You are on page 1of 12

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO.

10, OCTOBER 2020 4111

A Dynamic Adaptive Algorithm for Merging Into


Platoons in Connected Automated Environments
Sogand Karbalaieali , Osama A. Osman , and Sherif Ishak

Abstract— The automated merging control is one of the con- of applications such as adaptive cruise control, automated
nected and automated vehicle applications that is expected to lane sensors, or CAV’s Cooperative Adaptive Cruise Control
improve the operation and safety of freeway merging areas. This (CACC), vehicles would be able to travel in the form of
study introduces a dynamic adaptive algorithm for automated
merging control based on a cost function of travel time for road-trains with short gaps and high speeds [4], [5]. While
both on-ramp and mainline vehicles. The developed algorithm this form of operation can provide more flow stability and
is designed to guide on-ramp vehicles to merge efficiently, increased capacity, it could lead to challenges for maneuvers
without frequent slowdown or wait for merging gaps at the that rely on gap availability such as merging or lane changing.
end of the ramp and with minimal disruption to the mainline When safe gaps are not available, on-ramp vehicles are forced
traffic. The algorithm was evaluated under different mainline
traffic demands, ranging from light to heavy conditions, using a to slow down and wait until a proper gaps become available.
simulation model for a one-mile freeway segment with two lanes Merging becomes even more problematic as the relative speed
and a single-lane on-ramp. Platooning is implemented in the between the mainline and on-ramp vehicles increases due to
simulation based on Cooperative Adaptive Cruise Control. The slowing down. Therefore, when a slow on-ramp vehicle finds
performance of the algorithm is compared to a base case where a proper gap and starts to speed up to merge, the platoons
mainline vehicles cooperatively decelerate to help on-ramp vehi-
cles merge. The results show that the proposed algorithm reduced upstream of the ramp could be heavily interrupted. As a
ramp delays in the range of 38% to 91% for different traffic result, shockwaves could form especially when the mainline
conditions, without disrupting mainline operation. Additionally, vehicles slow down to accommodate the high-speed difference
travel time reliability index improved in the range of 18% to 48% with the merging vehicle.
under different traffic conditions. According to the simulation This paper develops an algorithm for freeway merging
results, the proposed automated merging algorithm is henceforth
considered reliable, with its promising performance. assistance in CAV environments with platooned traffic based
on minimization of an objective function of travel time through
Index Terms— Connected automated vehicle, automated a merging junction. The proposed algorithm optimizes the pla-
merging control, platoons, VISSIM simulation.
tooning formation and lane changing maneuvers in a dynamic
I. I NTRODUCTION process to help on-ramp vehicles merge smoothly without
disturbing the mainline traffic.
T HE freeway system in the United States is facing many
challenges due to the continuously increasing traffic
demand, especially where capacity expansion is not possible. II. L ITERATURE R EVIEW
The Connected and Automated Vehicle (CAV) technology
offers promising solutions to such challenges through auto- Owing to Vehicle to Vehicle (V2V) and Vehicle to
mated control and real-time information dissemination capa- Infrastructure (V2I) communications, several applications
bility [1], [2]. The added capabilities will increase the freeway including CACC for traffic flow stability, platooning for lane
lane capacity as a result of the reduced headway between capacity increase, and Cooperative Merging Assistance (CMA)
automated vehicles. for merging negotiation are now possible [6]. Cooperative
In CAV environments, platooning represents a viable merging is defined as a longitudinal and lateral control of
solution for capacity problems. The concept of platooning vehicles in a target lane to provide the required gap for a
was introduced by Varaiya [3] as an alternative mode of merging vehicle [7], [8]. CMA aims to improve operation and
operation that will maximize roadway capacity. With the aid safety in merging areas [9].
Past research shows several attempts to develop merg-
Manuscript received April 2, 2018; revised December 9, 2018 and June 17, ing control algorithms for CMA application. For instance,
2019; accepted August 28, 2019. Date of publication September 20, 2019; Kachroo and Li [10] proposed one of the earliest merging
date of current version October 2, 2020. The Associate Editor for this article
was F. Qu. (Corresponding author: Osama A. Osman.) control systems for automated vehicles. The authors developed
S. Karbalaieali is with the Department of Civil and Environmental Engi- three control assistance algorithms with linear, optimal and
neering, Louisiana State University, Baton Rouge, LA 70803 USA (e-mail: parabolic speed advisory functions to guide merging vehicles
s.karbalaieali@fehrandpeers.com).
O. A. Osman is with the Civil and Chemical Engineering Department, to track the speed of an available gap on the mainline. The
University of Tennessee at Chattanooga, Chattanooga, TN 37403 USA three algorithms were evaluated using simulation, and the
(e-mail: osama-osman@utc.edu). results provided evidence that the parabolic algorithm outper-
S. Ishak is with the Department of Civil and Environmental Engineering,
Old Dominion University, Norfolk, VA 23529 USA (e-mail: sishak@odu.edu). formed the other two. Kato et al. [11] studied the feasibility
Digital Object Identifier 10.1109/TITS.2019.2938728 of cooperative driving using V2V and V2I communications.
1524-9050 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
4112 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

On a test track, the research was conducted using a platoon TABLE I


of five vehicles with speed range of 24.8 to 37.3 mph. S UMMARY OF P REVIOUS R ESEARCH W ORKS
The intra-platoon gap distances were 66 and 111 feet in a
single-lane and two-lane test tracks, respectively. The study
showed that vehicles in the platoon succeeded in lane chang-
ing, merging, and leaving the platoon. In another study by
Lu et al. [12], an adaptive merging algorithm was proposed
relying on an advisory speed system for the following vehicle
in a two-vehicle platoon to allow for one on-ramp vehicle to
merge smoothly. The experiment was conducted for platoons
in a test track with the range of speeds from 13 through
17 mph. The speed of the merging vehicle was limited to the
same as that of the vehicles in the platoon. The experiment
showed that the proposed algorithm was successful in assisting
the on-ramp vehicle to merge smoothly and safely.
Rathgeber et al. [13] developed a trajectory estimation
approach for Automated Vehicle’s (AV) lane changing and by 6.4%. When implementing the large safe gap, the average
emergency braking based on non-linear models. They con- speed decreased up to 2.2%. While this study suggested that
cluded that trajectory planning for the driver assistance system merging assistance algorithms could be successful in providing
was necessary to guarantee safety and comfort. Torres and assistance for smooth and safe merging without deteriorating
Malikopoulos [14] presented an analytical closed-form merg- mainline operation, the sustainability of proposed algorithm
ing solution for CAVs’ fuel consumption with the assumptions still needs to be tested under heavier traffic conditions.
of one vehicle merging at a time, and the constant speed Marinescu et al. [16] proposed a slot-based merging algo-
in the merging area. Simulations in MATLAB/Simulink were rithm using V2I in CAV environments which works based
performed for sets of two and 15 vehicles on the mainline and on lane change advisory messages sent at a frequency of
the on-ramp with constant speed in the range of 24 to 64 mph. one second from a roadside unit (RSU) to mainline vehi-
The solution is tested against a baseline scenario where the cles in order to map gaps with merging vehicles. In a
mainline traffic has the right of way in a CAV environment. VISSIM simulation model, the algorithm was tested in a
When the number of vehicles increases, then the vehicles are 3-lane non-platooned freeway with medium to heavy mainline
forced to decelerate or fully stop. Ntousakis et al. [4] presented traffic conditions of 3,600 vph and 4,700 vph, respectively.
another method for merging assistance based on longitudinal The different on-ramp traffic flows of 200 to 2,000 vph
trajectory planning for a pair of vehicles at a merging junction. were implemented. The proposed algorithm’s performance
This study [4] aimed to minimize acceleration and jerk rate was compared to the VISSIM’s human driver model. The
using a discrete time quadratic programming formulation. The results showed that the slot-based algorithm outperformed
model predictive control assumed that each merging vehicle VISSIM’s model by improving the average travel time of
knows the final speed and time of arrival to a predetermined merging vehicles. This study shows promising potential for
merging point downstream with the constant speed equal to CMA; however, the proposed algorithm was not tested for
its final speed. The simulation of six vehicles with predefined platoon traffic. Considering the VISSIM’s human driving
merging sequences and constant speed was performed to eval- model without any modifications to account for AV tech-
uate the proposed model. While the aforementioned studies nology appears insufficient. Also, the mainline traffic condi-
provide useful insight into several algorithms supporting the tion was not evaluated for the automated merging assistance
merging assistance application of the CAV technology, they application.
do not account for the variations of traffic conditions in the The summary shown in III indicates that the literature
evaluation of the algorithms. On the contrary, other studies addressed different aspects of merging controllers to inves-
account for traffic conditions in designing and testing several tigate the potential improvements in the presence of CAV
merging assistance algorithms using simulation in connected technology. Yet, more aspects of CAV merging remain to
vehicle (CV) environments. For instance, a lane changing be explored. This study aims to overcome a few shortcom-
advisory algorithm was developed by Park et al. [15] based ings in previous studies by developing a dynamic adaptive
on V2V and V2I communications. The proposed algorithm automated merging control algorithm in CAV environments
was tested for different safe gap values (small, medium, and with two main objectives: (1) assist on-ramp vehicles in
large) using VISSIM simulation and took into consideration merging smoothly and safely to the mainline platoons under
mixed traffic of connected cars and trucks in a two-lane non- low to heavy traffic conditions, and (2) maintain the mainline
platooned freeway. The algorithm was tested in a simulation conditions with fewer disruptions or extended shockwave
of 15 minutes with demand values of 2,774 and 613 vehicles formations. In doing so, the proposed algorithm is developed
per hour (vph) for the mainline and on-ramp traffic, respec- based on a cost function of travel time to minimize the
tively. The study showed that while the choice of small gap travel time at a merging junction. The developed algorithm is
resulted in marginal improvement in the mainline operational evaluated in a traffic simulation environment with a platooned
condition, medium gap improved the average mainline speed two-lane freeway under different traffic conditions.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
KARBALAIEALI et al.: DYNAMIC ADAPTIVE ALGORITHM FOR MERGING INTO PLATOONS 4113

III. M ETHODOLOGY
The algorithm is developed to optimize merging maneu-
vers without disturbing the mainline operational and safety
conditions. The developed algorithm relies on a cost function
of the travel time to identify the optimal set of required
actions for a smooth merging process. The automated merg-
ing algorithm in this study is developed considering several
assumptions: fully automated vehicles, one vehicle merging at
a time, and communication range of 1000 feet. One of the
main concerns for applications similar to cooperative merge
is communication latency. As reported in [17], such latency
can extend from milliseconds to seconds which could be
problematic especially for safety applications. One way to
overcome that is integration with cellular networks, especially
with the introduction of 5G technology, latency problem can
be completely overcome. Considering that, our study assumes
zero latency in information communication.
In the flowchart shown in Figure 1, mainline vehicles are
set to not change lanes once they establish communication
via V2I with a central control unit located right upstream
the on-ramp. Then, an approaching on-ramp vehicle searches
for other vehicles in the rightmost lane via V2V commu-
nication. If no communication can be established until the
vehicle reaches the auxiliary lane, the vehicle will proceed to
merge. If communication is established, the on-ramp vehicle’s
onboard unit will start to collect information about all vehicles
within its communication range. This information includes the
size of the approaching platoon(s), the speed and position of
all vehicles, and the available gaps in mainline traffic.
The estimator in Figure 1 calculates the travel times required
for the on-ramp vehicle and the platoon vehicles to reach a
predetermined merging point (M) downstream the auxiliary
lane. This point is selected at a distance of one-third of the
auxiliary lane length, measured from the end of the ramp. If by
the time the on-ramp vehicle reaches point M the required
safe gap is available, the on-ramp vehicle will receive a call
to proceed with merging and the platoon vehicles continue at
the same speed. Otherwise, several options will be evaluated
to identify the alternatives that minimize the travel time in
the merging area. As shown in Figure 1, these alternatives
may include slowing down, speeding up, or changing lanes,
for a part of the platoon or an entire platoon in the mainline
rightmost lane, and slowing down or speeding up for an on-
ramp vehicle. The estimator first checks for the possibility of
each alternative and flags the possible ones for estimation of
travel time. For an alternative to be viable, enough gaps should
be available ahead of and behind the vehicles making an
action, and hence, a collision-free environment is maintained. Fig. 1. Demonstration of dynamic adaptive merging algorithm.
The estimator comparatively evaluates the flagged decisions
using the travel time cost function. Then, the decision with An alternative for the rightmost platoon vehicles is defined
the least travel time cost is consequently identified. Based as Px , an alternative for the on-ramp vehicle is defined as
on a dynamic real-time feedback approach, these decisions O y , and each selected decision is a combination of Px and
are re-evaluated every time step (δt) using the real-time data O y . To implement the algorithm as a real-time controller,
about vehicles’ status in the previous time step (t − δt). the decisions for each vehicle are limited to driving with
In other words, the algorithm adapts the merging maneu- the desired acceleration, speeding up with the maximum
ver to the current condition at each time step to select an acceleration, and slowing down with the desired deceleration.
alternative until the merging vehicle is ready to join the In the following sections, all alternatives and the associated
mainline. costs are discussed in detail.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
4114 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

A. Cost Estimation for On-Ramp Vehicles available distance d j (t), and therefore, the acceleration is
The algorithm is activated when communication is initiated increased by one increment a such that a j () = a ∗j +a and
between an on-ramp vehicle and the mainline vehicles at the distance is d1e j recalculated.
a time (t). The estimator then extracts information on the The second component T2ej is calculated using equation 5:
location and speed of the on-ramp vehicle as well as the loca- d2e j
tion, speed, platoon size, and available gaps for the mainline T2ej = (5)
vehicles. Based on this information, all available alternatives v
(O j ) for the on-ramp vehicle ( j ) at the next time step (t + δt) where the distance d2e j ,traveled in time T2ej , is calculated using
are identified. For each alternative, the associated cost (travel equation 6:
time) is also estimated.
CAVs transmit status data, such as speed, acceleration, d2e j = d tj − d1e j (6)
position, to nearby vehicles through V2V communication [18]. The on-ramp travel time T je is then calculated using
Each vehicle is assumed to have 360 degree awareness. Given equation 7:
the current position (x tj , y tj ) of the on-ramp vehicle, the net-
work stretch from that location to the merging point (x M , y M ) T je = T1ej + T2ej (7)
is divided into smaller segments (ds). Then, the distance d tj
to be traveled by an on-ramp vehicle to the merging point is The first alternative may not lead to the best overall travel
estimated using equation 1 [19]. time in the merging area. Thus, another alternative (O2 ) of
speeding up with the maximum acceleration (a max ) is consid-
 B  (x M ,y M )  2  2
j
ered to reach to the merging point faster. However, to avoid
dj =
t
ds =   d x tj + d y tj (1) the sudden acceleration changes in the time estimation, a max
A x tj ,y tj j
is assumed to be 30% of the potential vehicle’s maximum
The algorithm assumes that each on-ramp vehicle desires acceleration. In this alternative, equations (2) and (4) are
to reach a speed equal to the estimated speed v i (t + T je ) of modified according to equation 8:
the mainline vehicle (i ) at a predetermined merging point (M)
at time t + T je , where T je is the estimated arrival time of the a j = a max
j (8)
on-ramp vehicle at point (M). The speed v i (t + T je ) is the These on-ramp decisions are then evaluated against those by

target speed, which is assumed to be the speed limit (v ) in this the mainline platoon vehicles as discussed in the next section.
study. Considering this assumption, several possible alternative
combinations are evaluated for the next time step (t + δt).
In the first alternative (O1 ), during the time T je the on-ramp B. Cost Estimation for Platoon Vehicles
vehicle is assumed to accelerate until it reaches the target For each decision (O y ), the on-ramp vehicle may merge in
speed and then merges. To evaluate the possibility of this alter- one of three conditions: (a) ahead of a platoon in the rightmost
native, T je is estimated. This travel time is broken down into lane, (b) behind the platoon, or (c) the middle of the platoon.
two components. The first component is T1ej , over which the In any of the three conditions, several actions may be required
vehicle accelerates until it reaches the target speed. The second by the platoon vehicles. Any of these actions must provide
component is T2ej , over which the vehicle travels with the the minimum safe gap for the on-ramp vehicle to merge with
target speed until merging is completed. The first component collision-free condition. The vehicle in a platoon may need to
T1ej is calculated using equation 2: slow down in (a), speed up in (b), or split up in (c) as shown
in the estimator in Figure 1. Breaking up the platoon also
v j requires vehicles to either slow down or change lanes. The
T1ej = (2)
a ∗j choice of any alternative depends on the estimated location of
the vehicles in the platoon relative to the on-ramp vehicle at the
where a ∗j is the desired acceleration for vehicle j to gradually
point M. This condition is determined by comparing estimated
speeding up. The value of a ∗j is determined based on the
travel times of the platoon vehicles and the on-ramp vehicle up
vehicle characteristics and its current speed [20], [21] v j is
to the point (M). In other words, the estimated on-ramp travel
the difference between the target speed, and the current speed
time (T je ) is compared to the estimated travel times (Tie ) for
v j (t) and calculated using equation 3:
each platoon vehicle (i ) that falls within the communication

v j = v − v j (t) (3) range of the on-ramp vehicle at time t before arriving to the

merging point.
Intuitively, if v j (t) is equal to v , then T1ej is zero. The estimated travel time for a platoon vehicle is divided
The distance d1e j to be traveled by the on-ramp vehicle into two components. The first component T1ie is calculated
during time T je is then calculated by equation 4: by considering the current platoon vehicle speed v i (t) and
the current acceleration ai (t), as shown in equation (9). Since
1
d1e j = a j (T1ej )2 + v j (t)T1ej (4) each platoon vehicle is accelerating to attain a specific head-
2 way with its leader [4], [5]. After reaching the desired speed
If d1e j > d tj , then the desired acceleration is not high enough and headway, the rest of the distance to point M, d2i e , will

for the on-ramp vehicle to reach the target speed over the be traveled with a constant speed. For this reason, the second

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
KARBALAIEALI et al.: DYNAMIC ADAPTIVE ALGORITHM FOR MERGING INTO PLATOONS 4115

component T2ie is calculated by considering a constant speed


until the platoon vehicle reaches the point (M).

⎨ 0, ai (t) = 0
T1ie = v i ∀i ∈ K (9)
⎩ , ai (t) = ‚0
ai (t)
where K is the zone where platoon vehicles (i ) are within
the communication range of the on-ramp vehicle ( j ) at time Fig. 2. Schematic representation of z1 , z1 , z3 , and z4 on the mainline.
t, and v i is the current speed difference between vehicle (i )
and its leader (i −1). The speed difference for the lead platoon upstream and downstream of the merging point, respectively;
vehicle is estimated differently. In the case of i = 1, v i is  
g f and gl are the minimum acceptable gaps ahead and behind
the difference between the lead vehicle’s speed and the speed a vehicle, respectively. In this study, these zones are defined as
limit on the freeway. The speed limit is considered since this upcoming platoon vehicles’ location on the mainline relative
vehicle is assumed to speed up if and only if it is traveling to the on-ramp vehicle while it is about to merge.
below the speed limit. The speed difference is hence calculated Based on the platoon’s location at time t + T je in the zones,
using equation 10: different scenarios are evaluated. In each scenario, different
v i = v f (t) − v i (t) ∀i ∈ K (10) sets of alternatives (Pi ) could be available to the platoon
vehicles at the next time step (t +δt), and the best combination
where v f is the estimated speed that vehicle (i ) is trying to of alternatives are determined based on the minimum cost
reach and is calculated by function of travel time. The notation |z| represents the zone’s
occupancy.
v i−1 (t), i > 1
vf =  (11) Scenario (1): if |z 2 | = ∅& |z3 | = ∅, no decision is required
v, i =1 for the platoons. For the on-ramp vehicle, the alternative (out
Accordingly, the estimated distance d1i e to be traveled by of O1 and O2 ) that leads to this scenario is considered for
each vehicle (i ) in the time (T1i ) is calculated using equa-
e the next time step. If both alternatives are feasible, then O1 is
tion 12: selected.
Scenario (2): if |z 2 | = ∅& |z3 | = ∅, the following alterna-
1
e
d1i = ai (t)(T1ie )2 + v i (t)T1ie ∀i ∈ K (12) tives are evaluated: All vehicles in zones z 2 and z 1 are set to
2 continue traveling in platoons, except for the very first vehicle
The second component of the travel time (T2ie ) is calculated (i = 1) in zone z 2 (upstream). This vehicle will be set to
using equation 13: change lanes to the left in the next time step t + δt provided
e that the condition in equation 17 is satisfied. This alternative
d2i
T2ie = ∀i ∈ K (13) is referred to as (P1 ).
vf
 
e
And the distance d2i is calculated using equation 14: glli=1 ≥ gl and glf i=1 ≥ g f (17)
e
d2i = dit − d1i
e
∀i ∈ K (14) where glli=1 and glf i=1 are the estimated lag and lead gaps,
respectively, in the adjacent lane to assure the collision-
where dit is the current distance that vehicle (i ) has from the free lane change. Therefore, between every two consecutive
merging point (M). vehicles traveling on the same lane or between a prospective
The estimated travel time Tie for each mainline vehicle to leader and a follower on the adjacent lane, the gap should be
point (M) is calculated using equation 15:  
greater than an acceptable gap (gl or g f ).
Tie = T1ie + T2ie ∀i ∈ K (15) If the condition in equation 17 is not satisfied, vehicles
in zones z 2 and z 1 (upstream) may slow down as shown in
Considering T je , the estimated time on-ramp vehicle arrive equation 18. This alternative is referred to as (P2 ).
to the merging point at each time step t, as a reference value,  
the platoon can be divided into four zones as shown in equation v i (t + Tidec ) = v i (t) + ci∗ Tidec ∀i ∈ (z 1 ∪ z 2 ) (18)
16:
⎧  where ci∗ is the desired deceleration for vehicle i in zones

⎪ z 1 , Tie − T je ≥gl z 2 and z 1 which is determined based on the vehicle char-

⎨ z , 0 ≤T e − T e < g 
2 l acteristics and its current speed [20], [21]. Moreover, for
vehi cle (i ) ∈ i j
 (16)

⎪ z , 0 ≤T e
− T e
< g the vehicle not to decelerate aggressively, only 30% of the


3 j i

f
vehicle’s maximum deceleration is utilized. Tidec is the time
z 4 , T je − Tie ≥ g f
upstream vehicle (i ) decelerates to create a lag gap which is
In Figure 2, z 1 and z 4 represent the zones along the calculated as the minimum positive root of equation 19.
 
rightmost lane where an acceptable gap (gl org f ) is available 1 ∗  dec 2
to the on-ramp vehicle upstream and downstream of the c T + v i (t) Tidec
merging point, respectively; z 2 and z 3 denote the zones where 2 i i   
an acceptable gap is not available to the on-ramp vehicle +ci∗ Timod + v i (t) Timod − Tidec − dit = 0 (19)

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
4116 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

and where, the second term on the right-hand side of the equation
is the estimated available gap at time t. For alternative P4 , the
Timod = Tie + tl (20) available gap (H ) ahead of the first vehicle (i = 1) in z 4 has
where Timod is the modified travel time that upstream vehicle to be large enough to accommodate the safety of the mainline.
(i ) needs to reach the merging point; and tl is a travel time This is defined by
increase to generate the acceptable safe lead-gap for the on- H ≥ gs + t f (29)
ramp vehicle. The latter is calculated using equation 21:
 If alternative P3 does not provide enough merging gap, a
tl = gl −(T je − Ti=1
e
) (21) combination of alternatives P3 and P4 will be implemented to
where, the second term on the right-hand side of the equation satisfy the condition in equation 30:

is the estimated gap at time t. For alternative P2 , the available glf j < g f (30)
lag gap (Q) behind the last vehicle (i = n) in space (z 1 ) has
to be large enough. This is defined by In other words, vehicle i = n changes lanes, while vehicles
i = 1 : n −1 speed up according to equation (25). This combi-
Q ≥ gs + tl (22) nation could become possible if the condition in equation (29)
is satisfied.
where gs is the minimum safe gap for platooning to maintain
Scenario (4): if |z 2 | = ∅& |z 3 | = ∅, there could be a
collision-free traffic on the mainline.
potential conflict downstream and upstream the merging point
If alternative P1 does not provide a large enough gap,
at the time (t + t M ) when the on-ramp vehicles are about
as described in equation 23, a combination of alternatives P1
to join the mainline. In this case, four alternatives that are
and P2 will be implemented.
considered by combinations of Scenario (2) and Scenario (3).

l
gl j < gl (23) These alternatives (Px=5:8 ) are:
⎧ ⎧
In other words, vehicle i = 1 would change lanes, while ⎪
⎪ ⎪
⎨ P3


vehicles i = 2 : N in a platoon would slow down according ⎪
⎪ P1 + or

⎪ ⎪

to equation 18. Again, this combination is possible if the ⎪
⎪ P4

condition in equation 22 is satisfied. Px=5:8 = or (31)
Scenario (3): if |z 2 | = ∅& |z 3 | = ∅, the following ⎪
⎪ ⎧

⎪ ⎪
⎨ P3
alternatives are evaluated. All vehicles in zones z 3 and z 4 ⎪



(downstream) may continue traveling in platoons, except for ⎪
⎪ P2 or
⎩ ⎪ ⎩
the very last vehicle (i = n) in zone z 4 . This vehicle would P4
change lanes to the left if the condition in equation 24 is where P5 is both immediate downstream (z 2 ) and upstream
satisfied. This alternative is referred to as (P3 ). (z 3 ) vehicles on the mainline change lanes to the left. P6
 
glli=n ≥ gl and glf i=n ≥ g f (24) is the upstream vehicle changes lanes while the downstream
vehicle speeds up as shown in Figure 3(c). P7 is the upstream
If the condition in equation 24 is not satisfied, vehicles in vehicle slows down and the downstream vehicle changes lanes
zones z 3 and z 4 would speed up as shown in equation (25). in Figure 3(d). Finally, P8 is the upstream vehicle slows
This alternative is referred to as (P4 ). down while the downstream vehicle speeds up as shown
in Figure 3(e).
v i (t + Tiacc ) = v i (t)+ai∗ Tiacc ∀i ∈ (z 3 ∪ z 4 ) (25)
In all scenarios, the possible alternatives are evaluated based
where Tiacc is the time downstream vehicles accelerate until on the resulting estimated travel times in the merging area.
the required merging gap is available. This acceleration time In other words, the cost associated with a combination of
is calculated as the minimum positive root of equation (26). alternatives assigned concurrently for an on-ramp vehicle and
the platoon vehicles is estimated. The cost Ci j for each set of
1 ∗ acc 2
a T + v i (t) Tiacc + concurrent alternatives (Oy and Px ) is calculated as the travel
2 i i    time for the merging on-ramp vehicle ( j ) added to the travel
ai∗ Timod + v i (t) Timod − Tiacc − di (t)= 0 (26) time for all the affected platoon vehicles (i ) by alternative Pi .
This is represented in equation 32:
and
N
Timod = Tie − t f (27) Ci j |O yi &Pxi = T je + Tie (32)
i=1

where Timodis the modified travel time downstream vehicles where T je is the estimated on-ramp vehicle travel time associ-
need to reach the merging point, and t f is the amount ated with alternative O y and Tie is the estimated travel time for
of time reduction required to generate the safe lead-gap for platoon vehicle (i ) associated with alternative Px . In FIGURE
the on-ramp vehicle to merge. The latter is calculated using 4 a tree of alternative decisions is illustrated.
equation 28: The combination of alternatives that leads to the minimum

travel time (Ci j ) is selected as the optimal decision at each
t f = g f −(T je − Ti=n
e
) (28) time step. This optimal decision is sent from the controller

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
KARBALAIEALI et al.: DYNAMIC ADAPTIVE ALGORITHM FOR MERGING INTO PLATOONS 4117

Fig. 5. Simulation network layout.

For the vehicle to be able to speed up smoothly, it is necessary


for the vehicle not to decelerate aggressively so a moderate
deceleration rate should be considered similar to the boundary
condition described for P2 .

IV. N UMERICAL A NALYSIS


To evaluate the performance of the algorithm, traffic
microsimulation is considered in this research. The developed
automated merging algorithm shown in Figure 1 has several
parts. In the previous section, the estimator, cost function and
decisions were explained. The last part of the algorithm is
the controller which is responsible for sending instructions
to vehicles in the CAV network. In the microsimulation,
the instructions include the speed value v (t + δt) in next time
step (t + δt) determined by equation 33:

Fig. 3. Merging maneuver with different alternatives. a ∗ T1i,
e
j + v(t), T1i, j < δt
e
v (t + δt) = e (33)
a ∗ δt + v(t), T1i, j ≥ δt

where a is acceleration or deceleration depending on the


selected alternative. The time step δt is 2 seconds. With
a smaller time step, the acceleration/deceleration rate can
become high which is not desired. Moreover, in the simulation,
lane change requests for mainline vehicles can be made for
selected alternatives.
On the mainline, platooning is considered the main mode of
operation for traffic. While platooning increases the highway
capacity, it also affects the distribution of merging gaps, mak-
ing it more challenging especially with higher traffic volumes.
In this paper, the platoon car following on the mainline is
Fig. 4. Decision tree in dynamic adaptive automated merging algorithm.
based on Bart van Arem’s model [22] that accounts for the
CACC as expressed by equations 33 and 34.
to all affected vehicles in the communication range of the

on-ramp vehicle. Based on the disseminated instructions, all aca = al + kv vl − v f + kd (dc − v c tg ) (34)
vehicles will respond at the next time step (t + δt). Because acacc = max[amin , mi n(aca , amax )] (35)
that decision is based on estimated travel times that could have
been over- or under-estimated in some cases, the available where aca is the control acceleration for a new vehicle to join
alternatives are re-evaluated each time step to guarantee that a platoon, al is the lead vehicle acceleration,kv is a speed gain
the optimal decision is made by the time the on-ramp vehicle factor, vl is the lead vehicle speed, v f is the lag vehicle speed,
is merging. The controller terminates re-evaluation, when a kd is a distance gain factor, dc is the current distance between
merging vehicle departures the auxiliary lane since there is no the lead and the new vehicle, v c is the current speed of the
merging request. new vehicle, tg is the time gap ahead of the new vehicle, amin
If none of the alternatives are feasible at time step t, the on- is the minimum acceleration, amax is maximum acceleration,
ramp vehicle is ordered to make a third alternative (P3 ). In this and acacc is the new vehicle acceleration in the next time step.
alternative, the on-ramp vehicle slows down for one-time step According to a previous study on CACC [5], the recommended
(t + δt). Then, in the following time step (t + 2δt), the sit- kv and kd are 0.85 and 0.1, respectively.
uation is re-evaluated to determine whether this deceleration Figure 5 shows the geometric characteristics of the one-mile
is to be continued or if the vehicle can begin to speed up. freeway merging area used in the VISSIM simulation model.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
4118 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

Fig. 6. Simulation in VISSIM using COM programming in MATLAB.

TABLE II
S IMULATION PARAMETERS AND B OUNDARY C ONDITIONS

Fig. 7. Mainline demand vs throughput without merging control.

to the mainline. Then, the CD is initiated when the on-ramp


vehicle reaches half the auxiliary lane length, the mainline
platoons cooperatively decelerate to accommodate the merging
vehicle. It should be noted that the VISSIM driving behavior
model, which only considers the right of way for the mainline
vehicles, is not used for comparison because it would lead to
The algorithm, vehicle type, and simulation parameters are prolonged delays for the vehicles on the auxiliary lane. This
coded in VISSIM using Component Object Model (COM) situation also could not be realistic in a CAV environment.
programming in MATLAB as shown in FIGURE 6. More Several traffic demands of light (2,000), moderate
specifically, the driving behavior is controlled through an (3,000 and 3,500), and heavy (4,000) vehicles per hour per
external behavior dynamic-link library (DLL). This DLL is lane (vphpl) are considered in this study. These demand values
compiled to form platooning based on a modified C++ are assumed based on the lane capacity in the existence of
code from the Open Source Application Development Por- platoons according to the study by Carbaugh et al. [26]. This
tal (OSADP). Vehicles inside a platoon can accelerate and capacity is calculated using equation 35:
decelerate to keep a desired headway and speed according to v∗N
C = 3600 (36)
TABLE 2. Vehicles also perform a lane change command via L ∗ N + i ∗ (N − 1) + I
DLL.
where C is the lane capacity, v is the platoon speed, N is the
TABLE 2 shows the boundary conditions and parameters
platoon size, L is the vehicle length,i is the intra-platoon space
used in the simulated model. According to Baskar et al. [23],
gap, and I is the inter-platoon space gap. Given the assumed
the vehicle’s speed in platoons is reported from 40 to 74 mph.
parameters in this study as shown in V, the lane capacity can
Thus, the desired speed value of 60 mph and speeding up with
go up to 4,000 vphpl.
a maximum speed of 65 mph is assumed. According to the lit-
Several performance measures are used for the evaluation
erature [5], [24], [25], the intra-platoon and inter-platoon time
including travel time, travel time reliability, and delay for both
gaps are set to 0.6 and 1.3 seconds, respectively. To comply
the on-ramp and mainline traffic. The travel time reliability is
with the main assumption of one on-ramp vehicle arriving at
calculated by
a time, the on-ramp demand of 360 vph is considered, which
σT
is equivalent to one vehicle arriving every 14 seconds. Hence, IT = ∗ 100 (37)
the controller helps one merging at a time. μT
The V2V communication is also coded in COM with a where IT is the travel time reliability, σT is the standard
transmission range of 1,000 feet at a frequency of 0.1 seconds. deviation of travel time, and μT is average travel time [27],
The process of identification of the alternative decisions with [28].A low travel time reliability interprets as more reliable
the minimum travel time in the algorithm was set to be updated while individual travel time is closer to an average value, and
every 2 seconds until an on-ramp vehicle joins the mainline. vice versa for a high travel time reliability.
The performance of the proposed dynamic adaptive merging
algorithm was compared to the Cooperative Deceleration (CD) V. R ESULTS
behavior. In this simulation if an on-ramp vehicle does not find The mainline traffic conditions are explored first to under-
a safe gap available, then it slows down to delay the arrival stand the corridor operation with each traffic demand. Then,

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
KARBALAIEALI et al.: DYNAMIC ADAPTIVE ALGORITHM FOR MERGING INTO PLATOONS 4119

Fig. 8. Ramp travel time with dynamic adaptive merging (DA) and CD.

Fig. 10. Mainline travel time with dynamic adaptive merging and CD.

large variations. The figures indicate that both the merging


models have comparable travel time and delay values at lower
traffic demands (≤3,000 vphpl). This comparable performance
is expected since gaps are widely available along the corridor
with platooning. Similarly, at a demand value of 3,500 vphpl,
on-ramp vehicles were still merging smoothly with low delay
and reasonable travel time with both models. This demand
value is identified as a marginal point at which gap availabil-
ities on the mainline start to decrease. For this reason, both
approaches are sufficiently able to provide the required safe
merging gaps for the on-ramp vehicles.
As traffic demand reaches capacity, the CD can no longer
help on-ramp vehicles merge without prolonged delays, as evi-
denced by the high travel time and delay of 58.94 and
42.6 seconds, respectively. On the other hand, the dynamic
adaptive merging significantly outperforms CD in facilitating
Fig. 9. Ramp delay with dynamic adaptive merging (DA) and CD.
on-ramp vehicle merging by successfully maintaining the
operational performance similar to that of the lower demand.
the throughput values in vphpl are collected on the mainline At this demand, gap availability on the mainline becomes
upstream of the ramp. The higher value of throughput is an extensive problem that on-ramp vehicles are forced to
desired with platooning. As shown in Figure 7, the throughput wait a long time before merging. Therefore, vehicles are
and the demand are almost the same until the demand reaches required to proactively cooperate to provide the required
capacity (4,000 vphpl) where the throughput starts to drop merging gaps. This allows the on-ramp vehicles to merge
slightly. smoothly without prolonged waits, which is the case in the
In addition, Figure 7 shows that demand values more than dynamic adaptive automated merging. This performance is
3,500 vphpl cannot be totally accommodated where gaps’ demonstrated by the low average on-ramp travel time and
length is considerably reduced and their availability signif- delay of 21.99 and 3.63 seconds, respectively. While CD
icantly affected. This gap reduction indicates that merging fails to help on-ramp vehicles merge, the dynamic adaptive
maneuvers would become more difficult to perform at higher merging algorithm exhibits a steady promising performance in
traffic demands. In other words, on-ramp vehicles may expe- providing safe gaps to on-ramp vehicles by the time they arrive
rience a prolonged delay until they find the proper merging at the merging point. Moreover, DA maintained the onramp
gap. To avoid this delay, a proactive cooperation between the speed in the range of 21 to 26 m/s, while the CD dropped the
mainline and the on-ramp vehicles is required to accommodate onramp speed by 63% to 7.76 m/s.
merging. In addition to improving the average travel time and delay
The main performance measures in this study are the for the on-ramp vehicles, implementing the dynamic adap-
average travel time and average delay for both the on-ramp tive merging algorithm significantly improves the travel time
and the mainline during the simulation period. Figure 8 and reliability. As shown in VI, the reliability index for proposed
Figure 9 illustrate the on-ramp operational performance. In the algorithm is significantly less than CD’s across all demand
figures, a log transformation is assigned to y-axis due to the values.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
4120 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

TABLE III
R AMP T RAVEL T IME R ELIABILITY I NDEX ITT

The mainline actions in the merging controllers have a


direct impact on the mainline traffic operation. Therefore,
the mainline traffic should be investigated for both merging Fig. 11. Trajectories of merging vehicles and the mainline vehicles with
models. According to Figure 10, at lower traffic demands of dynamic adaptive merging algorithm.
2,000 and 3,000 vphpl, the average mainline travel time is
around 30 seconds for both dynamic adaptive merging and flow speed. In addition, the on-ramp vehicles do not wait
CD. This confirms that at the lower demands large gaps are a long time on the auxiliary lane with the help of dynamic
widely available along the mainline such that platoon vehicles adaptive merging algorithm. Moreover, Figure 11, an excerpt
are able to slow down, speed up, or change lanes with the of trajectories in the merging junction, illustrates that the
dynamic adaptive merging algorithm, or only slow down with merging maneuvers happen smoothly in between the platoons
CD without disturbing the mainline traffic. (grey lines). While the suggested slowdown and speedup
As demand approaches capacity (3500 vphpl), gap avail- actions could imply that sudden acceleration change may exist,
ability starts to become scarce. Although the CD is able a thorough review of vehicle trajectories proves otherwise.
to help on-ramp vehicles merge without experiencing severe The reason is that an approaching on-ramp vehicle has a
delays (4.78 seconds in Figure 7), the lack of gaps results reasonably high speed; thus, changing the acceleration to
in fast spreading shockwave formations on the mainline. a higher desired value can help the on-ramp vehicle reach
In addition, when more vehicles travel in platoons with the target merging speed smoothly. Moreover, the range of
short headways, a slow-down action to accommodate merging acceleration and deceleration values are bounded to moderate
vehicles could propagate backward. The outcome could be values rather than the maximum values in order to limit the
a significant delay to upstream vehicles and eventually to the sudden changes over a small time step.
entire corridor due to the extreme slow-downs on the mainline Overall, the findings of this study indicate that CAV tech-
traffic. The high average mainline travel time of 231.9 seconds nology has the potential to revolutionize transportation system
in Figure 10 supports this outcome. management and operation, yet there remain uncertainties
On the other hand, implementation of the dynamic adap- about how they will be employed on the roads. The results
tive merging algorithm maintains a low travel time value indicate that this algorithm successfully controls the merging
of 30.36 seconds on the mainline. The DA well-maintained area for the modeled network. However, further study needs
the mainline speed in range of 25.14 to 26.7 m/, where the CD to be considered for designing a real-world application. The
reduced the speed for higher demands heavily to 3.25 m/s. This merging of more than one vehicle at a time, more realistic
promising performance is the result of dynamically evaluating lateral control of vehicles, the latency in the communication,
the alternatives and proposing actions in an adaptive way. This the fuel consumption and the passenger comfort will be
approach leads to the re-distribution of gaps along the corridor addressed in future research.
which, in turn, provides the required gaps for on-ramp vehicles
to efficiently merge without disrupting the mainline condition.
VI. C ONCLUSION
As the mainline demand goes beyond 3,500 vphpl, the CD
performance drops. The cooperative slowing down actions This paper introduces an algorithm for automated merging
result in more severe shockwave formations, which are sup- control via connected automated vehicle (CAV) technology.
ported by the mainline travel time of more than 230 seconds as The dynamic adaptive algorithm is developed to help on-ramp
the demand increases from 3,500 to 4,000 vphpl (Figure 10). vehicles merge smoothly into mainline platoons under dif-
In contrast, the dynamic adaptive algorithm continues to ferent traffic conditions. The proposed algorithm accounts
maintain a stable performance by proactively rearranging the for dynamic evaluation of the alternatives and execution of
available gaps on the mainline, and hence leads to a low the merging maneuver in an adaptive way such that safe
mainline travel time of around 31 seconds for traffic demands gaps become available for merging smoothly with less than
higher than 3,500 vphpl. 4 seconds average delay. The developed algorithm also opti-
In the simulation, no concurrent congestion is considered on mizes the actions made by the mainline and on-ramp vehicles
downstream of the ramp because platoons mainly improve the with no severe disruption to the mainline traffic conditions.
capacity with no lane expansion in highways. Hence, the main- The algorithm is evaluated using a simulation model for
line vehicles pass the merging area smoothly with the free a one-mile two-lane freeway merging area with a one-lane

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
KARBALAIEALI et al.: DYNAMIC ADAPTIVE ALGORITHM FOR MERGING INTO PLATOONS 4121

on-ramp followed by an auxiliary lane. The performance of [6] S. E. Shladover, “Connected and automated vehicle systems: Introduc-
dynamic adaptive merging is compared with the cooperative tion and overview,” J. Intell. Transp. Syst., vol. 22, no. 3, pp. 190–200,
2018.
deceleration (CD) actions. Different mainline traffic conditions [7] R. Popescu-Zeletin, I. Radusch, and M. A. Rigani, Vehicular-2-X Com-
are ranging from low (2,000 vphpl) to heavy (4,000 vphpl) munication: State-of-the-Art and Research in Mobile Vehicular Ad hoc
demand. Networks. Berlin, Germany: Springer-Verlag, 2010.
[8] C. Tampère, J. H. Hogema, R. T. Van Katwijk, and B. Van Arem,
The simulation results show that both merging approaches “Exploration of the impact of intelligent speed adaptation and co-
have comparable performance at lower traffic demands operative following and merging on highways using MIXIC,” KU
(<3,500 vphpl). However, the performance of the dynamic Leuven, Leuven, Belgium, Tech. Rep., 1999.
[9] C. Diakaki, M. Papageorgiou, I. Papamichail, and I. Nikolos, “Overview
adaptive merging algorithm is significantly superior to that and analysis of vehicle automation and communication systems from
of CD at heavier traffic conditions. At lower traffic demands, a motorway traffic management perspective,” Transp. Res. A, Policy
more acceptable gaps for merging synchronize with the arrival Pract., vol. 75, pp. 147–165, May 2015.
[10] P. Kachroo and Z. Li, “Vehicle merging control design for an automated
of on-ramp vehicles to the auxiliary lane. When safe gaps highway system,” in Proc. IEEE Conf. Intell. Transp. Syst. (ITSC),
are not available on arrival of on-ramp vehicles, slowdown Nov. 1997, pp. 224–229.
actions by the mainline vehicles in CD can adequately help [11] S. Kato, S. Tsugawa, K. Tokuda, T. Matsui, and H. Fujii, “Vehicle
control algorithms for cooperative driving with automated vehicles and
on-ramp vehicles merge after experiencing minor delays (less intervehicle communications,” IEEE Trans. Intell. Transp. Syst., vol. 3,
than 5 seconds). More importantly, merging at lower demands no. 3, pp. 155–161, Sep. 2002.
does not impose significant delays to the mainline traffic, with [12] X.-Y. Lu, H.-S. Tan, S. E. Shladover, and J. K. Hedrick, “Automated
vehicle merging maneuver implementation for AHS,” Vehicle Syst. Dyn.,
either approach which is evident from their average mainline vol. 41, no. 2, pp. 85–107, 2004.
travel time being consistently around 30 seconds. [13] C. Rathgeber, F. Winkler, X. Kang, and S. Müller, “Optimal trajec-
At heavier traffic demands with significantly less safe gap tories for highly automated driving,” World Acad. Sci., Eng. Technol.
Int. J. Mech., Aerosp., Ind., Mechatronic Manuf. Eng., vol. 9, no. 6,
availability, slow down actions by the mainline vehicles in pp. 969–975, 2015.
CD lead to severe shockwave formations that extend farther [14] J. Rios-Torres and A. A. Malikopoulos, “Automated and cooperative
upstream on the mainline. Additionally, by the time merging vehicle merging at highway on-ramps,” IEEE Trans. Intell. Transp. Syst.,
vol. 18, no. 4, pp. 780–789, Apr. 2017.
gaps become available, the on-ramp vehicles have already [15] H. Park, C. S. Bhamidipati, and B. L. Smith, “Development and evalu-
suffered extreme delays. To avoid this situation, the vehicles ation of enhanced intellidrive-enabled lane changing advisory algorithm
should adapt to the traffic conditions through dynamic adaptive to address freeway merge conflict,” Transp. Res. Rec., J. Transp. Res.
Board, vol. 2243, no. 1, pp. 146–157, 2011.
merging. As a result, a stable operational performance for [16] D. Marinescu, J. Čurn, M. Bouroche, and V. Cahill, “On-ramp traffic
both mainline and on-ramp vehicles was well-maintained at merging using cooperative intelligent vehicles: A slot-based approach,”
all demands. Overall, the results of this study prove that the in Proc. 15th IEEE Conf. Intell. Transp. Syst. (ITSC), Sep. 2012,
pp. 900–906.
developed algorithm is a promising tool for the merging assis- [17] D. Jia, K. Lu, J. Wang, X. Zhang, and X. Shen, “A survey on platoon-
tance application of CAV technology. It is important however based vehicular cyber-physical systems,” IEEE Commun. Surveys Tuts.,
to mention that the model considered only one vehicle merging vol. 18, no. 1, pp. 263–284, 1st Quart., 2016.
[18] N. J. Goodall, B. L. Smith, and B. Park, “Microscopic estimation of
at a time with a set of pre-identified merging alternatives. Thus, freeway vehicle positions from the behavior of connected vehicles,”
future work will consider (a) adding platoon merging, and (b) J. Intell. Transp. Syst., vol. 20, no. 1, pp. 45–54, 2016.
solving merging as a continuous problem rather than discrete [19] D. Tian et al., “A dynamic travel time estimation model based on
connected vehicles,” Math. Problems Eng., vol. 2015, Apr. 2015,
with pre-identified merging alternatives. Art. no. 903962.
[20] A. Mehar, S. Chandra, and S. Velmurugan, “Speed and acceleration
characteristics of different types of vehicles on multi-lane highways,”
ACKNOWLEDGMENT Eur. Transp., vol. 55, no. 1, pp. 1–2, 2013.
[21] P. S. Bokare and A. K. Maurya, “Acceleration-deceleration behaviour of
The authors would like to acknowledge Dr. Sameer AbuAsal various vehicle types,” Transp. Res. Procedia, vol. 25, pp. 4733–4749,
for his IT advice, Bita Hashemi for her assistance with the pro- Jun. 2017.
gramming, and Mathew Theriot for his technical editing. This [22] B. van Arem, A. P. de Vos, and M. Vanderschuren, “The microscopic
traffic simulation model MIXIC 1.3,” TNO Inro, Dept. Traffic Transp.,
research did not receive any specific grant from funding agen- Delft, The Netherlands, TNO-Rep. INRO-VVG 1997-02b, 1997.
cies in the public, commercial, or not-for-profit organizations. [23] L. D. Baskar, B. De Schutter, and H. Hellendoorn, “Traffic management
for automated highway systems using model-based predictive control,”
IEEE Trans. Intell. Transp. Syst., vol. 13, no. 2, pp. 838–847, Jun. 2012.
R EFERENCES [24] S. E. Shladover, D. Su, and X.-Y. Lu, “Impacts of cooperative adaptive
cruise control on freeway traffic flow,” Transp. Res. Rec., J. Transp. Res.
[1] Y. Feng, K. L. Head, S. Khoshmagham, and M. Zamanipour, Board, vol. 2324, no. 1, pp. 63–70, 2012.
“A real-time adaptive signal control in a connected vehicle environment,” [25] M. Amoozadeh, H. Deng, C.-N. Chuah, H. M. Zhang, and D. Ghosal,
Transp. Res. C, Emerg. Technol., vol. 55, pp. 460–473, Jun. 2015. “Platoon management with cooperative adaptive cruise control enabled
[2] C. Roncoli, M. Papageorgiou, and I. Papamichail, “Traffic flow optimi- by VANET,” Veh. Commun., vol. 2, no. 2, pp. 110–123, 2015.
sation in presence of vehicle automation and communication systems— [26] J. Carbaugh, D. N. Godbole, and R. Sengupta, “Safety and capacity
Part II: Optimal control for multi-lane motorways,” Transp. Res. C, analysis of automated and manual highway systems,” Transp. Res. C,
Emerg. Technol., vol. 57, pp. 260–275, Aug. 2015. Emerg. Technol., vol. 6, nos. 1–2, pp. 69–99, 1998.
[3] P. Varaiya, “Smart cars on smart roads: Problems of control,” IEEE [27] M. Martchouk and F. Mannering, “Analysis of travel time reliabil-
Trans. Autom. Control, vol. 38, no. 2, pp. 195–207, Feb. 1993. ity on Indiana interstates,” NEXTRANS Center, Purdue Univ., West
[4] I. A. Ntousakis, I. K. Nikolos, and M. Papageorgiou, “Optimal vehicle Lafayette, IN, USA, Tech. Rep. NEXTRANS Project No 014PY01,
trajectory planning in the context of cooperative merging on highways,” 2009.
Transp. Res. C, Emerg. Technol., vol. 71, pp. 464–488, Oct. 2016. [28] D. Fha, “Travel time reliability: Making it there on time, all
[5] L. Zhao and J. Sun, “Simulation framework for vehicle platooning the time,” U.S. Dept. Transp., Federal Highway Admin., Wash-
and car-following behaviors under connected-vehicle environment,” ington, DC, USA, Tech. Rep., 2010. [Online]. Available: https://
Procedia-Social Behav. Sci., vol. 96, pp. 914–924, Nov. 2013. ops.fhwa.dot.gov/publications/tt_reliability/TTR_Report.htm

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.
4122 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

Sogand Karbalaieali received the B.S. and M.S. Sherif Ishak received the Ph.D. degree.
degrees in civil engineering in Iran in 2008 and He is a Professor and the Department Chair of the
2011, respectively, and the Ph.D. degree in trans- Civil and Environmental Engineering, Old Dominion
portation engineering. University. Prior to joining ODU, he was a Professor
She joined Louisiana State University Graduate of civil engineering with Louisiana State University
School in 2014. She is currently an Engineer with (LSU), and a Professor and the Department Chair
Fehr & Peers DC, Washington, DC, USA. Her of The University of Alabama, Huntsville. He has
research interests include the application of con- over 25 years of experience in the field of trans-
nected automated vehicle technology in transporta- portation engineering with emphasis on intelligent
tion, traffic flow modeling, traffic microsimulation, transportation systems, traffic operation and control,
traffic safety data explanatory analysis, and intelli- traffic flow modeling and simulation, traffic safety,
gent transportation systems. She is a member of the Institute of Transportation human factors and driving behavior, artificial intelligence and advanced
Engineers. She was a recipient of the National Operation Center of Excellence computing applications in transportation, and the new emerging area of
Award in 2017, the Gulf Region ITS Award in 2017, and Lifesavers National connected and automated vehicles. Since 2015, he has been serving as
Conference on Highway Safety Priorities Scholarship in 2018. the Chair of the Transportation Research Board Standing Committee on
Artificial Intelligence and Advanced Computing Applications (ABJ70), which
Osama A. Osman received the Ph.D. degree. promotes and advances interdisciplinary research at the intersection of the
He is an Assistant Professor in intelligent trans- fields of transportation engineering and computer science and information
portation systems and data analytics with the Depart- technology. He is also a member of the five-year National Cooperative
ment of Civil and Chemical Engineering, UTC. Prior Highway Research Program (NCHRP) panel for maintaining and executing
to joining UTC, he was a Research Associate with a research agenda for connected and automated vehicles (CAV) roadmap.
the Center for Sustainable Mobility, Virginia Tech He was the Civil Engineering Program Director of the master’s degree
Transportation Institute. From 2016 to 2018, he was students at LSU and the Associate Dean of Academic Programs and the
also a Research Associate of civil and environmental Director of the Interdisciplinary Graduate Engineering Science Program. His
engineering with Louisiana State University. He has extensive service record includes serving on the Faculty Senate, the Discover
extensive research experience in several transporta- Advisory Board for the master’s degree research, the Communication Across
tion research areas including traffic modeling and the Curriculum Advisory Council, the College Academic Matters Committee,
simulation, traffic safety and human factors research, traffic operations, the College Policy Committee, and other university, college, and departmental
intelligent transportation systems, connected/automated vehicles, and the committees. He is also an Associate Editor of the Canadian Journal of Civil
applications of artificial intelligence in transportation. He has served as the Engineering.
PI or Co-PI of several research projects supported by the USDOT-UTC
Program and Louisiana DOTD in the areas of active traffic management,
modeling and experimental testing of connected/automated vehicle technol-
ogy, and advanced data analytics using machine learning and deep learning
techniques. He has over 45 publications in peer-reviewed journals, national
and international conference proceedings, technical reports, and a book
chapter. He is also the Committee Research Coordinator of TRB-ABJ70. He is
a member of the Committee for Shared and Digital Mobility of the Society
of Automotive Engineers (SAE). He serves as the Chair for the Commu-
nications Subcommittee, TRB Standing Committee on Artificial Intelligence
and Advanced Computing Applications (TRB-ABJ70). He is also an active
reviewer of several journals including the Journal of Intelligent Transportation
Systems, the IEEE T RANSACTIONS OF I NTELLIGENT T RANSPORTATION
S YSTEMS , the IEEE TRANSACTIONS ON I NTELLIGENT V EHICLES , the Jour-
nal of Accident Analysis and Prevention, the Journal of Safety Science,
the Journal of the Transportation Research Record, the Canadian Journal
of Civil Engineering, the Journal of Transportation Research-Part C, and the
Journal of Transportation Engineering: Part A.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on October 22,2020 at 06:58:13 UTC from IEEE Xplore. Restrictions apply.

You might also like