14876 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 25, NO.

10, OCTOBER 2024

Optimizing Bus Operations at Autonomous

Intersection With Trajectory Planning
and Priority Control
Wei Wu , Mengfei Xiong , Tangzhi Liu, Jian Sun , and Yongfu Li , Senior Member, IEEE

Abstract— Existing studies on Autonomous Intersection Man- by CPLEX. Results demonstrate significant reductions in average
agement (AIM) primarily focus on regular vehicles (e.g., cars), bus delay and passenger delay under the proposed model, and
while ignoring bus priority demands. This paper aims to optimize sensitivity analysis further examines its effectiveness.
the bus operations at autonomous intersection with trajectory
planning and priority control. First, an intersection trajectory Index Terms— Autonomous intersection management, bus tra-
planning approach is proposed for turning movements within the jectory design, dynamic bus lane, bus priority control.
intersection considering significant passenger volume and large
size of buses. A two-stage trajectory planning method is adopted
that employs a combination of transition and circular curves to I. I NTRODUCTION
ensure the smooth turning movements for vehicles within the
intersection. Next, a bus priority control model for autonomous
intersection (AIM-BP) is developed to minimize weighted com-
binations of total bus delay and car delay. In particular, the
I NTERSECTIONS are widely recognized as the bottlenecks
for traffic flows in urban transportation networks [1].
Hence, ensuring effective management of intersections is
model incorporates the introduction of the dynamic bus lane important for alleviating traffic congestion. Improving the
designed to clear the vehicles in front of buses, thereby creating signal timings at intersections can significantly enhance the
a relatively exclusive space for the buses. The proposed model
simultaneously optimizes the lane choice on the road section, traffic efficiency of intersections and urban transportation net-
the route choice within the intersection and the time to enter works [2]. Research on intersection management has primarily
the intersection for each vehicle, while determining whether focused on signal control, which can be divided into fixed-time
to deploy the dynamic bus lane and which lane serves as the control, actuated control and adaptive control. A compre-
dynamic bus lane in the approach. The model is formulated as a hensive review of the above three types of signal control
Mixed Integer Linear Programming (MILP) problem, compiled
in AMPL (A Mathematical Programming Language) and solved strategies can be found in [3]. With the rapid advancement
of autonomous driving technology [4], a novel intersection
Manuscript received 5 November 2023; revised 21 February 2024 and control method is introduced, known as AIM (Autonomous
8 April 2024; accepted 22 April 2024. Date of publication 13 May 2024;
date of current version 4 October 2024. This work was supported in
Intersection Management), which is first proposed by Dres-
part by Changsha Science and Technology Plan Project Funding under ner and Stone [5], [6]. The significant potential of AIM to
Grant kq2107009, in part by the Ministry of Education Humanities and improve traffic efficiency at intersections has received substan-
Social Sciences Research Project under Grant 22YJCZH189, in part by
the National Natural Science Foundation of China under Grant 61773077, tial research attention [7], [8], [9]. Under AIM, autonomous
in part by the Natural Science Foundation of Chongqing under Grant vehicles can coordinate among each other to traverse the
CSTB2022NSCQ-LZX0025, in part by the Talent Program of Chongqing intersection without the control of traffic lights [10], [11].
under Grant cstc2024ycjh-bgzxm0037, and in part by the Science and Tech-
nology Research Program of Chongqing Municipal Education Commission Most research on AIM has been dedicated to vehicle trajectory
under Grant KJZD-M202300602. The Associate Editor for this article was planning and traffic control strategies.
X. Li. (Corresponding author: Tangzhi Liu.) In terms of vehicle trajectory planning, most studies have
Wei Wu is with Chongqing Key Laboratory of Intelligent Integrated and
Multidimensional Transportation System, Chongqing Jiaotong University, developed the related models and algorithms to optimize
Chongqing 400074, China, and also with the School of Traffic and Transporta- the entering time, passing sequence, and travel speeds and
tion Engineering, Changsha University of Science and Technology, Changsha accelerations for vehicles, which aims to avoid the conflicts
410114, China (e-mail: [email protected]).
Mengfei Xiong is with the School of Traffic and Transportation Engineering, among vehicles and reduce vehicle delays in the intersection.
Changsha University of Science and Technology, Changsha 410114, China Levin et al. [12] propose a conflict point model to decide in
(e-mail: [email protected]). which order vehicles should move. Fayazi and Vahidi [13]
Tangzhi Liu is with Chongqing Key Laboratory of Intelligent Integrated
and Multidimensional Transportation System, Chongqing Jiaotong University, formulate the MILP model to obtain the optimal arrival time
Chongqing 400074, China (e-mail: [email protected]). for each vehicle and propose a trajectory planning algorithm
Jian Sun is with the Department of Traffic Engineering and the Key to adjust vehicles’ speeds in order to make them access the
Laboratory of Road and Traffic Engineering, Ministry of Education, Tongji
University, Shanghai 201804, China (e-mail: [email protected]). intersection at scheduled arrival times. Li et al. [14] use a
Yongfu Li is with the College of Automation, Key Laboratory of Intelligent genetic algorithm to determine the optimum vehicle passing
Air-Ground Cooperative Control for Universities in Chongqing, Chongqing sequence and develop a trajectory optimization algorithm to
University of Posts and Telecommunications, Chongqing 400065, China
(e-mail: [email protected]). optimize vehicles’ speeds and accelerations in the intersection.
Digital Object Identifier 10.1109/TITS.2024.3395280 Mirheli et al. [15] develop a distributed cooperative control
1558-0016 © 2024 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: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 WU et al.: OPTIMIZING BUS OPERATIONS AT AUTONOMOUS INTERSECTION 14877

logic to find the optimal conflict-free trajectories by assigning In order to mitigate the adverse impacts on other vehicles
accelerations to vehicles to approach and cross the intersection, while improving bus operation efficiency, the conditional
which aims to minimize the travel time and speed varia- signal priority (CSP) is developed [28]. Christofa et al. [29]
tions. Liu et al. [16] perform a window searching algorithm propose to assign priority to the buses based on their passen-
to find the time and velocity to enter the intersection for ger occupancy and a mixed integer nonlinear programming
each vehicle and then a segmented dynamic programming (MINLP) model is formulated to minimize the total person
algorithm is applied to plan the vehicle’s velocity profile delay in the intersection. Hu et al. [30] claim that the priority
before entering the intersection. However, these studies have is granted only when the bus is behind schedule and no extra
mainly assumed that the vehicles’ route choices are exogenous total person delay is caused. Anderson et al. [31] propose a
input. Wu et al. [17], [18] simultaneously optimize the entering mathematical model based on Brownian motion to evaluate
time and route choice for each vehicle and the effectiveness of the conditional signal priority (CSP), wherein the buses send
route choice optimization in reducing potential conflicts and priority requests only when the requests improve reliabil-
average vehicle delay is proved. ity. Zhang et al. [32] introduce the Dueling Double Deep
Moreover, in order to describe the vehicle’s travel route Q-learning Network (D3QN) algorithm to determine whether
within the intersection, trajectory equations have been estab- to implement the TSP strategy or not, considering the goal of
lished. Levin et al. [12] restrict the approach lanes to certain minimizing the total passenger waiting time of buses and cars.
traffic movements (left-turn, through, or right-turn) and the cir- On the other hand, bus lane deployment achieves the spatial
cular curve is used to establish trajectory equations for turning separation between buses and other vehicles. It improves the
vehicles. However, He et al. [19] propose the all-direction turn speed and reliability of bus service, but reduces the road space
lane (ADTL), where left-turn, through, and right-turn traffic available to cars [33]. As the number of cars arriving at the
can be allowed at the same lane and autonomous vehicles intersection increases, it will exacerbate traffic congestion and
on any approach lane are able to turn onto any exit lane. reduce traffic capacity. To address this concern, the concept
Based on ADTL, the elliptic curve is used to establish the of dynamic bus lane is introduced, wherein the bus lane is
trajectory equations for turning vehicles in [17], [18], and accessible to other vehicles in the absence of buses [34]. Based
[20]. However, there is still room for improvement in how on this concept, Viegas et al. [35] propose the intermittent
to plan smoother trajectories for turning vehicles within the bus lane, which changes to a bus lane only when the bus is
intersection, especially for large vehicles like buses and trucks. approaching. However, it only prohibits the cars ahead of the
In terms of traffic control strategies, a tile-based reservation bus from entering the bus lane, but permits the existing cars in
(TBR) method is proposed in [5] and [6], where the intersec- the bus lane to continue using it. Therefore, Eichler et al. [36]
tion is divided into several grids of reservation tiles. Then propose that the cars should exit the bus lane when the bus is
the “First Come First Served” (FCFS) strategy is adopted present. Furthermore, Wu et al. [37] claim that only the cars
to process vehicles’ requests of occupying reservation tiles. ahead of the bus within a “Clear Distance” are required to exit
Based on the TBR method, Levin et al. [21] adopt the auction- the bus lane, while the cars beyond the “Clear Distance” or
based strategy, where high-value-of-time travelers are able to behind the bus are allowed to drive in the bus lane. However,
gain priority through intersection auctions. Bashiri et al. [22] these studies are conducted in traditional intersection involving
develop a platoon-based strategy, where the leading vehicle human-driven vehicles, rather than in autonomous intersection.
in each platoon is in charge of sending requests to cross the Emerging technologies of autonomous driving and V2X
intersection. To guarantee the traffic safety, only one platoon is communication make the accurate control of vehicles become
allowed in the conflict zone at any time. Wang et al. [23] use available [38], [39], [40], thereby creating more reliable con-
a game theory and formulate a Nash game model and a coop- ditions for bus priority control in the intersection. To realize
erative game model to capture the non-cooperative behavior the wide deployment of autonomous vehicles, one important
and cooperative behavior of conflicting vehicles respectively. effort lies in the developing of autonomous buses [41]. Hence,
The aforementioned control strategies have mainly focused it is necessary to conduct research on bus priority control
on regular vehicles (e.g., cars), with little attention paid to in autonomous intersection. Zhang et al. [39] design a bus
the issue of bus priority demands. As the number of private priority lane (BPL) for the buses and allow some cars to access
cars continues to increase, traffic congestion has become a through trajectory-based control. It aims to minimize the sacri-
pressing issue. Thus, developing bus priority control strategies fice of cars while ensuring absolute bus priority in autonomous
is of great importance. Such strategies have the potential to intersection, but still assumes that the vehicle trajectory is
improve bus operation efficiency and attract more travelers to subject to the signal plan constraints. Chen et al. [40] propose
take buses, thereby alleviating traffic congestion and reducing an innovative rhythmic control scheme for heterogeneous
energy consumption and pollutant emissions [24], [25], [26]. traffic (RC-H) to minimize the weighted travel time of buses
Bus priority control strategies can be broadly classified into and cars, where the regular virtual platoons are designed for
bus signal priority (BSP) and bus lane deployment. cars and the dedicated virtual platoons are designed for buses
BSP can be divided into passive priority and active priority. to provide exclusive right-of-ways.
Passive priority predetermines the signal timings to grant To summarize, previous studies on AIM have predominantly
priority to buses, without actually detecting the presence of focused on vehicle trajectory planning and traffic control
buses. However, active priority makes adjustments in signal strategies that effectively improve intersection traffic efficiency
timings to grant priority according to the arrival of buses [27]. while avoiding vehicle conflicts. However, some limitations

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 14878 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 25, NO. 10, OCTOBER 2024

still exist in these studies. First, in terms of vehicle trajectory

planning, existing studies have used elliptic curve to establish
trajectory equations for turning vehicles. However, given the
significant passenger volume and large size of buses, ensuring
traffic safety and passenger comfort is important when for-
mulating vehicle trajectory equations within the intersection.
Therefore, it is necessary to plan smoother trajectories for
turning movements within the intersection. Second, in terms
of traffic control strategies, most existing studies on AIM
often concentrate on regular vehicles (e.g., cars), with little
attention paid to the issue of bus priority demands, leading to
an increase in average passenger delay. To tackle the problems
mentioned above, this study aims to contribute to the literature
in the following aspects: Fig. 1. The layout of an autonomous intersection.
• This paper puts forward a new trajectory planning
approach for turning movements within the intersection
that takes into account the significant passenger volume Furthermore, while ensuring bus priority demands, it is critical
and large size of buses. A two-stage trajectory planning to optimize the routes and the time to enter the intersection
method is proposed that employs a combination of tran- for buses and cars, while determining whether to deploy the
sition and circular curves to ensure the smooth turning dynamic bus lane and which lane serves as the dynamic bus
movements and traffic safety for the vehicles within the lane in the approach.
intersection. Meanwhile, the golden section method is The main notations in the model are listed in Table I.
used to determine the intersection point of the above two
types of curves. III. M ODEL F ORMULATION
• This paper develops a bus priority control model for the A. Vehicle Trajectory Planning
autonomous intersection (AIM-BP). As the vehicles can This section proposes a new trajectory planning approach
coordinate among each other to traverse the autonomous for turning movements within the intersection that takes into
intersection, the proposed model aims to minimize the account the significant passenger volume and large size of
weighted combinations of the total bus delay and car buses. According to the number of approach and exit lanes
delay. In particular, the model incorporates the introduc- in each direction and the lane width at the intersection,
tion of the dynamic bus lane designed to clear the cars the vehicle trajectory equations within the intersection are
in front of buses, consequently establishing a relatively formulated.
exclusive space for the buses. The proposed model simul- It is assumed that there are n approach lanes and n exit
taneously optimizes the lane choice on the road section, lanes in each direction, as shown in Figure 2. Let O denote
the route choice within the intersection and the time to the approach direction, O ∈ {E, W, S, N } and D denote the
enter the intersection for each vehicle, while determining exit direction, D ∈ {E, W, S, N } where E, W, S, N indicate
whether to deploy the dynamic bus lane and which lane East, West, South and North, respectively. Let i O denote the
serves as the dynamic bus lane in the approach. index of the lane in approach O, i O ∈ {1, 2, . . . , n} and j D
The remainder of this paper is organized as follows. denote the index of the lane in exit D, j D ∈ {1, 2, . . . , n}.
Section II presents the research problems and the list of We establish a rectangular coordinate system and orient the
notations. Section III presents vehicle trajectory planning origin of coordinates at the center of the intersection. There
method, and then presents the bus priority control model are four vertices denoted by (a1 , b1 ), (a2 , b2 ), (a3 , b3 ), and
for autonomous intersection. Section IV presents numerical (a4 , b4 ), and the coordinates are (−(n L + L 0 ), n L + L 0 ),
analysis and Section V concludes the paper. (n L + L 0 , n L + L 0 ), (n L + L 0 , −(n L + L 0 )) and (−(n L +
L 0 ), −(n L + L 0 )), respectively, where L is the lane width
II. P ROBLEM D ESCRIPTION AND N OTATIONS and L 0 is the distance between the vertex and the outer
As shown in Figure 1, in the autonomous intersection, edge of the lane. Let PO,i (x O,i , y O,i ) denote the starting
the upcoming vehicles include both cars and buses, with the point of the vehicle trajectory, which means that the vehicle
straight lines denoting the trajectories of through vehicles, and enters the intersection from the i th lane of approach O.
the curved lines denoting the trajectories of turning vehicles. Let Q D, j (x D, j , y D, j ) denote the end point of the vehicle
To enhance model flexibility, we set each approach lane as trajectory, which means that the vehicle leaves the intersection
the all-direction turn lane (ADTL), where left-turn, through, from the j th lane of exit D. In Figure 2, the trajectories
and right-turn traffic can be allowed at the same lane and of the right-turning vehicles are represented by the orange
autonomous vehicles on any approach lane are able to turn curves, while the trajectories of the left-turning vehicles are
onto any exit lane (e.g., P → Q 1 , P → Q 2 and P → depicted by the green curves. The trajectories PW,2 → Q S,2
Q n (n>2)). Given the significant passenger volume and large and PW,2 → Q N ,2 are designed using circular curves. The
size of buses, it is important to design the trajectory for buses trajectories PW,2 → Q S,n , PW,2 → Q S,1 , PW,2 → Q N ,1 and
within the intersection especially for the turning movements. PW,2 → Q N ,n are designed by employing a combination of

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 WU et al.: OPTIMIZING BUS OPERATIONS AT AUTONOMOUS INTERSECTION 14879

TABLE I TABLE I
L IST OF M AIN N OTATIONS (Continued.) L IST OF M AIN N OTATIONS

Fig. 2. Intersection layout and vehicle trajectories.

transition and circular curves, where Ma , Mb , Mc and Md

represent the intersection points of the transition and circular
curves for the above four trajectories respectively.
1) Trajectory Equation of Through Vehicles: We take
the through vehicle in the west approach as an example, as the
blue lines shown in Figure 1. As the vehicle travels from the
approach lane i W to the exit lane j E , the starting point (i.e.,
PW,i (x W,i , yW,i )) and the end point (i.e., Q E, j (x E, j , y E, j )) of
the trajectory can be determined as follows.
x W,i , yW,i = (−(n · L + L 0 ) , −[(i W − 1) · L + L/2]) (1)

x E, j , y E, j = (n · L + L 0 , −[( j E − 1) · L + L/2])


(2)
The slope of the trajectory can be computed by Eq. (3),
and the trajectory equation for the through vehicle in the west
approach can be determined by Eq. (4).
k = (y E, j − yW,i ) (x E, j − x W,i )

(3)


y − yW,i = k · x − x W,i (4)
where the value range of x is: x ∈ [min(x W,i , x E, j ),
max(x W,i , x E, j )].

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 14880 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 25, NO. 10, OCTOBER 2024

Fig. 3. Trajectory I for right turning vehicles.

Similarly, the trajectory equations for through vehicles in

other approaches can be derived. Fig. 4. Trajectory II for right turning vehicles.
2) Trajectory Equation of Turning Vehicles: We take the
right turning vehicle in the west approach as an example, as the
yellow lines shown in Figure 1. As the vehicle travels from x Oa = − (n · L + L 0 ) (9)
the approach lane i W to the exit lane jS , the starting point(i.e., y Oa = yW,i − Ra (10)
PW,i (x W,i , yW,i )) of its trajectory can be determined by Eq.(1),
and the end point(i.e., Q S, j (x S, j , y S, j )) can be determined by To achieve a combination between the circular curve and
Eq.(5). transition curve, the golden section method is adopted to deter-
mine the intersection point of the two curves (i.e., Ma (xa , ya )),
x S, j , y S, j = (−[( jS − 1) · L + L/2], −(n · L + L 0 )) (5)


as shown in Figure 4. The golden ratio, also known as
the “divine proportion”, appears frequently as an aesthetic
If n − i W = n − jS , the distance between the starting point standard in nature, architecture, engineering and art because
PW,i and the vertex (a4 , b4 ) is equal to the distance between of its unique properties of proportionality and harmony [42].
the end point Q S, j and the vertex (a4 , b4 ), like the trajectory Through using the golden section method, the intersection
PW,2 → Q S,2 depicted in Figure 2. In this case, the vehicle’s point Ma can be computed by Eqs. (11)-(12). The equations
entrance and exit lanes are symmetrically arranged, and the of the circular curve and transition curve can be determined
circular curve is used as vehicle’s turning trajectory, as shown by Eqs. (13) and (14), respectively.
in Figure 3. The center of the circular curve is (a4 , b4 ), the
radius Rr can be calculated by Eq. (6), and the trajectory (xa , ya ) = (− [(n · L + L 0 ) − L a ] , ya ) (11)
equation can be determined by Eq. (7). L a = Ra × 0.618 (12)

2
2
Rr = (n − i W ) · L + L/2 + L 0
 
(6) x − x Oa + y − y Oa = Ra2 (13)

1/3
(x − a4 )2 + (y − b4 )2 = Rr2 (7) y = −6Ca · x − x S, j + y S, j (14)

where the value ranges of x and y are: The value of ya in Eq. (11) can be obtained by substituting
x ∈ [min(x W,i , x S, j ), max(x W,i , x S, j )] xa into Eq. (13). Eq. (13) depicts the vehicle trajectory of
y ∈ [min(yW,i , y S, j ), max(yW,i , y S, j )] PW,i → Ma , where the value ranges of x and y are:
If n − i W > n − jS , the distance between the starting x ∈ [min(x W,i , xa ), max(x W,i , xa )]
point PW,i and the vertex (a4 , b4 ) is longer than the distance y ∈ [min(yW,i , ya ), max(yW,i , ya )]
between the end point Q S, j and the vertex (a4 , b4 ), like Eq.(14) depicts the vehicle trajectory of Ma → Q S, j , where
the trajectory PW,2 → Q S,n depicted in Figure 2. In this the value ranges of x and y are:
case, the vehicle’s entrance and exit lanes are not symmet- x ∈ [min(xa , x S, j ), max(xa , x S, j )]
rically arranged, and a two-stage trajectory planning method y ∈ [min(ya , y S, j ), max(ya , y S, j )]
is proposed that employs a combination of transition and Ca is the parameter of the transition curve, which is equal
circular curves to guarantee the smooth turning movements to the product of the radius of the circular curve and the
and traffic safety for vehicles within the intersection, as shown arc length of the transition curve. The value of Ca can be
in Figure 4. Ma (xa , ya ) is the intersection point of circular calculated by substituting the coordinates of Ma into Eq. (14).
curve and transition curve. The initial segment of the vehicle If n − i W < n − jS , the distance between the starting
trajectory (i.e., PW,i → Ma ) is defined by the circular curve, point PW,i and the vertex (a4 , b4 ) is shorter than the distance
while the subsequent segment (i.e., Ma → Q S, j ) is defined between the end point Q S, j and the vertex (a4 , b4 ), like the
by the transition curve. trajectory PW,2 → Q S,1 depicted in Figure 2. Similarly, the
The radius of the circular curve (i.e., Ra ) can be calculated two-stage trajectory planning method is proposed that employs
by Eq. (8), and the center of the circular curve (i.e., Oa ) can a combination of transition and circular curves, as shown
be determined by Eqs. (9)-(10). in Figure 5. Mb (xb , yb ) is the intersection point of the two
curves. The initial segment of the vehicle trajectory (i.e.,
Ra = (n − jS ) · L + L/2 + L 0 (8) PW,i → Mb ) is defined by the transition curve, while the

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 WU et al.: OPTIMIZING BUS OPERATIONS AT AUTONOMOUS INTERSECTION 14881

B. Bus Priority Control Model

Based on vehicle trajectory planning, a bus priority control
model for the autonomous intersection (AIM-BP) is devel-
oped. The model inputs include: the distance between the
starting point and the conflict point along the vehicle trajectory,
the approach and exit directions of each vehicle, and the
planned arrival time at the stop line for each vehicle. The pro-
Fig. 5. Trajectory III for right turning vehicles.
posed model aims to minimize the weighted sum of the total
bus delay and car delay. It simultaneously optimizes the
lane choice on the road section, the route choice within the
subsequent segment (i.e., Mb → Q S, j ) is defined by the intersection and the time to enter the intersection for each
circular curve. vehicle, while determining whether to deploy the dynamic bus
The radius of the circular curve (i.e., Rb ) can be calculated lane and which lane serves as the dynamic bus lane in the
by Eq. (15), and the center of the circular curve (i.e., Ob ) can approach direction. The objective and the constraints are as
be determined by Eqs. (16)-(17). follows.
1) The Objective: The delay for both buses and cars can
Rb = (n − i W ) · L + L/2 + L 0 (15)
be computed by Eq. (24), and the objective can be defined by
x Ob = x S, j − Rb (16) Eq. (25).
y Ob = − (n · L + L 0 ) (17)
The intersection point Mb (xb , yb ) can be computed by Eqs. dk = Tk − tk ∀k ∈ A ∪ B (24)
(18)-(19). The equations of the circular curve and transition X X
!
curve can be determined by Eqs. (20) and (21), respectively. min w db + (1 − w) da (25)
(xb , yb ) = (xb , − [(n · L + L 0 ) − L b ]) (18) b∈B a∈A

L b = Rb × 0.618 (19) where A represents the set of all upcoming cars and B

2
2 represents the set of all upcoming buses. Tk is the time point to
x − x Ob + y − y Ob = Rb2 (20) enter the intersection and tk is the planned arrival time

3 . P at the
y = − x − x W,i 6Cb + yW,i (21) stop line for buses and cars. a represents the car,Pand a∈A da
is the total car delay. b represents the bus, and b∈B db is the
The value of xb in Eq. (18) can be obtained by substituting total bus delay. Eq. (24) indicates that the delays for both buses
yb into Eq. (20). Eq.(20) depicts the vehicle trajectory of and cars are computed by subtracting the planned arrival time
Mb → Q S, j , where the value ranges of x and y are: from the entering time. The objective function in Eq. (25) is
x ∈ [min(xb , x S, j ), max(xb , x S, j )] to minimize the weighted sum of the total bus delay and car
y ∈ [min(yb , y S, j ), max(yb , y S, j )] delay, where w is a weighting factor.
Eq.(21) depicts the vehicle trajectory of PW,i → Mb , where 2) The Constraints:
the value ranges of x and y are: a) Entering time: For the buses and cars, the entering
x ∈ [min(x W,i , xb ), max(x W,i , xb )] time should be no less than the planned arrival time, i.e.,
y ∈ [min(yW,i , yb ), max(yW,i , yb )]
Cb is the parameter of the transition curve, which can be Tk ≥ tk k ∈ A∪B (26)
calculated by substituting the coordinates of Mb into Eq. (21).
For any two vehicles (e.g., vehicle k and p, k, p ∈ A∪ B) in
Similarly, the trajectory equations for the left turning and
the same approach, if they choose the same lane, the vehicle
right turning vehicles in other approaches can be derived.
in front should enter the intersection earlier than the vehicle
3) The Distance Between the Starting Point and the Conflict
behind, i.e.,
Point: Based on vehicle trajectory equations, the set of all con-
flict points can be determined, and subsequently, the distance T p ≥ Tk + L k /v
between the starting point and the conflict point along the (k, p) ∈ {(k, p)|tk < t p , Ok = O p , Ik = I p } (27)
vehicle trajectory can be calculated using Eqs. (22) and (23).
q

2
2 where L k is the body length of vehicle k and v is the vehicle
Src = x O,i − xc + y O,i − yc (22) speed. tk < t p means that k is the vehicle in front and p is
Z xc q the vehicle behind. Ok and O p are the approach directions of
Src = 1 + (yx′ )2 d x (23) vehicle k and p respectively. Ik and I p are the approach lanes
x O,i
chosen by vehicle k and p respectively.
where (x O,i , y O,i ) represents the coordinates of the starting b) The time to arrive at/leave the conflict point: For the
point, and (x c , yc ) represents the coordinates of the conflict buses and cars, trc is the time to travel from the starting point
point. Eq. (22) pertains to the trajectory of through vehicles, to the conflict point c along route r , which can be computed
and Eq. (23) pertains to the trajectory of turning vehicles. by Eq. (28).
The trajectory equations can be expressed by y(x), and yx′ in
Eq. (23) is the derivative function of y(x). trc = Src /v (28)

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 14882 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 25, NO. 10, OCTOBER 2024

Whether the bus or the car clears the conflict point can be
determined by Eqs. (29) and (30).
X
λkc = σ k µc k ∈ A ∪ B, c ∈ C (29)
r ∈R r r
X
σrk = 1, k ∈ A ∪ B (30)
r ∈R
where µrc is a binary parameter. µrc = 1 means that the conflict
point c is on the route r , and zero otherwise. σrk is a binary
variable. σrk = 1 when vehicle k chooses the route r , and Fig. 6. The implementation of dynamic bus lane.
zero otherwise. Meanwhile, each vehicle can only choose one
route to traverse the intersection, which can be specified as
Eq. (30). λkc is a binary variable. λkc = 1 when vehicle k introduction of the dynamic bus lane is under consideration,
clears the conflict point c, and zero otherwise. which would prohibit the cars ahead of the bus from selecting
If the bus or the car passes the conflict point c along its and driving in the dynamic bus lane, i.e.,
route, i.e., λkc = 1, the time to arrive at/leave the conflict point
can be determined by Eq. (31) and Eq. (32), respectively. If Ib − Ia ̸= 0 ∀a ∈ Fb and Ib = i,
then 1 Ob ,i = 1 b ∈ B, i ∈ {1, 2, . . . , n} (37)
X
tk,c = Tk + σrk trc k ∈ A ∪ B, c ∈ C (31) where Ia and Ib represent the approach lanes chosen by the
r ∈R
Tk,c = tk,c + L k /v k ∈ A ∪ B, c ∈ C (32) car a and bus b respectively. Ob is the approach direction of
bus b, and Fb is the set of all the cars preceding bus b in the
where tk,c is the time for vehicle k to arrive at the conflict approach. Eq. (37) indicates that if all the cars preceding the
point c, and Tk,c is the time for vehicle k to leave the conflict bus don’t choose the same approach lane as the bus, then
point c. the approach lane of the bus is deployed as the dynamic bus
c) Conflict separation: For any two vehicles passing the lane, as illustrated in Figure 6.
same conflict point (e.g., vehicle e and f , e, f ∈ A ∪ B), the The approach lanes chosen by the bus and the car can be
conflict between them should be avoided. determined by Eqs. (38) and (39) respectively:
When vehicle e arrives at the conflict point c earlier than
vehicle f , vehicle f should arrive at the conflict point c after X
vehicle e leaves, i.e., Ib = σrb lr b∈B (38)
r
X
Ia = σra lr a∈A (39)
Te,c − t f,c ≤ 0 e, f ∈ A ∪ B, c ∈ C (33) r
When vehicle f arrives at the conflict point c earlier than The exit lanes chosen by the bus and the car can be
vehicle e, vehicle e should arrive at the conflict point c after determined by Eqs. (40) and (41) respectively:
vehicle f leaves, i.e., X
Jb = σrb lr′ b ∈ B (40)
T f,c − te,c ≤ 0 e, f ∈ A ∪ B, c ∈ C (34) r
X
d) The number of dynamic bus lanes: The number of Ja = σra lr′ a∈A (41)
dynamic bus lanes in each approach should be no less than r
zero and no more than n − 1, where n is the number where lr is the approach lane associated with route r , and lr′
of the approach lanes in each direction. The corresponding is the exit lane associated with route r . σra is the route choice
constraints can be written as: of car a, which means whether car a chooses route r . σrb is
the route choice of bus b, which means whether bus b chooses
Xn route r .
1 O,i ≥ 0 O ∈ {E, W, S, N } (35)
Xi=1 f) The lane choices for the bus: If the dynamic bus lane
n
1 O,i ≤ n − 1 O ∈ {E, W, S, N } (36) is deployed in the approach, the bus should opt to use and
i=1
travel on it, i.e.,
where 1 O,i is a binary variable. 1 O,i = 1 when the i th lane Xn
of the approach O is a dynamic bus lane, and zero otherwise, If 1 ≤ 1 Ob ,i ≤ n − 1 and 1 Ob ,i = 0,
i=1
i ∈ {1, 2, . . . , n}. then σrb = 0 b ∈ B, i ∈ {1, 2, . . . , n}, r ∈ Ri (42)
e) The implementation of dynamic bus lane: In order
to effectively minimize bus delays and mitigate the negative where Ri is the set of the routes associated with the i th
impact of the traffic constraint described in Eq. (27), where the approach lane. Eq. (42) indicates that if there is a dynamic
following vehicle should enter the intersection after the vehicle bus lane in the approach direction of bus b and the i th lane
in front when they opt for the same approach lane. A viable in this approach is not a dynamic bus lane, then the bus b is
solution entails preventing the cars in front from sharing the not allowed to choose the routes associated with the i th lane
same approach lane with the following bus. As a result, the (i.e., Ri ), as illustrated in Figure 7.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 WU et al.: OPTIMIZING BUS OPERATIONS AT AUTONOMOUS INTERSECTION 14883

The nonlinear constraint in Eq. (27) can be rewritten into

the linear form as follows:
TK + L k /v − T p − M(Ik − I p ) ≤ M2 ρk, p (44)
TK + L k /v − T p − M(I p − Ik ) ≤ M2 (1 − ρ k, p ) (45)
where M and M2 are large positive numbers, M2 ≫ M. ρk, p is
a binary variable. ρk, p = 1 when Ik ≤ I p , and ρk, p = 0 when
Ik > I p . If vehicle k and p are on the same approach lane
(i.e., Ik = I p ), then Eq. (45) becomes TK + L k /v − T p ≤ 0,
which is consistent with Eq. (27). In this case, Eq. (44) is
automatically satisfied. If vehicle k and p are not on the same
approach lane (i.e., Ik > I p or Ik < I p ), Eq. (44) and Eq. (45)
will be automatically satisfied.
Fig. 7. The lane choice for buses. The nonlinear constraints in Eq. (31) and (32) can be
rewritten into the linear forms as follows:
X  
Tk + σrk trc − M 1 − λkc ≤ tk,c
r
X  
≤ Tk + σrk trc + M 1 − λkc (46)
r
 
tk,c + L k /v − M 1 − λkc ≤ Tk,c
 
≤ tk,c + L k /v + M 1 − λkc (47)
where M is a large positive number. If the vehicle k passes
its route(i.e., λkc = 1), Eq. (46)
the conflict point c along P
is reduced to tk,c = Tk + σrk trc , and Eq. (47) is reduced
r
to Tk,c = tk,c + L k /v. Otherwise, Eq. (46) and Eq. (47) are
automatically satisfied.
The nonlinear constraints in Eq. (33) and (34) can be
Fig. 8. The lane choice for cars.
rewritten into the linear forms as follows:
 
f
Te,c − t f,c − M 1 − λec − M 1 − λc ≤ M ye, c


g) The lane choice for the car: If there is a bus behind f (48)
the car, the car is not allowed to choose and drive on the 
f

T f,c − te,c − M 1 − λec − M 1 − λc ≤ M(1 − ye, c
f)


dynamic bus lane. If there is no bus behind the car, it is allowed
to choose and drive on the dynamic bus lane, i.e., (49)
1 − ϕ a 1 Oa ,i ≥ σra , a ∈ A, i ∈ {1, 2, . . . , n}, r ∈ Ri (43) where ye, c
f is a binary variable. For any two vehicles (e.g.,
where Oa is the approach direction of car a. ϕa is a binary vehicle e and f ) passing the same conflict point, ye, c
f = 1
parameter. ϕa = 1 means that there is a bus behind car a in the means that vehicle f arrives at the conflict point c earlier
approach, and zero otherwise. Eq. (43) indicates that if the i th than vehicle e, and zero otherwise. When vehicle e arrives
at the conflict point earlier than f (i.e., ye,c
lane is the dynamic bus lane and there is a bus behind the car, f = 0), Eq. (48)
then the car is not allowed to choose and drive on the routes becomes Te,c −t f,c ≤ 0 and Eq. (49) is automatically satisfied.
associated with the i th lane (i.e., Ri ), as illustrated in Figure 8. When vehicle f arrives at the conflict point earlier than e
c = 1), Eq. (49) becomes T
(i.e., ye,
“A1” and “A2” are the left-turning cars ahead of the bus. “A3” f f,c − te,c ≤ 0 and Eq. (48)
and “A4” are the right-turning cars ahead of the bus. “A5” is is automatically satisfied.
the right-turning car behind the bus. If lane 2 in the west The nonlinear constraint in Eq. (37) can be rewritten into
approach is deployed as the dynamic bus lane, the cars ahead the linear form as follows:
of the bus, i.e., “A1”, “A2”, “A3” and “A4”, are not allowed to g1b,a + g2b,a + g3b,a = 1, b ∈ B, a ∈ Fb (50)
choose and drive on the routes associated with lane 2, which
are denoted by the purple curves and blue curves respectively 1 − ng1b,a − g2b,a
≤ Ib − Ia ≤ g2b,a + ng3b,a −1
in Figure 8. In addition, the car behind the bus, i.e., “A5”, b ∈ B, a ∈ Fb (51)
is allowed to choose and drive on the routes associated with εb,a = g1b,a
+ g3b,a , b
∈ B, a ∈ Fb (52)
lane 2. X X
εb,a − Nb + σrb ≤ 1 Ob ,i
r ∈Ri
a∈Fb
C. Model Solving and Linearization
b ∈ B, i ∈ {1, 2, . . . , n} (53)
We can transform the nonlinear constraints into linear forms,
and the proposed model can be linearized as a mixed integer where Nb is the number of elements in the set Fb , and g1b,a ,
linear programming (MILP) problem. g2b,a , g3b,a and εb,a are binary variables. Eq. (50) indicates that

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 14884 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 25, NO. 10, OCTOBER 2024

there is only one variable equal to 1 among g1b,a , g2b,a , and TABLE II
g3b,a . In Eq. (51), g1b,a = 1 means that 1 − n ≤ Ib − Ia ≤ −1, BASIC N UMERICAL S ETTING

g2b,a = 1 means that Ib − Ia = 0, and g3b,a = 1 means that

1 ≤Ib − Ia ≤ n − 1. Therefore, if g1b,a = 1 or g3b,a = 1,
it indicates that the bus b and the car a don’t choose the same
approach lane. If g2b,a = 1, it indicates that the bus b and
the car a choose the same approach lane. In Eq. (52), εb,a =
1 when g1b,a = 1 or g3b,a = 1, and εb,a = 0 when g2b,a = 1.
Therefore, εb,a = 1 means that the bus b and car a don’t
choose the same approach lane, and zero otherwise. Therefore,
Eq. (53) indicates that if all the cars in the
P set Fb don’t choose
the same approach lane as bus b (i.e., a∈Fb εb,a − Nb = 0)
and the i th approach lane is chosen by bus b(i.e., r ∈Ri σrb =
P

1), then the i th approach lane is deployed as the dynamic bus

lane(i.e., 1 Ob ,i ≥ 1), which is consistent with Eq.(37).
The nonlinear constraint in Eq. (42) can be rewritten into
the linear form as follows:
Xn
(n − 1)δ O − 1 O,i ≥ 0, O ∈ {E, W, S, N } (54)
i=1
Xn
δO ≤ 1 O,i , O ∈ {E, W, S, N } (55)
i=1
σrb ≤ (1 − δ Ob ) + 1 Ob ,i b ∈ B, i ∈ {1, 2, . . . , n}, r ∈ Ri
(56)

where δ O is a binary variable. From Eq. (54) and (55), Fig. 9. Comparison of the trajectory.
n
it can be inferred that δ O = 0 when 1 O,i = 0 and
P
i=1
n
δ O = 1 when 1 ≤ 1 O,i ≤ n − 1. Thereby, δ O = 1 means
P
i=1
that the dynamic bus lane is deployed in the approach O,
and zero otherwise. Therefore, Eq.(56) indicates that if the
dynamic bus lane is deployed in the approach direction of
bus b (i.e., 1 − δ Ob = 0) and the i th lane in this approach is
not a dynamic bus lane(i.e., 1 Ob ,i = 0), then the bus b is not
allowed to choose the routes associated with the i th lane(i.e.,
σrb ≤ 0), which is consistent with Eq.(42).
After linearization, all the nonlinear constraints in the model
are transformed into linear forms, thus a Mixed-Integer Lin-
ear Programming (MILP) model considering bus priority is
established.
Fig. 10. Comparison of vehicle heading angle.
The MILP model can be solved by the branch and bound
method. The decision variables, the objective function and the
constraints in the model are as follows.
IV. N UMERICAL E XAMPLES
Decision Variables: The time for vehicle k to enter the
intersection (i.e., Tk ), whether the vehicle k chooses the route A. The Result of the Proposed Model
r (i.e., σrk ), and whether the i th lane of the approach O is a
dynamic bus lane (i.e., 1 O,i ). The numerical test is conducted based on the parameters
Objective Function: Minimization of the weighted com- of the four-way intersection between Xinyao North Road
binations
 of the total bus delay and total car delay, i.e., and Xinning Road in Changsha City, China, which has two
approach lanes and two exit lanes in each direction. The basic
min w b∈B db + (1 − w)
P P
da .
a∈A
numerical settings are listed in Table II.
Constraints: Eq. (26), (28)-(30), (35)-(36), (38)-(41), To assess the efficacy of the trajectory planning method
(43)-(56). introduced in this paper, we conduct a comparative analysis
The model is compiled in AMPL (A Mathematical Program- between the proposed curve and the conventional elliptic
ming Language) and solved by CPLEX on a desktop computer curve utilized in [17], [18], and [20]. As an illustrative
with Win-10 64-bit operating system and Montage Jintide(R) example, let’s consider a left-turning vehicle’s trajectory.
C5218R, 2.10 GHz, 16.0 GB. In Figure 9, 10, and 11, we present a comparison of the

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 WU et al.: OPTIMIZING BUS OPERATIONS AT AUTONOMOUS INTERSECTION 14885

TABLE III
T HE I NFORMATION OF THE U PCOMING V EHICLES

The objective function of the proposed model (AIM-BP) is

to minimize the weighted combinations of the total bus delay
and total car delay. Considering the difference in passenger
capacity between buses and cars, the value of the weighting
factor w is set at 0.8, which is conservative, and the sensitivity
analysis on this factor is presented in the subsequent section.
After solving the model, the outputs are shown in Table IV
and V, including the time to enter the intersection for each
vehicle (i.e., Tk ), the vehicle delay (i.e., dk ), the route choice
(the approach lane and the exit lane chosen by each vehicle),
and whether to deploy the dynamic bus lane in each approach.
Fig. 11. Comparison of the curvature. From Table V, it can be found that lane 2 in the north
approach and lane 1 in the east approach are deployed as
the dynamic bus lane. This is because when the dynamic bus
vehicle’s trajectory, heading angle, and curvature between the lane is deployed, it prohibits the cars ahead of the bus from
two curves, respectively. selecting and driving in the same lane as the bus, thus reducing
In Figure 9, it can be found that the trajectory of the pro- the bus queueing delays at the intersection.
posed curve is smoother than the elliptic curve. In Figure 10,
the vehicle’s heading angle generally increases from 0 to
π/2. The heading angle along the proposed curve is smaller
B. Model Comparison
compared to the elliptic curve at the beginning, while becom-
ing larger compared to the elliptic curve near the exit lane. To examine the effectiveness of the proposed bus priority
In Figure 11, compared to the elliptic curve, the curvature of control model in this paper (AIM-BP), we add the regular
the proposed curve is smaller near the approach and exit lanes, AIM model (AIM-R) without considering bus priority for
while it is larger within the central region of the intersection. comparative analysis. Under AIM-R, the objective function is
It can be inferred that the advantage of the proposed curve to minimize the total delay of all the vehicles, without taking
over the elliptic curve is as follows: when the vehicle travels into account any constraints related to bus priority control
along the proposed curve, its heading angle will experience a (i.e., the constraints on the number of dynamic bus lanes, the
smoother transition when it enters and exits the intersection, implementation of dynamic bus lane, and the lane choices for
and the vehicle primarily completes the turning maneuver the buses and cars). Figure 12 and 13 illustrate the comparison
within the central region of the intersection, which ensures of vehicle delays between the two models.
the smooth turning movements and traffic safety for vehicles In Figure 12 and 13, vehicle 3, 8, 11 and 16 are buses,
to traverse the intersection. and other vehicles are cars. Compared with AIM-R, the
The arrival of the vehicles is assumed to follow the Poisson implementation of AIM-BP leads to a significant decrease in
distribution. Take a random snapshot of vehicles approaching the delays of all the buses, where the delays of vehicle 11 and
the intersection at a certain time and the information of the 16 are reduced by 1.28s and 1.56s respectively. This is because
vehicles can be obtained, including the vehicle type, the under AIM-BP, while the priority is accorded to the buses, the
approach and exit directions of each vehicle, the planned cars in front of the bus are not allowed to choose and drive
arrival time at the stop line for each vehicle (i.e., tk ), and the on the dynamic bus lane, and vehicles’ entry times and route
set of all the cars preceding a bus in the approach (i.e., Fb ), choices within the intersection are optimized to reduce the
as shown in Table III. delays for buses.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 14886 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 25, NO. 10, OCTOBER 2024

TABLE IV
T HE O PTIMIZATION R ESULT OF THE P ROPOSED M ODEL

TABLE V
DYNAMIC B US L ANE D EPLOYMENT IN E ACH A PPROACH

Fig. 13. The comparison of vehicle delay distribution.

Fig. 12. The comparison of vehicle delays.

to vehicles 11 (bus) and 16 (bus), respectively. This further

As shown in Figure 12, it is noteworthy that under reinforces the effectiveness of AIM-BP in improving bus effi-
AIM-R, the maximum vehicle delay is 2.45s, corresponding to ciency. In addition, the average vehicle delays under AIM-BP
vehicle 16(bus), whereas under AIM-BP, the maximum vehicle and AIM-R are 0.49s and 0.35s respectively. This can be
delay is 1.62s, corresponding to vehicle 15 (car). In addition, attributed to the bus priority control under AIM-BP, which
under AIM-BP, delays exceeding 1s are observed for vehi- inevitably leads to a moderate increase in the total car delay.
cles 7, 12, 14, and 15, all of which are cars. However, under As a result, the average vehicle delay is slightly increased
AIM-R, delays exceeding 1s are observed for vehicles 11 and under AIM-BP.
16, both of which are buses. As shown in Figure 13, the The average car delay, the average bus delay and the
outliers in AIM-R are delays of 1.28s and 2.45s, corresponding average passenger delay (assuming the average occupancies

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 WU et al.: OPTIMIZING BUS OPERATIONS AT AUTONOMOUS INTERSECTION 14887

TABLE VI
S ENSITIVITY A NALYSIS ON w

Fig. 15. Sensitivity analysis on w.

Fig. 14. The comparison of average delay.

of 40 passengers per bus and 1.25 passengers per car [29]) from 0.5 to 1, the corresponding changes in average bus
under AIM-BP and AIM-R are presented in Figure 14. delay, average car delay and average passenger delay can be
Figure 14 illustrates that, while the average car delay observed, as illustrated in Table VI and Figure 15.
increases under AIM-BP compared to AIM-R, there is a From Table VI and Figure 15, it can be found that with the
notable decrease in both average bus delay and average increase in w, there is an increasing trend in average car delay,
passenger delay. The average bus delays under AIM-BP and while a decreasing trend in average bus delay and average
AIM-R are 0.28s and 1.15s respectively, and the average passenger delay. When w = 0.5, the values of the key metrics
passenger delays under AIM-BP and AIM-R are 0.30s and are equal to those in AIM-R. This is because when w = 0.5,
1.06s respectively. The implementation of AIM-BP results the weight assigned to total bus delay is equal to that assigned
in a substantial reduction of 75.65% in average bus delay to total car delay in the objective function. When the value of
and 71.7% in average passenger delay when compared to w is increased from 0.5 to 0.55, the average bus delay and
AIM-R, which performs well under a conservative setting average passenger delay is decreased by 6.96% and 6.60%
of the weighting factor w. This verifies the effectiveness of respectively. This is because the value of w is relatively small,
AIM-BP in reducing both average bus delay and average the priority accorded to the bus is not sufficient. Thus, there
passenger delay. is only a slight decrease in the average bus delay and average
passenger delay. When the value of w is increased from 0.55 to
0.7, there is a significant decrease in the average bus delay and
C. Sensitivity Analysis average passenger delay by 73.83% and 69.70% respectively,
Under AIM-BP, we analyze the impact of the weighting while the average car delay is only increased by 0.45s. When
factor w on several key metrics, including average bus delay, the value of w is increased from 0.8 to 0.85, the average bus
average car delay, and average passenger delay. Notably, w delay and average passenger delay is decreased by 25% and
represents the weight assigned to total bus delay. Given that 16.67% respectively. When w = 1, it can be found that there
AIM-BP incorporates bus priority control, the weight assigned is a significant increase in average car delay, which leads to
to total bus delay should be no less than that assigned to an increase in average passenger delay. This is because when
total car delay in the objective function. Therefore, we set w = 1, the objective function is reduced to the minimization
the value range of w between 0.5 and 1. By increasing w of total bus delay, with no consideration for total car delay.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 14888 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 25, NO. 10, OCTOBER 2024

V. C ONCLUSION [2] S. Ilgin Guler, M. Menendez, and L. Meier, “Using connected vehicle
technology to improve the efficiency of intersections,” Transp. Res. C,
Bus priority control at the intersection represents a sig- Emerg. Technol., vol. 46, pp. 121–131, Sep. 2014.
nificant approach to reduce bus delay, make better use of [3] M. Eom and B.-I. Kim, “The traffic signal control problem for inter-
spatial-temporal road resources and alleviate urban traffic con- sections: A review,” Eur. Transp. Res. Rev., vol. 12, no. 1, pp. 1–20,
Dec. 2020.
gestion. In recent years, the rapid advancement of autonomous [4] Y. Xiao, Y. An, T. Li, N. Wu, W. He, and P. Li, “Autonomous driving
driving technology has given rise to a novel intersection con- policy learning from demonstration using regression loss function,”
trol method, known as autonomous intersection management Knowl.-Based Syst., vol. 295, Jul. 2024, Art. no. 111766.
[5] K. Dresner and P. Stone, “Multiagent traffic management: An improved
(AIM), which holds the potential to significantly improve intersection control mechanism,” in Proc. 4th Int. Joint Conf. Auto.
traffic efficiency at the intersection. However, most existing Agents Multiagent Syst., Jul. 2005, pp. 471–477.
studies on AIM ignore the bus priority demands. [6] K. Dresner and P. Stone, “A multiagent approach to autonomous
Therefore, this paper optimizes bus operations at intersection management,” J. Artif. Intell. Res., vol. 31, pp. 591–656,
Mar. 2008.
autonomous intersection with trajectory planning and priority [7] E. Namazi, J. Li, and C. Lu, “Intelligent intersection management sys-
control. First, this paper puts forward a new trajectory planning tems considering autonomous vehicles: A systematic literature review,”
approach for turning movements within the intersection that IEEE Access, vol. 7, pp. 91946–91965, 2019.
[8] G. Wu and R. Jiang, “Joint optimization for autonomous intersection
takes into account the significant passenger volume and management and trajectory smoothing design with connected auto-
large size of buses. A two-stage trajectory planning method mated vehicles,” Transportmetrica B, Transp. Dyn., vol. 11, no. 1,
is proposed that employs a combination of transition and pp. 1234–1255, Dec. 2023.
circular curves to ensure the smooth turning movements [9] Z. Yao, H. Jiang, Y. Jiang, and B. Ran, “A two-stage optimiza-
tion method for schedule and trajectory of CAVs at an isolated
and traffic safety for vehicles within the intersection. autonomous intersection,” IEEE Trans. Intell. Transp. Syst., vol. 24,
Meanwhile, the golden section method is used to determine no. 3, pp. 3263–3281, Mar. 2023.
the intersection point of the above two types of curves. The [10] A. Mirheli, L. Hajibabai, and A. Hajbabaie, “Development of a
signal-head-free intersection control logic in a fully connected and
efficacy of the proposed trajectory planning method is verified autonomous vehicle environment,” Transp. Res. C, Emerg. Technol.,
compared with the conventional elliptic curve in the numerical vol. 92, pp. 412–425, Jul. 2018.
analysis. Second, based on vehicle trajectory planning, a bus [11] D. Miculescu and S. Karaman, “Polling-systems-based autonomous
vehicle coordination in traffic intersections with no traffic signals,” IEEE
priority control model for the autonomous intersection Trans. Autom. Control, vol. 65, no. 2, pp. 680–694, Feb. 2020.
is developed. The proposed model aims to minimize the [12] M. W. Levin and D. Rey, “Conflict-point formulation of intersection
weighted combinations of the total bus delay and car delay. control for autonomous vehicles,” Transp. Res. C, Emerg. Technol.,
In particular, the model incorporates the introduction of the vol. 85, pp. 528–547, Dec. 2017.
[13] S. A. Fayazi and A. Vahidi, “Mixed-integer linear programming for
dynamic bus lane designed to clear the cars in front of buses, optimal scheduling of autonomous vehicle intersection crossing,” IEEE
thereby creating a relatively exclusive space for the buses. Trans. Intell. Vehicles, vol. 3, no. 3, pp. 287–299, Sep. 2018.
The proposed model simultaneously optimizes the lane choice [14] Z. Li, M. Pourmehrab, L. Elefteriadou, and S. Ranka, “Intersection
control optimization for automated vehicles using genetic algorithm,”
on the road section, the route choice within the intersection J. Transp. Eng., A, Syst., vol. 144, no. 12, pp. 12–26, Dec. 2018.
and the time to enter the intersection for each vehicle, while [15] A. Mirheli, M. Tajalli, L. Hajibabai, and A. Hajbabaie, “A consensus-
determining whether to deploy the dynamic bus lane and based distributed trajectory control in a signal-free intersection,”
which lane serves as the dynamic bus lane in the approach Transp. Res. C, Emerg. Technol., vol. 100, pp. 161–176, Mar. 2019.
[16] B. Liu, Q. Shi, Z. Song, and A. El Kamel, “Trajectory planning for
direction. Finally the proposed model is formulated as a autonomous intersection management of connected vehicles,” Simul.
mixed integer linear programming (MILP) problem, which is Model. Pract. Theory, vol. 90, pp. 16–30, Jan. 2019.
compiled in AMPL (A Mathematical Programming Language) [17] W. Wu, Y. Liu, W. Liu, F. Zhang, V. Dixit, and S. T. Waller,
“Autonomous intersection management for connected and automated
and solved by CPLEX. The results indicate that the proposed vehicles: A lane-based method,” IEEE Trans. Intell. Transp. Syst.,
model significantly reduces average bus delay and average vol. 23, no. 9, pp. 15091–15106, Sep. 2022.
passenger delay at the intersection. The sensitivity analysis [18] W. Wu, S. Chen, M. Xiong, and L. Xing, “Enhancing intersection safety
further examines the effectiveness of the proposed model. in, autonomous traffic: A grid-based approach with risk quantification,”
Accident Anal. Prevention, vol. 200, Jun. 2024, Art. no. 107559.
In addition, this paper can be extended in several ways. [19] Z. He, L. Zheng, L. Lu, and W. Guan, “Erasing lane changes from roads:
First, as this paper only focuses on the bus priority control A design of future road intersections,” IEEE Trans. Intell. Vehicles,
at a single intersection, future studies can be conducted on vol. 3, no. 2, pp. 173–184, Jun. 2018.
[20] W. Wu, Y. Liu, W. Hao, G. A. Giannopoulos, and Y.-J. Byon,
the coordinated bus priority control between multiple inter- “Autonomous intersection management with pedestrians crossing,”
sections. Second, this paper only considers the upcoming Transp. Res. C, Emerg. Technol., vol. 135, Feb. 2022, Art. no. 103521.
buses and cars at the autonomous intersection, while the [21] M. W. Levin and S. D. Boyles, “Intersection auctions and reservation-
based control in dynamic traffic assignment,” Transp. Res. Rec., J.
traffic demands of emergency vehicles and pedestrian crossing Transp. Res. Board, vol. 2497, pp. 35–44, 2015.
are not incorporated. Thus, future studies may formulate a [22] M. Bashiri and C. H. Fleming, “A platoon-based intersection manage-
multi-modal priority control model under the mixed-traffic ment system for autonomous vehicles,” in Proc. IEEE Intell. Vehicles
flow scenario with cars, buses, emergency vehicles, and pedes- Symp. (IV), Jun. 2017, pp. 667–672.
[23] H. Wang, Q. Meng, S. Chen, and X. Zhang, “Competitive and cooper-
trians at the autonomous intersection. ative behaviour analysis of connected and autonomous vehicles across
unsignalised intersections: A game-theoretic approach,” Transp. Res. B,
Methodol., vol. 149, pp. 322–346, Jul. 2021.
R EFERENCES
[24] S. Liang et al., “Optimal control to improve reliability of demand
[1] C. Yu, Y. Feng, H. X. Liu, W. Ma, and X. Yang, “Integrated optimization responsive transport priority at signalized intersections considering
of traffic signals and vehicle trajectories at isolated urban intersections,” the stochastic process,” Rel. Eng. Syst. Saf., vol. 218, Feb. 2022,
Transp. Res. B, Methodol., vol. 112, pp. 89–112, Jun. 2018. Art. no. 108192.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.
 WU et al.: OPTIMIZING BUS OPERATIONS AT AUTONOMOUS INTERSECTION 14889

[25] J. Li, Y. Liu, N. Zheng, L. Tang, and H. Yi, “Regional coordinated Wei Wu is currently a Professor with Chongqing
bus priority signal control considering pedestrian and vehicle delays at Key Laboratory of Intelligent Integrated and Mul-
urban intersections,” IEEE Trans. Intell. Transp. Syst., vol. 23, no. 9, tidimensional Transportation System, Chongqing
pp. 16690–16700, Sep. 2022. Jiaotong University. His primary research interests
[26] X. Zeng, Y. Zhang, J. Jiao, and K. Yin, “Route-based transit signal include traffic control and management within con-
priority using connected vehicle technology to promote bus sched- nected and autonomous vehicles environment.
ule adherence,” IEEE Trans. Intell. Transp. Syst., vol. 22, no. 2,
pp. 1174–1184, Feb. 2021.
[27] N. T. Imran, A. Girijan, and L. D. Vanajakshi, “A review on bus
signal priority systems,” Transp. Developing Economies, vol. 8, no. 1,
pp. 1–13, Apr. 2022.
[28] P. G. Furth and T. H. J. Müller, “Conditional bus priority at signalized
intersections: Better service with less traffic disruption,” Transp. Res. Mengfei Xiong received the bachelor’s degree from
Rec., J. Transp. Res. Board, vol. 1731, no. 1, pp. 23–30, Jan. 2000. Changsha University of Science and Technology in
[29] E. Christofa, I. Papamichail, and A. Skabardonis, “Person-based traffic 2022, where she is currently pursuing the master’s
responsive signal control optimization,” IEEE Trans. Intell. Transp. Syst., degree with the School of Traffic and Transportation
vol. 14, no. 3, pp. 1278–1289, Sep. 2013. Engineering. Her primary research interests include
[30] J. Hu, B. B. Park, and Y.-J. Lee, “Transit signal priority accommodating autonomous intersection management.
conflicting requests under connected vehicles technology,” Transp. Res.
C, Emerg. Technol., vol. 69, pp. 173–192, Aug. 2016.
[31] P. Anderson and C. F. Daganzo, “Effect of transit signal priority on
bus service reliability,” in Proc. 23rd Int. Symp. Transp. Traffic Theory
(ISTTT), 2019, pp. 2–14.
[32] X. Zhang, Z. He, Y. Zhu, and L. You, “DRL-based adaptive signal
control for bus priority service under connected vehicle environment,”
Transportmetrica B, Transp. Dyn., vol. 11, no. 1, pp. 1455–1477, Tangzhi Liu is currently a Professor with
Dec. 2023. Chongqing Key Laboratory of Intelligent Inte-
[33] A. Russo, M. W. Adler, and J. N. van Ommeren, “Dedicated bus lanes, grated and Multidimensional Transportation System,
bus speed and traffic congestion in Rome,” Transp. Res. A, Policy Pract., Chongqing Jiaotong University. His research inter-
vol. 160, pp. 298–310, Jun. 2022. ests include mountain road traffic safety, intelligent
[34] N. Dadashzadeh and M. Ergun, “Spatial bus priority schemes, imple- transportation, emergency rescue, and integration of
mentation challenges and needs: An overview and directions for future safety control and rescue on mountain highway.
studies,” Public Transp., vol. 10, no. 3, pp. 545–570, Dec. 2018.
[35] J. Viegas and B. Lu, “Traffic control system with intermittent bus lanes,”
IFAC Proc. Volumes, vol. 30, no. 8, pp. 865–870, Jun. 1997.
[36] M. Eichler and C. F. Daganzo, “Bus lanes with intermittent priority:
Strategy formulae and an evaluation,” Transp. Res. B, Methodol., vol. 40,
no. 9, pp. 731–744, Nov. 2006. Jian Sun received the Ph.D. degree from Tongji
[37] W. Wu, L. Head, S. Yan, and W. Ma, “Development and evaluation of University in 2006. He was with Tongji University
bus lanes with intermittent and dynamic priority in connected vehicle as a Lecturer and then promoted to a Full Profes-
environment,” J. Intell. Transp. Syst., vol. 22, no. 4, pp. 301–310, sor in 2011. He is currently a Professor with the
Jul. 2018. College of Transportation Engineering and the Chair
[38] J. Hu, Z. Zhang, Y. Feng, Z. Sun, X. Li, and X. Yang, “Transit of the Department of Traffic Engineering, Tongji
signal priority enabling connected and automated buses to cut through University. His main research interests include traffic
traffic,” IEEE Trans. Intell. Transp. Syst., vol. 23, no. 7, pp. 8782–8792, flow theory, traffic simulation, and connected and
Jul. 2022. autonomous vehicles.
[39] Z. Zhang, J. Lai, C. Ma, J. Zhu, and X. Yang, “Ensuring absolute transit
priority through trajectory based control of connected and automated
traffic,” in Proc. 6th Int. Conf. Transp. Inf. Saf. (ICTIS), Oct. 2021,
pp. 1132–1135.
[40] X. Chen, X. Lin, M. Li, and F. He, “Network-level control of hetero- Yongfu Li (Senior Member, IEEE) is currently
geneous automated traffic guaranteeing bus priority,” Transp. Res. C, a Professor with the School of Automation,
Emerg. Technol., vol. 140, Jul. 2022, Art. no. 103671. Chongqing University of Posts and Telecommunica-
[41] L. Zhang, G. Qian, Z. Song, and D. Wang, “Deploying dedicated tions. His main research interests include connected
lanes for connected and autonomous buses in urban transportation vehicles, intelligent transportation systems, intel-
networks,” Transportmetrica A, Transp. Sci., vol. 19, no. 2, Mar. 2023, ligent sensing and automated control, artificial
Art. no. 2005181. intelligence, and big data.
[42] I. Crăciun, D. Inoan, D. Popa, and L. Tudose, “Generalized golden
ratios defined by means,” Appl. Math. Comput., vol. 250, pp. 221–227,
Jan. 2015.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on November 11,2024 at 00:05:31 UTC from IEEE Xplore. Restrictions apply.