectiveness of each individual CAV application
and their combinations. On the infrastructure operational side, the case study evaluates
the potential benefits of dedicated ramps and a realistic management lane concept –
CV/CAV eligible HOV lanes – where CVs, CAVs, and HOVs (which can be human-driven or
CV/CAV) can access the left-side managed lane. The remainder of this paper is organized
as follows: Section 2 provides detailed descriptions of the technologies and operational
strategies. Section 3 covers the simulation experiments and results, including the model
calibration, design of simulation, simulation results for different scenarios, and discussion
of implications. Section 4 concludes this paper and proposes future research topics.
CAV applications and operational strategies
Cooperative adaptive cruise control/platooning
The car-following behavior of CACC-equipped vehicles is significantly different from
human-driven behaviors. CACC-equipped vehicles can form platoons with stable and
short following gaps. This study adopts the CACC control logic developed by Milanés and
1378 Y. GUO AND J. MA
Figure 1. Car following logic for vehicles equipped with cooperative adaptive cruise control.
Shladover (2014)andLiuetal.(2018) and incorporates recent results from the Federal High-
way Administration (FHWA) High-Performance Vehicle Stream project (2015). This study
focuses on the impact of the car-following behavior of CACC operations at the early stages
of deployment and does not explicitly consider lane changes for platooning; therefore the
lateral behaviors are modeled the same as human-driven behavior. If a CACC-equipped
vehicle intends to change lane, the longitude behavior will be regulated by CACC con-
trol logic, and the lateral behavior is controlled by the conventional lane change model.
Therefore, the CACC behavior is SAE Level 1 Automation with longitudinal control. The
cooperative merge and speed harmonization applications discussed later are also Level 1.
As shown in Figure 1, if there is no vehicle in front of the subject CAV, it will apply the
speed regulation mode to regulate the driving behavior. This mode keeps the subject CAV
cruising with target speed to reduce unnecessary oscillations, as shown in Equation (1) (Liu
et al. 2018):
asv =k1(vf−vsv)(1)
where k1is the control gain of the difference between current speed vsv and free-flow speed
vfand determines the acceleration asv. The control gain k1is set to 0.4s−1in this study (Liu
et al. 2018).
If the preceding vehicle is a conventional vehicle, the subject CAV will switch to the ACC
mode to regulate the driving behavior. If the subject CAV is too close to the preceding
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1379
vehicle (i.e. the detected clearance distance is smaller than a given minimum following
threshold), it will switch to the ACC gap regulation mode to maintain a safe following
time gap thw, as shown in Equation (2) (Liu et al. 2018). Otherwise, the CAV will repeatedly
implement previous control logic to ensure consistent driving behavior.
asv =k2(d−thwvsv −L)+k3(vl−vsv )(2)
where k2=0.23s−2and k3=0.07s−1are control gains on following distance difference
and speed difference, respectively (Liu et al. 2018). The headway d, preceding vehicle length
L, and preceding vehicle speed vlare considered in Equation (2).
If the preceding vehicle is a CAV, the subject vehicle will switch to the CACC mode and
communicate with the preceding vehicle to exchange critical information (e.g. speed, loca-
tion, platoon size). If the length of the previous CACC platoon is less than the maximum
allowable platoon length, the subject CAV will catch up with the preceding CACC platoon
and become a platoon follower; therefore the intra-platoon gap t2(0.7 s in this study) is
applied to tightly follow the preceding CAV. Otherwise, the subject CAV becomes a CACC
platoon leader and applies the inter-platoon gap t1(1.5 s in this study) to follow the pre-
ceding CAV. The specific regulation mode depends on the actual time gap between the
subject CAV and its preceding CAV. If the time gap is larger than a given threshold (2 s in
this study), the subject CAV will apply speed regulation mode, as shown in Equation (1). Oth-
erwise, it will apply the CACC gap regulation mode to keep a safe following distance with
the determined following gap (i.e. inter-platoon gap or intra-platoon gap) by implementing
Equations (3–6) (Liu et al. 2018).
vsv(t)=vsv (t−t)+kpek(t)+kd˙
ek(t)(3)
asv(t)=(vsv (t)−vsv (t−t))/t(4)
ek(t)=d(t−t)−t1vsv(t−t)−L(5)
˙
ek(t)=vl(t−t)−vsv(t−t)−t1asv (t−t)(6)
where kp=0.45s−1and kd=0.0125 are gap error control gains (Milanés and Shladover
2014).
Due to the linearity of the above models, the vehicles cannot handle emergency braking
to avoid collisions. The forward collision warning algorithm (Kiefer et al. 2003) devel-
oped by the Collision Avoidance Metrics Partnership (CAMP) is included in the C/ACC car
following modes to determine whether the gap between the subject vehicle and the pre-
ceding vehicle is sufficient for safe car following. If the crash warning is activated, it implies
that a crash will happen if both the subject vehicle and the preceding vehicle keep their
current acceleration speeds for the next few seconds. The algorithm will use a conven-
tional car-following model (e.g. Wiedermann 99) that guarantees collision-free to generate
emergency deceleration commands until the crash warning is deactivated.
In this study, as a benchmark, we chose a maximum string length of 10 vehicles, as rec-
ommended in Liu et al. (2018). Shorter string lengths would result in more CACC strings,
which can lead to lower freeway capacity because inter-string gaps are larger than the
gaps between consecutive vehicles within the string (Bujanovic and Lochrane 2018). On
the other hand, long CACC strings would make it more difficult for other vehicles to per-
form certain maneuvers, such as merging and lane changing, particularly when they need
to merge into or leave the freeway.
1380 Y. GUO AND J. MA
Speed harmonization
This study adopts the segment-based I2 V speed harmonization, as proposed in FHWA.
When an imminent or existing congestion at a bottleneck is detected, to avoid hitting the
downstream queue at a sudden full stop, the upstream CAV should slow down moderately
and pass the bottleneck smoothly at a reasonable speed just as the downstream queue dis-
sipates. This speed harmonization strategy not only smooths the CAV’s trajectory but also
helps any type of following vehicles on the mainline to move in a similar smooth manner.
As a result, the platoon of vehicles following this CAV will pass the bottleneck with a larger
throughput rate due to reduced time headway at high speed, less fuel consumption due to
smoothed trajectories, and less collision risk due to harmonized vehicle speeds.
This speed-based algorithm determines advisory speeds for freeway segments upstream
and downstream of a known bottleneck location based on measured speeds within the
bottleneck area. It is assumed that traffic detectors are available at bottleneck locations
to monitor the real-time traffic condition to calculate the arithmetic mean of speed and
occupancy of the past 3 min at these locations. Within a bottleneck area, the speed-based
algorithm tends to generate advisory speeds 10–50 percent higher than measured bottle-
neck speeds. This approach does not claim system optimization yet emphasizes simplicity
and practical field implementation. A simple linear algorithm is applied to generate the
recommended speed um(k)at time step k, as shown in Equation (7):
um(k)=αmׯ
vm(k)(7)
where αmis proportional control gain of the arithmetic mean of speed in bottleneck area m.
The value of αmis set to αm=1.3 in this study after evaluations of the effects of alternative
values.
Second, when an imminent or existing congestion is detected, this algorithm intends to
generate a lower recommended speed than the measured arithmetic mean of the speed
of the bottleneck area to smooth the upstream traffic. This algorithm is triggered by the
measured occupancy of the bottleneck area, as shown in Equation (8):
um+1(k)=Vfree,if om(k)<σ
sw
βmׯ
vm(k)if om(k)≥σsw (8)
Where,
um+1(k): the recommended speed at time step kin upstream section m+1,
um+1(k)is no less than 80% of the speed limit in section m+1. Vfree: free-flow speed.
βm: proportional control gain in section m, where βm∈[0.7, 0.9],
βm=0.8 in this study om(k): occupancy in bottleneck section m.ocri: critical occupancy
in bottleneck section m.
σsw: switch threshold of occupancy close to the critical occupancy ocri , i.e.
σsw =1−om(k)
ocratical
,
where σsw ∈[0.1, 0.125]; σsw =0.125 in this study.
It is critical to mention that in a legacy algorithm of speed harmonization for human-
driven traffic, the value of critical occupancy in the bottleneck section ocri stays constant
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1381
Figure 2. Illustration of different cooperative merging cases. (a) Case 1 (b) Case 2 (c) Case 3 (d) Case 4.
as a basic attribute of the traffic stream. With CAV traffic, however, ocri varies depending
on the market penetration of CAVs because the fundamental diagrams shift with different
market penetration rates. In this study, initial simulation runs are conducted to obtain the
capacity and critical density values under different market penetration rates of CAVs.
Cooperative merge
Cooperative merge aims to cooperatively operate both mainline vehicles and merging
vehicles to create qualified gaps at merging areas through V2 V and/or V2I communica-
tions. When the merging vehicle requests to merge or a merging vehicle is detected, a gap
should be created to let merging vehicle to merge into the mainline. The creation of gaps
relies on the situations described below, letting mainline vehicles to cooperatively change
the lane or slow down to create qualified gaps.
There are four scenarios that the vehicle may potentially encounter to activate the
cooperative merge. Please note that the algorithm process uses many parameters, and
their values have been determined by selecting those that can generate the best system
performance during initial simulation runs on simplified and actual I-66 networks.
Case 1
As shown in Figure 2(a), there is a vehicle (Vehicle B) on the target lane in front of the merg-
ing vehicle (Vehicle A). If Vehicles A and B have similar speed (i.e. small speed difference v),
the merging Vehicle A will slightly slow down and merge into the mainline. In this study, the
range of speed difference is −1.0m/s<v<0.1m/sand the deceleration rate of merging
Vehicle A is a constant value of −0.5m/s2.
Case 2
As shown in Figure 2(b), if Vehicle A intends to merge into mainline and Vehicle B on the
target lane is behind A, then B will be advised to cooperatively change lane to adjacent lane
to create a safety and acceptable gap for Vehicle A to merge into. This cooperative merge
will be activated once the following conditions are met:
1382 Y. GUO AND J. MA
(1) The new lane will not affect Vehicle B to complete its original route.
(2) The speed difference vAB between Vehicle A and Vehicle B is less than the maxi-
mum speed difference vmax, where vAB =vA−vB. The maximum speed difference
is vmax =10.8 km/h (i.e. 3 m/s) in this study.
(3) The collision time does not exceed the maximum collision time tσ, and B’s speed has
increased less than the maximum speed difference vmax. The maximum collision time
is tσ=10sin this study recommended by PTV VISSIM.
Case 3 and Case 4 involve the situation that the mainline vehicle cannot change lane into
the adjacent lane due to the requirements in Case 2 cannot be met. Therefore, the mainline
vehicle needs to reduce speed to create a gap and allow the merging vehicle to merge into
the mainline.
Case 3
As shown in Figure 2(c), on-ramp Vehicle A intends to merge into mainline and the mainline
vehicle B is behind A. Then Vehicle B will slow down and create an acceptable gap to let
on-ramp Vehicle A merge into the mainline. Meanwhile, the following vehicle C will also
cooperatively slow down to keep a safe following distance from Vehicle B. Vehicle B and C
can be independent vehicles and can also possibly be a CACC platoon in many cases.
Case 4
Figure 2(d) shows another situation that when the on-ramp Vehicle A requests to merge,
the mainline Vehicle B is too close to slow down. Then B will take no action and keep its
speed. The following Vehicle C will slow down and let A merge into the mainline.
Bundled CAV application
The aforementioned CAV applications, when bundled together, are expected to generate
higher benefits (Ma et al. 2018). This section briefly describes the operational process of the
bundled application.
The desired speed of CAV vdes is initially set to speed limit ¯
vwhen it enters the net-
work. The CACC logic determines the platoon status (i.e. platoon leader or follower) first
and decides which mode should be applied based on the preceding vehicle type, platoon
status, and current following gap. Then the CACC logic calculates acceleration or deceler-
ation rate with the desired speed vdes =¯
vand control the vehicle to stably maintain the
optimal following gap.
If speed harmonization is activated, an updated speed guidance vSH will be calculated
and distributed to CAVs within the speed harmonization zone instead of the original speed
limit ¯
v. Cooperative merge can also impact speed guidance. If the mainline traffic is signaled
to implement cooperative merge by reducing speed (i.e. Case 3 or 4), an updated speed
guidance vCM will be calculated and sent to the subject CAV. Therefore, an alternative speed
guidance ˆ
vwill be applied instead of ¯
v, where ˆ
vis defined by Equation (9).
ˆ
v=min{¯
v,vSH,vCM }(9)
Then vdes is set to vdes =ˆ
vand applied in CACC control logic to regulate the car following
behavior.
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1383
If CAV needs to change lane for the cooperative merge (i.e. Case 2 of cooperative merge),
the lane change behavior is regulated by conventional lane change model. However, during
the lane change process, all the longitudinal maneuvers are still governed by the above-
mentioned process, including collision avoidance. This means that this study focuses on
longitudinal control of all automated vehicles.
Connected vehicle behavior
Compared with CAV, CV is only capable of communicating wirelessly to exchange critical
vehicle information (e.g. speed, location) with other CVs, CAVs, and infrastructure. The CVs
cannot be automatically controlled and are still maneuvered by human drivers. In this study,
a driving behavior model is proposed to adapt CV driving behavior, which is different from
either CAVs or conventional vehicles.
The proposed CV driving behavior model is based on Wiedemann 99 model with cali-
brated parameters (Miller-Hooks, Tariverdi, and Zhang 2012),whichisalsousedtomodel
conventional vehicle driving behavior in this study. For the CACC application, a CV can be
a platoon leader since it can transmit vehicular information to the following CAVs, which
enables the CAVs to perform CACC logic. However, the CV itself cannot be the platoon
follower because it cannot be automatically controlled.
In this study, we assume all CV drivers comply with the speed harmonization and coop-
erative merge guidance because of the personalized message delivered to the vehicle. For
speed harmonization, a CV driver will reduce its speed once it received the speed guid-
ance. However, unlike the CAV, CVs’ desired speeds are heterogeneous in the traffic stream.
The speed harmonization command is only the desired speed for all CVs. The actual CV
speeds follow a pre-specified distribution with the desired speed (i.e. commands from the
speed harmonization algorithm) as the distribution mean. Therefore, CVs still show het-
erogeneous behaviors in the traffic. For cooperative merge, CVs can either send merge
requests to mainline traffic or receive merge requests from merging vehicles. If a CV in the
mainline traffic receives a cooperative merge request, it will respond to this request and
implement the cooperative merge manually (i.e. implementing the recommended speed
by the cooperative merge algorithm). If a CV intends to merge into the mainline, it will the
signal mainline traffic and finds a qualified gap to merge into the mainline manually.
Operational strategies with managed lanes
The goals of managed lane operations overlap with the goals of the CAV applications. While
each managed lane project is implemented with its own set of goals and performance mea-
sures, in general, all managed lane operations aim to provide superior traffic performance
(typically measured in terms of travel speeds and travel-time reliability) to that of adjacent
general-purpose lanes, which are not subject to the same level of active management. The
CAV strategies proposed in this study present a mechanism to increase the efficiency of
available managed lane capacity and to improve traffic performance without significant
capital investment.
Three scenarios for infrastructure are analyzed in this study. The main idea is to convert
the existing HOV2 lanes, on which only vehicles with 2 or more than 2 persons in the car
eligible, as the managed lane that also allows CVs or CAVs to access the lane. Different from
1384 Y. GUO AND J. MA
Figure 3. Illustration of three operational scenarios. (a) ML Scenario 1 (b) ML Scenario 2 (c)ML S cenario3.
the CAV-dedicated managed lane (Liu et al. 2018), this study investigates a CV/CAV eligi-
ble HOV lane, meaning that CVs and CAVs can have access to the existing HOV lane, even
if the CVs or CAVs are single-occupancy vehicles. Also note that any HOVs, human-driven
or CV/CAV, can still access the managed lanes. Therefore, this study investigates the mixed
traffic scenario where both human-driven vehicles or CVs/CAVs may co-exist on any part of
the network. Although the existence of human-driven traffic on managed lanes may neg-
atively impact the CAV traffic performance, the mixed traffic condition in this case study is
considered more realistic in the short run when early deployment of CAVs can be directly
incorporated into the transportation network and the benefits of important network users,
particularly HOVs, are not sacrificed. And it is also important to know how significant the
negative impacts of human-driven vehicles are.
Another infrastructure component of interest is the left-side dedicated ramps that are
connected to the managed lane. Because more vehicles will have access to the left-side
managed lane, vehicles may make multiple lane change maneuvers to access the managed
lane and therefore create a man-made weaving section that can reduce the highway capac-
ity and increase safety risks. Therefore, constructing the dedicated ramps can reduce such
negative effects and can be considered as a part of the CAV infrastructure strategy.
The three scenarios are illustrated in Figure 3. The key features of them are summarized
below:
•ML Scenario 1
oDedicated Ramp for HOVs, CVs, and CAVs
oExisting one-lane managed lane for HOVs, CVs, and CAVs
oHOVs, CVs, and CAVs can access ramps on both sides
•ML Scenario 2
oExisting ramp for all vehicles
oExisting one-lane managed lane for HOVs, CVs, and CAV
•ML Scenario 3
oExisting ramp for all vehicles
oExisting one-lane managed lane for HOVs only
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1385
Figure 4. I-66 Westbound simulation network.
Simulation results and analysis
Model calibration
The experimental simulation network is I-66 Westbound between interchange I495 (MM64)
and US-29 (MM 51), six interchanges on this 13-mile long section as shown in Figure 4.
And there is an HOV lane on the left side through this network. Speed and volume data
were collected by RTMS trailers along with major mainline segments. On- and off-ramp
volume data are collected by cameras. Initial calibration was performed to narrow down
parameter set candidates by using a Latin Hypercube Sampling Design (LHD) approach.
A total of 500 scenarios created by LHD were evaluated by PTV VISSIM with 5 replications
for each scenario to choose the best candidate scenario. The selected candidate was fine-
tuned to obtain the final simulation model. OD matrices are used by VISSIM to specify
travel demand. The I-66 freeway network has ten zones. Zones 1 and 10 are the start-
ing and ending points of the corridor. Zones 2–9 contain the intermediate interchanges.
Two of the zones are only applicable to the existing HOV vehicles (i.e. exits for westbound,
entrances for eastbound). The field-collected data in this study could identify how many
vehicles traveled between some, but not all, OD pairs. To fill in the gaps, OD matrices
were estimated using the QueensOD software. Results indicate an excellent correlation
between estimated and field-measured OD trips. Figure 5(a) and (b) show example results
of simulation validation. On the selected links, the comparison between estimated traf-
fic flow and traffic counts match each other well. We also use INRIX data to validate the
1386 Y. GUO AND J. MA
Figure 5. Calibration Results of the I-66 Simulation Network. (a) Comparison of Estimated Flow and
Observed Flow: 3:00 pm – 3:15 pm (b) Calibration Results of traffic counts. (c) Calibration Results of Speed
and Travel Time.
simulation, as shown in Figure 5(c), and the results show that the simulated link travel times
also match the data well. Simulation experiments were performed 10 times to account for
stochasticity.
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1387
Simulation results
In this study, three scenarios are simulated to investigate different operational strategies of
existing or upgraded freeway facilities. Four different market penetrations (MP) of 0%, 33%,
67%, and 100% are tested, which the MP is calculated by Equation (10).
MP =number of CVs + number of CAVs
number of vehicles ×100% (10)
Since CV has different driving behaviors with CAV, the CAV market penetration (MPA) is pro-
posed to represent the proportion of CAV in the traffic flow and is defined by Equation (11).
CAV market penetration can be also regarded as the autonomy level of the traffic system.
MPA = number of CAVs
number of vehicles ×100% (11)
Therefore the CV market penetration (MPC) can be defined as,
MPC = MP - MPA = number of CVs
number of vehicles ×100% (12)
Five CAV market penetration levels of 0%, 25%, 50%. 75% and 100% are applied in
this study to represent different autonomy levels. Eleven different traffic compositions
are conducted by combining different MPs and MPAs and selecting all possible pairs
={(mp,mpa)|mp ∈MP, mpa ∈MPA, mpa ≤mp}. For example, (0.33, 0.25) indicates
that 33% of all traffic is CV or CAV, which 25% of all traffic is CAV and 8% of all traffics is CV.
If there are 100 vehicles in the traffic stream, there are 8 CVs, 25 CAVs, and 67 conventional
vehicles.
Combining three different operational strategies and two different traffic demand lev-
els (100% and 130% of calibrated volume), 66 different scenarios are investigated. Each
scenario is run three times with different random seeds, which affect the oncoming traffic
pattern. The simulation period is 5 h and PTV VISSIM is used in this study as the simulation
platform. The VISSIM driver model DLL interface and COM interface are used to realize the
bundled CAV application logic.
Two measurements, network throughput and total delay, are used to evaluate the traffic
system performance. The network throughput is defined as the total number of vehicles
that arrived at their destinations within the simulation period. And the total delay is defined
as the sum of individual vehicle delays. The individual delay is calculated by Equation (13),
ti
s=d
vfree −vavg
(13)
where tsis accumulated delay of vehicle iat the time step s,dis traveled distance, vfree is
free-flow speed, and vavg is the average speed for traversing distance d. Therefore the total
delay at time step s can be calculated by Equation (14).
Ts=
I
i=1
ti
s(14)
where Iis the total number of vehicles that are in or has left the network.
Since the results of two different traffic demand levels show the similar trends, we only
present the results of 130% demand level scenarios in Analysis 2–4 to ensure the readability
of the paper.
1388 Y. GUO AND J. MA
Figure 6. Cooperative adaptive cruise control pipeline capacity under different CAV market penetration
rates.
Analysis 1 – CACC pipeline capacity
CACC can significantly improve throughput. Since this study involves five different MPA
levels of 0%, 25%, 50%, 75%, and 100%, the CACC pipeline capacity under each of those
different percentages of MPA rates is estimated. The capacity values are particularly use-
ful for the speed harmonization algorithm, for which the system’s critical density values
need to be specified under different scenarios. The capacity is tested on a 7-mile simplified
freeway segment with 4 lanes. The traffic input demand varies from 1,400 vphpl to 4,000
vphpl to generate different conditions. The data are collected every 15 min after a simula-
tion warm-up period of 15 min. The capacity is represented by four times of the maximum
15-minute-volume collected from all simulation runs for one market penetration scenario.
As shown in Figure 6, the results demonstrate a significant increase in capacity with the
increase of CAV market penetration. The benchmark (0 percent) capacity is 1,780 vphpl, and
the maximum CACC pipeline capacity is around 3,227 vphpl, an increase by 81.2 percent.
At the CAV market penetration rates of 25, 50, and 75 percent, the capacity has increased
by 14.0, 25.9, and 48.0 percent, respectively.
Analysis 2 – cooperative adaptive cruise control, speed harmonization, and
cooperative merge
CACC can significantly improve traffic performance in terms of both network throughput
and total delay. As shown in Figure 7, the increasing trend of throughput is obvious when
the market penetration increased from 0 percent to 50 percent. From 50 percent to 100
percent, the throughputs have not changed significantly with the increasing CAV market
penetration because the input volume is less than the capacity of corresponding mixed
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1389
Figure 7. Performance of applying CACC.
traffic. The descending trend of delay is significant. This is mainly because that CACC can
form CAVs strings with small gaps, maximizing the use of the existing facility and reduces
disturbances or oscillations in traffic flow. Also, it is noted that CACC platoons also have the
capability of stabilizing traffic flows because of the unique control algorithms and faster,
smoother response of any CACC vehicle to the front vehicle’s changes in speed.
As shown in Figure 8, speed harmonization can help reduce the delay and increase
throughput by smoothing the upstream traffic and discharging existing queues at the bot-
tleneck areas. Comparing (0, 0), (0.33, 0), (0.67, 0) and (1, 0) cases, delay decreases with the
increase of market penetration, but the throughput does not significantly change. As dis-
cussed before, if the speed harmonization algorithms are not designed carefully, the slow-
down effect of speed harmonization may cause negative benefits in throughput and delay.
When the market penetration is fixed and the CAV penetration rates increases, such as (0.67,
0), (0.67, 0.25), (0.67, 0.5), both throughput and delay decrease significantly. This is mainly
because AVs assume the deterministic behavior of ACC/CACC mode and can maintain rel-
atively stable following distances with prespecified, deterministic inter- and intra-platoon
gaps. This stabilizing effect can directly impact the traffic system performance.
Cooperative merge aims to facilitate merge area operations by creating gaps for merging
vehicles. As shown in Figure 9, the cooperative merge can also improve traffic performance
in terms of delay and throughput. This proves the effectiveness of the cooperative merge
algorithm and the corresponding parameters in this study. Note that if the parameters
selected in the algorithm are not optimized, initial simulation runs show that the creation of
gaps may become too frequently and then negatively impact the entire traffic performance.
Comparing cases (0.33, 0.25), (0.67, 0.25) and (1, 0.25), an interesting phenomenon can
be found that at the same low CAV market penetration rate of 0.25, the traffic perfor-
mance is getting worse with the increase of the market penetration. This is because in
this study CVs also perform cooperative merge. When the CV market penetration increases,
1390 Y. GUO AND J. MA
Figure 8. Performances of applying speed harmonization.
Figure 9. Performances of applying cooperative merge.
more vehicles are eligible for cooperative merge, and they create gaps for on-ramp vehi-
cles. As CV’s driving behavior (i.e. manually driving behavior) is quite stochastic and incurs
errors, the lane change, acceleration, and deceleration process can have an impact on the
mainline traffic performance. However, this phenomenon is gone when the CAV market
penetration becomes high because of the stabilizing effect of AV traffic flow originated from
deterministic machine driving behavior coded in the ACC/CACC vehicle algorithms.
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1391
Tab le 1. Results of CACC, Speed Harmonization, and Cooperative Merge Cases.
CACC Speed Harmonization Cooperative Merge
Network TH Delay Network TH Delay Network TH Delay
Case (veh) (h) (veh) (h) (veh) (h)
0,0 61,828 23,808.20 61,828 23,808.20 61,828 23,808.20
(0.00%) (0.00%) (0.00%) (0.00%) (0.00%) (0.00%)
0.33,0 62,136 22,495.15 62,133 22,493.52 62,804 20,963.30
(0.50%) (−5.52%) (0.49%) (−5.52%) (1.58%) (−11.95%)
0.33,0.25 69,516 4,367.58 69,020 5,032.74 69,695 3,412.22
(12.43%) (−81.66%) (11.63%) (−78.86%) (12.72%) (−85.67%)
0.67,0 62,136 22,495.15 62,491 22,015.76 63,119 20,395.98
(0.50%) (−5.52%) (1.07%) (−7.53%) (2.09%) (−14.33%)
0.67,0.25 69,248 5,110.94 67,655 8,828.23 67,774 8,542.69
(12.00%) (−78.53%) (9.42%) (−62.92%) (9.62%) (−64.12%)
0.67,0.5 71,237 572.71 70,822 1,172.15 70,758 1,427.67
(15.22%) (−97.59%) (14.55%) (−95.08%) (14.44%) (−94.00%)
1,0 62,136 22,495.15 62,545 20,864.48 63,743 18,389.62
(0.50%) (−5.52%) (1.16%) (−12.36%) (3.10%) (−22.76%)
1,0.25 68,988 5,511.25 67,318 9,558.41 66,416 12,019.20
(11.58%) (−76.85%) (8.88%) (−59.85%) (7.42%) (−49.52%)
1,0.5 71,047 737.72 70,233 2,475.64 69,556 4,498.56
(14.91%) (−96.90%) (13.59%) (−89.60%) (12.50%) (−81.11%)
1,0.75 71,093 470.25 70,424 2,793.13 70,273 2,964.61
(14.99%) (−98.02%) (13.90%) (−88.27%) (13.66%) (−87.55%)
1,1 71,166 238.41 70,098 4,116.97 70,064 4,100.12
(15.10%) (−99.00%) (13.38%) (−82.71%) (13.32%) (−82.78%)
Network TH =Network Throughput.
Table 1shows the results of applying CACC, speed harmonization, and cooperative
merge separately for the ML Scenario 1. For the overall performance, CACC performs signif-
icantly better than the other two applications, because CACC can efficiently utilize existing
facility capacity and reduce disturbances on the mainline. Note that when CAV penetration
is 0 percent, i.e. (0, 0), (0.33, 0), (0.67, 0) and (1,0) cases, there is no CAV and thus CACC is
not applied. The results of these cases can be regarded as base cases. It can be found that
there is a slight difference in traffic performance between (0, 0) and other 0 percent CAV
penetration cases. This is because, in the (0, 0) case, only HOVs, about 30 percent of total
traffic volume, can use the managed lane and dedicated ramps; and in other 0 percent CAV
cases, about 50 percent of total traffic volume, including both HOVs and CVs, can utilize the
managed lane and dedicated ramps. This rebalanced volume could slightly relief the con-
gestion at the merge area and results in a 0.5 percent throughput increase and 5.52 percent
of delay reduction.
Analysis 3 – performances of bundled CAV applications
Figure 10 shows the comparison of traffic performance between the bundled applications
and single CAV applications of ML Scenario 1. As shown in Figure 10, bundling CACC and
speed harmonization can further improve traffic performance than applying only CACC or
SH. As Error! Reference source not found. shows, bundling CACC and speed harmoniza-
tion can help increase 0.49–15.38 percent of throughput and reduce 5.52– 100.18 percent of
delay. The bundled application can improve 1.37–15.73 percent of throughput and reduce
10.53–100.17 percent of delay. The performance of the bundled application is better than
1392 Y. GUO AND J. MA
Figure 10. Performances comparison between single application and bundled applications, Scenario 1.
Figure 11. Performances comparison between single application and bundled applications, Scenario 2.
the combination of CACC and speed harmonization only when the CAV market penetra-
tion is less than 50 percent. When the CAV market penetration is greater than 50 percent, a
longer CACC string can be formed than the low CAV market penetration scenario, efficiently
using the facility capacity and improving the capacity the facility can support. Under low
CAV penetration conditions, the mainline is not congested, and there are more qualified
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1393
Figure 12. Performances comparison between single application and bundled applications, Scenario 3.
gaps for on-ramp vehicles to use, while there are less qualified gaps when the CAV market
penetration is high.
As shown in Figures 11 and 12, Scenario 2 and 3 have shown the same trends when
the applications are compared with one ML scenario, and therefore their results are not
discussed separately (Table 2).
Analysis 4 – comparison of different ML scenarios
In this section, results from three ML scenarios are compared to understand if there are
significant differences between the infrastructure scenarios. As shown in Figure 13,theML
1 scenario’s performance is significantly better than the ML 2 and ML 3 scenarios, indicating
the benefits of constructing dedicated ramps for CVs, CAVs, and HOVs. This reduces the
formation of weaving traffic at the on- and off-ramp areas and therefore is beneficial in
terms of both delay and throughput.
Comparison of ML 2 and ML 3 results indicates that allowing CVs/CAVs to access the
existing HOV lane is beneficial to the system delay and throughput, even when the market
penetration rates of CVs/CAVs are small. This benefit, however, is relatively minor compared
to the additional benefit brought by the construction of dedicated ramps (ML 1). In other
words, the weaving effects in ML 2 are quite dramatic and cancels a large portion of the
CAV benefits.
The fact that both ML 1 and ML 2 outperforms ML 3 in terms of both throughput and
delay implies the benefit of the CAV managed lane strategies. Concentrating CAVs in a sin-
gle lane can help realize early deployment opportunities. Additionally, it is proved through
this simulation that CAV dedicated lanes are not necessary for realizing early deployment
benefits. Even if CAVs are influenced by human-driven traffic (e.g. making it impossible for
certain application operations such as cooperative merge when human-driven vehicles will
not create gaps), the benefits of deploying CAVs can still be achieved.
1394 Y. GUO AND J. MA
Tab le 2. Performance Results of Scenario 1 at 130 Percent Demand Level.
CACC Only SH Only Cooperative Merge Only CACC +SH Bundled Application
Network TH Delay Network TH Delay Network TH Delay Network TH Delay Network TH Delay
Case (veh) (h) (veh) (h) (veh) (h) (veh) (h) (veh) (h)
0,0 61,828 23,808.20 61,828 23,808.20 61,828 23,808.20 61,828 23,808.20 61,828 23,808.20
(0.00%) (0.00%) (0.00%) (0.00%) (0.00%) (0.00%) (0.00%) (0.00%) (0.00%) (0.00%)
0.33,0 62,136 22,495.15 62,133 22,493.52 62,804 20,963.30 62,133 22,493.52 62,675 21,301.99
(0.50%) (−5.52%) (0.49%) (−5.52%) (1.58%) (−11.95%) (0.49%) (−5.52%) (1.37%) (−10.53%)
0.33,0.25 69,516 4,367.58 69,020 5,032.74 69,695 3,412.22 70,234 1,482.48 70,406 1,251.18
(12.43%) (−81.66%) (11.63%) (−78.86%) (12.72%) (−85.67%) (13.60%) (−93.77%) (13.87%) (−94.74%)
0.67,0 62,136 22,495.15 62,491 22,015.76 63,119 20,395.98 62,491 22,015.76 63,090 20,445.62
(0.50%) (−5.52%) (1.07%) (−7.53%) (2.09%) (−14.33%) (1.07%) (−7.53%) (2.04%) (−14.12%)
0.67,0.25 69,248 5,110.94 67,655 8,828.23 67,774 8,542.69 70,250 1,802.01 70,414 1,595.93
(12.00%) (−78.53%) (9.42%) (−62.92%) (9.62%) (−64.12%) (13.62%) (−92.43%) (13.89%) (−93.30%)
0.67,0.5 71,237 572.71 70,822 1,172.15 70,758 1,427.67 71,340 240.08 71,329 240.20
(15.22%) (−97.59%) (14.55%) (−95.08%) (14.44%) (−94.00%) (15.38%) (−98.99%) (15.37%) (−98.99%)
1,0 62,136 22,495.15 62,545 20,864.48 63,743 18,389.62 62,545 20,864.48 63,781 18,387.38
(0.50%) (−5.52%) (1.16%) (−12.36%) (3.10%) (−22.76%) (1.16%) (−12.36%) (3.16%) (−22.77%)
1,0.25 68,988 5,511.25 67,318 9,558.41 66,416 12,019.20 70,241 1,637.05 70,150 2,064.93
(11.58%) (−76.85%) (8.88%) (−59.85%) (7.42%) (−49.52%) (13.61%) (−93.12%) (13.46%) (−91.33%)
1,0.5 71,047 737.72 70,233 2,475.64 69,556 4,498.56 71,194 226.84 71,186 243.37
(14.91%) (−96.90%) (13.59%) (−89.60%) (12.50%) (−81.11%) (15.15%) (−99.05%) (15.14%) (−98.98%)
1,0.75 71,093 470.25 70,424 2,793.13 70,273 2,964.61 71,228 85.93 71,227 89.51
(14.99%) (−98.02%) (13.90%) (−88.27%) (13.66%) (−87.55%) (15.20%) (−99.64%) (15.20%) (−99.62%)
1,1 71,166 238.41 70,098 4,116.97 70,064 4,100.12 71,255 −43.68 71,268 42.22
(15.10%) (−99.00%) (13.38%) (−82.71%) (13.32%) (−82.78%) (15.25%) (−100.18%) (15.27%) (−100.18%)
Network TH =Network Throughput.
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1395
Figure 13. Performances comparison among different scenarios.
Discussion of implications
Based on the simulation results, multiple key observations and implications are summarized
as follows:
•For all scenarios, individual and bundled CAV applications can significantly improve the
traffic performance in terms of delay and throughput in most of the cases.
•CACC platooning is the most effective individual strategy because it directly reduces the
gaps between vehicles and stabilizes the traffic flow with unique platoon control algo-
rithms. Through the capacity analysis, the effect of CACC continues to increase as the
market penetration rates increase (81.2 percent increase for the 100 percent penetration
scenario).
•Speed harmonization, by its nature, can help smooth mainline traffic, increase through-
put and reduce delay in most of the cases. Because the speed harmonization relies on
monitoring of the real-time traffic state to avoid breakdown, it is critical that the exist-
ing speed harmonization algorithm is updated to reflect the mixed-autonomy traffic
conditions. Legacy algorithms may deteriorate the traffic performance.
•Cooperative merge can positively impact traffic performance by reducing force-in merge
occurrences and smooth the merging process. In the I-66 case study, the mainline
vehicles will slow down or make lane changes, when appropriate, to create safe gaps for
the merging vehicles. However, it is important to note that this process (i.e. algorithm
parameters) needs to be tweaked and optimized to reflect local geometric and traffic
conditions to ensure that the cooperative merge will not negatively impact the overall
system performance of the merge areas and the entire corridor.
•The I-66 case study shows that the bundled CAV applications with high market pene-
tration rates, if deployed, can handle the 130 percent demand scenario, indicating that
the resultant highway capacity with high market penetrations of CAVs is greater than
1396 Y. GUO AND J. MA
the 130 percent of the current demands. Additionally, even with low CAV market pene-
tration rates, there are still system benefits in early deployment stages, and this applies
to both V2 V applications (i.e. CACC) or I2 V applications (i.e. speed harmonization and
cooperative merge).
•In many cases, even though when CACC is not implemented, the system performance
still improves when CAV market penetration increases. This is because of the traffic
stabilizing effect originated from the deterministic behavior of CAVs as compared to
the stochastic behavior of human drivers. For example, cooperative merge with CVs is
implementable, but it deteriorates the system performance because the slowdown and
lane change processes of human drivers create too much disturbance to the system.
Meanwhile, the deterministic behavior of AVs makes these processes more stable.
•Scenario ML 1 performs better than ML 2 and ML 3, and ML 2 outperforms ML 3 for low
and medium market penetration cases, indicating that the dedicated ramps and man-
aged lane operation strategy are beneficial. While this conclusion is intuitive because
these two infrastructure-side enhancements reduce weaving and increase the chance
of platooning, the significance of the result is that it proves the effectiveness of a more
realistic scenario (at least for the I-66 case study) where human HOV traffic is still allowed
to access the dedicated ramps and managed lane to continue to enjoy their benefits. This
conclusion was only made for CAV-dedicated managed lanes in the past (e.g. Liu et al.,
2017). There are a large number of such existing HOV facilities in the country, and the
results of this case study prove the early deployment benefits with limited infrastructure
adjustment.
•In this case study, it is found that all three applications, when applied individually or
bundled together, are all beneficial to the system performance. Although for the rea-
sons mentioned above, CACC platooning (V2 V CAV operations) generates most of the
benefits, it is interesting to find that speed harmonization and cooperative merge (two
I2 V traffic control strategies) further improves this benefit when the CAV market pene-
tration is low to medium. This reiterates the agency’s role in realizing early deployment
benefits of CAV applications. Also, even if CACC is not implemented for certain reasons,
speed harmonization and cooperative merge, when applied individually, have signifi-
cantly positive impacts, indicating the feasibility of incorporating them as parts of the
next-generation active traffic management systems.
Conclusion
Connected and automated vehicles hold the potential for substantial improvements to traf-
fic safety, travel time reliability, driver comfort, roadway capacity, environmental impacts,
and users’ overall travel experience. As managed lanes have evolved from simple restrip-
ing and signage improvements to more sophisticated intelligent transportation systems
(ITS) and toll system deployments, they present as ideal testbeds for V2I and V2 V commu-
nications, and vehicle automation technologies as well as potential first locations for their
deployment.
This study conducts simulation on a real-world corridor I-66 in Northern Virginia and
aims to investigate the effectiveness of CAV deployment in enhancing existing traffic
system performance. The simulation results show that all three CAV applications, when
applied individually or bundled together, are all beneficial to the system performance.
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1397
CACC platooning generates most of the benefits, and it is also interesting to find that
speed harmonization and cooperative merge further improves this benefit when the CAV
market penetration is low to medium. Also, even if CACC is not implemented for certain
reasons, speed harmonization and cooperative merge, when applied individually, have sig-
nificantly positive impacts, indicating the feasibility of incorporating them as parts of the
next-generation active traffic management systems.
Last but not least, while this study is simulation-based analysis with assumed or partially
calibrated CV/CAV behavior models, some of the simulation results and insights have been
proved and validated by previous experiments conducted at FHWA. The stabilizing effect
of AVs has been validated during the speed harmonization experiment on I-66 inside the
Beltway conducted in 2015 (Ma et al. 2016) and the bundled CACC and cooperative merge
experiment (Ma et al., 2019) conducted in 2018. The efficiency of platooning and coop-
erative merge has also been tested and confirmed in the previous experiment (Ma et al.,
2019), which is in line with the simulation assumption and has the potential to generate
system-level benefits.
Although many insights have been obtained, multiple areas of the future of research
are recommended to improve the modeling and simulation. As the connected automated
vehicle data become increasingly available, the aforementioned CAV models need to be
constantly enhanced to increase model validity. Besides, in this case study, only selected
infrastructure and CAV technological strategies are simulated and evaluated as a first-step
analysis. There are other interesting scenarios that may be further simulated. And more per-
formance measurements, such as safety and environmental impact, need to be considered.
Therefore, the multiple goals need to be optimized simultaneously and the complexity of
operation will increase significantly (Li and Sun 2018;LiandSun2019).
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
This work was supported by Federal Highway Administration [grant number DTFH61-12-D-00030].
ORCID
Yi Guo http://orcid.org/0000-0002-4778-1823
References
Bujanovic, Pavle, and Taylor Lochrane. 2018. “Capacity Predictions and Capacity Passenger Car Equiv-
alents of Platooning Vehicles on Basic Segments.” Journal of Transportation Engineering, Part A:
Systems 144 (10): 04018063.
Chou, Fang-Chieh, Steven E. Shladover, and Gaurav Bansal. 2016. “Coordinated Merge Control Based
on V2 V Communication.” In 2016 IEEE Vehicular Networking Conference (VNC), 1–8. IEEE.
de Palma, Andre, and Robin Lindsey. 2011. “Traffic Congestion Pricing Methodologies and Technolo-
gies.” Transportation Research Part C: Emerging Technologies 19 (6): 1377–1399.
Elliott, David, Walter Keen, and Lei Miao. 2019. “Recent Advances in Connected and Automated
Vehicles.” Journal of Traffic and Transportation Engineering (English Edition) 6 (2): 109–131.
Guanetti, Jacopo, Yeojun Kim, and Francesco Borrelli. 2018. “Control of Connected and Automated
Vehicles: State of the Art and Future Challenges.” Annual Reviews in Control 45: 18–40.
1398 Y. GUO AND J. MA
Guo, Yi, Jiaqi Ma, Chenfeng Xiong, Xiaopeng Li, Fang Zhou, and Wei Hao. 2019. “Joint Optimization
of Vehicle Trajectories and Intersection Controllers with Connected Automated Vehicles: Com-
bined Dynamic Programming and Shooting Heuristic Approach.” Transportation Research Part C:
Emerging Technologies 98: 54–72.
ITSJPO. 2019. “Connected Vehicle Applications.” Accessed 31 August, 2019. https://www.its.dot.gov/
pilots/cv_pilot_apps.htm.
Kiefer, R. J., M. T. Cassar, C. A. Flannagan, D. J. LeBlanc, M. D. Palmer, R. K. Deering, and M. A. Shul-
man. 2003.Forward Collision Warning Requirements Project: Refining the CAMP Crash Alert Timing
Approach by Examining” Last Second” Braking and Lane Change Maneuvers Under Various Kinematic
Conditions. No. DOT HS 809 574. Washington, DC: National Highway Traffic Safety Administration.
Learn, Stacy, Jiaqi Ma, Kelli Raboy, Fang Zhou, and Yi Guo. 2017. “Freeway Speed Harmonisation
Experiment Using Connected and Automated Vehicles.” IET Intelligent Transport Systems 12 (5):
319–326.
Letter, Clark, and Lily Elefteriadou. 2017. “Efficient Control of Fully Automated Connected Vehicles at
Freeway Merge Segments.” Transportation Research Part C: Emerging Technologies 80: 190–205.
Li, Xiang, and Jian-Qiao Sun. 2018. “Signal Multiobjective Optimization for Urban Traffic Network.”
IEEE Transactions on Intelligent Transportation Systems 19 (11): 3529–3537.
Li, Xiang, and Jian-Qiao Sun. 2019. “Turning-lane and Signal Optimization at Intersections with
Multiple Objectives.” Engineering Optimization 51 (3): 484–502.
Litman, Todd. 2017.Autonomous Vehicle Implementation Predictions. Victoria, Canada: Victoria Trans-
port Policy Institute.
Liu, Hao, Lin Xiao, Xingan David Kan, Steven E. Shladover, Xiao-Yun Lu, Meng Wang, Wouter
Schakel, and Bart van Arem. 2018.Using Cooperative Adaptive Cruise Control (CACC) to Form High-
Performance Vehicle Streams. FINAL REPORT. McLean, VA: Publication.
Ma, J., E. Leslie, A. Ghiasi, S. Sethi, D. Hale, S. Shladover, X. Lu, and Z. Huang. July 2018.Applying Bundled
Speed Harmonization, Cooperative Adaptive Cruise Control, and Cooperative Merging Applications to
Managed Lane Facilities, Final Report (Draft). Washington, DC: U.S. Department of Transportation,
Federal Highway Administration.
Ma, Jiaqi, Edward Leslie, Amir Ghiasi, Zhitong Huang, and Yi Guo. 2019. “Empirical Analysis of a Free-
way Bundled Connected Automated Vehicle Application Using Experimental Data.” ASCE Journal
of Transportation Engineering.
Ma, Jiaqi, Xiaopeng Li, Steven Shladover, Hesham A. Rakha, Xiao-Yun Lu, Ramanujan Jagannathan,
and Daniel J. Dailey. 2016. “Freeway Speed Harmonization.” IEEE Transactions on Intelligent Vehicles
1 (1): 78–89.
Ma, Jiaqi, Xiaopeng Li, Fang Zhou, Jia Hu, and B. Brian Park. 2017. “Parsimonious Shooting Heuristic for
Trajectory Design of Connected Automated Traffic Part II: Computational Issues and Optimization.”
Transportation Research Part B: Methodological 95: 421–441.
Malikopoulos, Andreas A., Seongah Hong, B. Brian Park, Joyoung Lee, and Seunghan Ryu. 2018.
“Optimal Control for Speed Harmonization of Automated Vehicles.” IEEE Transactions on Intelligent
Transportation Systems 20 (7): 2405–2417.
Milanés, Vicente, and Steven E. Shladover. 2014. “Modeling Cooperative and Autonomous Adap-
tive Cruise Control Dynamic Responses Using Experimental Data.” Transportation Research Part C:
Emerging Technologies 48: 285–300.
Milanés, Vicente, Steven E. Shladover, John Spring, Christopher Nowakowski, Hiroshi Kawazoe, and
Masahide Nakamura. 2013. “Cooperative Adaptive Cruise Control in Real Traffic Situations.” IEEE
Transactions on Intelligent Transportation Systems 15 (1): 296–305.
Miller-Hooks, E., M. Tariverdi, and X. Zhang. 2012.Standardizing and Simplifying Safety Service Patrol
Benefit–Cost Ratio Estimation. Report to the I-95 Corridor Coalition.
Naus, Gerrit JL, Rene PA Vugts, Jeroen Ploeg, Marinus JG van De Molengraft, and Maarten Steinbuch.
2010. “String-stable CACC Design and Experimental Validation: A Frequency-Domain Approach.”
IEEE Transactions on Vehicular Technology 59 (9): 4268–4279.
Nowakowski Christopher, Shladover, E. Steven, Jessica O’Connell, and Delphine Cody. 2010. “Cooper-
ative Adaptive Cruise Control: Driver Selection of Car-Following Gaps.” In 17th ITS World CongressITS
JapanITS AmericaERTICO.
TRANSPORTMETRICA A: TRANSPORT SCIENCE 1399
Ntousakis, Ioannis A., Ioannis K. Nikolos, and Markos Papageorgiou. 2016. “Optimal Vehicle Trajec-
tory Planning in the Context of Cooperative Merging on Highways.” Transportation Research Part
C: Emerging Technologies 71: 464–488.
Raboy, Kelli, Jiaqi Ma, John Stark, Fang Zhou, Kyle Rush, and Ed Leslie. 2017. Cooperative Control for
Lane Change Maneuvers with Connected Automated Vehicles: A Field Experiment. No. 17-05142.
Shladover, S., D. Su, and X. Y. Lu. 2012. “Impacts of Cooperative Adaptive Cruise Control on Freeway
Traffic Flow.” Transportation Research Record: Journal of the Transportation Research Board 2324:
63–70.
Talebpour, Alireza, Hani S. Mahmassani, and Samer H. Hamdar. 2013. “Speed Harmonization: Evalua-
tion of Effectiveness Under Congested Conditions.” Transportation Research Record: Journal of the
Transportation Research Board 2391 (1): 69–79.
Yin, Yafeng, and Yingyan Lou. 2009. “Dynamic Tolling Strategies for Managed Lanes.” Journal of
Transportation Engineering 135 (2): 45–52.
Zhang, Guohui, Shuming Yan, and Yinhai Wang. 2009. “Simulation-Based Investigation on High-
Occupancy Toll Lane Operations for Washington State Route 167.” Journal of Transportation
Engineering 135 (10): 677–686.
Zhou, Fang, Xiaopeng Li, and Jiaqi Ma. 2017. “Parsimonious Shooting Heuristic for Trajectory Design
of Connected Automated Traffic Part I: Theoretical Analysis with Generalized Time Geography.”
Transportation Research Part B: Methodological 95: 394–420.
