Review Article

Proc IMechE Part D:

J Automobile Engineering
2021, Vol. 235(7) 1808–1824
A dynamic trajectory planning method Ó IMechE 2020
Article reuse guidelines:
for lane-changing maneuver of sagepub.com/journals-permissions
DOI: 10.1177/0954407020982712

connected and automated vehicles journals.sagepub.com/home/pid

Xiao Liu , Jun Liang and Junwei Fu

Abstract
This paper describes a dynamic trajectory planning method for lane-changing maneuver of connected and automated
vehicles (CAVs). The proposed dynamic lane-changing trajectory planning (DLTP) model adopts vehicle-to-vehicle (V2V)
communication to generate an automated lane-changing maneuver with avoiding potential collisions and rollovers during
the lane-changing process. The novelty of this method is that the DLTP model combines a detailed velocity planning
strategy and considers more complete driving environment information. Besides, a lane-changing safety monitoring algo-
rithm and a lane-changing starting-point determination algorithm are presented to guarantee the lane-changing safety,
efficiency and stability of automated vehicles. Moreover, a trajectory-tracking controller based on model predictive con-
trol (MPC) is introduced to make the automated vehicle travel along the reference trajectory. The field traffic data from
NGSIM are selected as the target dataset to simulate a real-world lane-changing driving environment. The simulations
are performed in CarSim-Simulink platform and the experimental results show that the proposed method is effective for
lane-changing maneuver.

Keywords
Trajectory planning, connected and automated vehicles, lane-changing maneuver, model predictive control

Date received: 31 May 2020; accepted: 9 November 2020

Introduction recognize the surrounding environment by vehicular

sensors, such as Millimeter wave radars and camera
Employing autonomous driving technology to improve sensors, but limited distance measurement of sensors
traffic safety and efficiency has been one of the most would affect the acquisition of needed data.16 However,
extensive research topics in academia and industry.1–8 CAVs are intelligent vehicles that equipped with V2V
Compared with the lane-keeping process, the lane- communication. Through V2V technology, CAVs are
changing process of autonomous vehicles is more com- allowed not only to collect more sufficient driving
plicated. Especially in a dynamic traffic environment, a
information, but also to collaborate with surrounding
wrong lane-changing decision, unreasonable path plan-
vehicles.17,18 Combining the above information, the
ning or invalid control algorithm may result in traffic
decision-making module can make reasonable lane-
congestion and accidents.9–12 Therefore, it is necessary
changing instructions.9,19
to conduct research on the active lane-changing system.
This paper mainly focuses on the lane-changing
A complete lane-changing system could be divided
maneuver of CAVs, which includes path planning and
into four modules: a sensory and communication mod-
path following. The details of the lane-changing deci-
ule, a lane-changing decision module, a trajectory plan-
sion-making process will not be covered in this article.
ning module and an execution module.13 The driving
Some researchers have studied the lane-changing
environmental information (such as the vehicle speed
and acceleration on vehicles in the adjacent lane and
the current lane, the relative distance between vehicles) State Key Lab of Industrial Control Technology, Zhejiang University,
can be observed and collected through the sensory and Hangzhou, China
communication module. Especially in recent years, the
Corresponding author:
emergence of the connected and automated vehicles Jun Liang, School of Zhejiang University, No. 38, Zheda Road, Hangzhou,
(CAVs) has greatly accelerated the development of the Zhejiang Province 310027, China.
perception module.14,15 Traditional automated vehicles Email: [email protected]
Liu et al. 1809

trajectory planning (LTP) algorithms, but the lane focuses more on ‘‘data-driven,’’ but lack of solid theore-
change environments assumed by the above algorithms tical approval.
are static, which means that the speed of the surround- Based on the above algorithms, this paper proposes
ing vehicles remains constant throughout the pro- a dynamic trajectory planning method for lane-
cess.20–24 The above situations are not consistent with changing maneuver of CAVs. The main contributions
the real-world lane-changing environment. Moreover, are as follows:
the algorithms mentioned above do not reflect their
adaptability to the dynamical state changes of the sur- (1) A detailed velocity planning strategy in the DLTP
rounding vehicles, which may result in accidents when process is proposed based on complete driving
the intelligent vehicles encounter emergencies during environment information. The proposed strategy
the lane-changing maneuver. ensures that velocity of CAVs does not change
To overcome these limitations, some researchers greatly when a potential driving safety risk occurs.
have focused on the dynamic lane-changing trajectory (2) During the DLTP process, a lane-changing safety
planning (DLTP) models. Luo et al.24 proposed a monitoring (LSM) algorithm and a lane-changing
DLTP model for the first time. The model could cap- starting-point determination (LSD) algorithm are
ture real-time information of the surrounding vehicles adopted to detect whether the automated vehicle’s
through V2V communication and adopt the informa- current speed and the planned final longitudinal
tion to plan a dynamic lane-changing trajectory. position harmonize the conditions for safe driving.
Furthermore, Yang et al.25 proposed a novel DLTP The detection range of these algorithms includes
model, which could guarantee the lane-changing safety vehicles in both target and current lane.
under emergency braking. However, these two models (3) To validate the feasibility of the dynamic trajec-
lacked a detailed description of speed planning in the tory and the velocity planning method presented
lane-changing process. Luo et al.’s model only limited in this study, model predictive control (MPC)
the range of speed. Yang et al.’s model introduced the algorithm is introduced to simulate the various
rules of speed change, but it ignored detailed regulation lane-changing circumstances.
instructions and the speed selection was not optimal. In
addition, Yang et al.’s model took into account the sit- The rest of the paper is organized as follows. Section
uation of collision-avoidance and rollover-avoidance, 2 describes the design of the dynamic lane-changing
but the model only paid attention to vehicles on the tar- trajectory planning process, which includes trajectory
get lane, neglecting the current lane vehicles’ influence curve and planning algorithms; In section 3, the simula-
on the safety of lane-changing maneuver. Jiang et al.26 tion and experimental results of automated lane-
focused on the trajectory planning method of intelligent changing maneuver based on CarSim-Simulink plat-
vehicle lane changing emergently. The model divides form are exhibited. Section 4 presents the paper’s
the emergency lane-changing process into the initial conclusions.
stage and tracking stage for trajectory planning based
on road steering experiment and sigmoid functions.
Nevertheless, the model assumes that there are no inter- Dynamic lane-changing trajectory
ference vehicles in the adjacent lane, which obviously planning process
does not accord with the actual situation. Zheng et al.13
decoupled the longitudinal and the lateral motion plan- As shown in Figure 1, lane-changing starting-point
ning to realize the trajectory re-planning in a normal determination algorithm, lane-changing safety monitor-
lane-changing process, while the velocity determination ing algorithm and dynamic lane-changing trajectory
process of this model contained many uncertain para- planning algorithm are the main components in the
meters. Ding et al.27 proposed an integrated lane- DLTP model. In this process, when receiving the lane-
change trajectory planning method for CAVs. In this changing decision, CAVs adopt the lane-changing
model, all feasible reference trajectories compose a tra- starting-point determination algorithm to determine
jectory map, which includes different driving situations. whether a safety lane change can be completed based
Within the derived trajectory map, an optimal lane- on the current position and the velocity information.
change trajectory eliminating potential collisions is cal- After that, the initial reference trajectory will be gener-
culated by minimizing the quantitative indicators. ated. During each sampling period before the lane
Under various road line-types, Wang et al.28 con- changing is complete, if the current vehicle information
structed a lane-changing trajectory planning method to (such as the velocity, the terminal position) meets the
generate the trajectory. However, both methods con- requirements of the lane-changing safety monitoring
sider some unnecessary trajectory re-planning pro- algorithm, the automated vehicle would maintain the
cesses, which may increase the time consumption costs current state and continue to track the trajectory.
of the motion planner. Moreover, by collecting the Otherwise, this system will call the dynamic lane-
drivers’ vision data, Zhou et al.29 proposed a hyperbolic changing trajectory planning algorithm to re-plan the
tangent lane-change trajectory model to describe driv- trajectory. The new determined velocity and terminal
ers’ detailed lane-change trajectories. This model position will generate a new reference trajectory. It
1810 Proc IMechE Part D: J Automobile Engineering 235(7)

Figure 1. Framework of a proposed DLTP process.

must be noted that the input information for these two

algorithms is the real-time states of the surrounding
vehicles, which can be obtained by V2V communica-
tion. Following the dynamic planned trajectory, CAVs
Figure 2. The DLTP process: (a) an example of the complete
can safely complete the lane-changing maneuver.
DTLP process and (b) the final trajectory.
Figure 2(a) displays an example of the complete
DLTP process. For each time step, the lane-changing
system means that the initial coordinate of each sam-
trajectories of CAVs are planned according to the infor-
pling period is set as (0,0), and the coordinates of the
mation of surrounding vehicles. The final trajectory is
vehicles around the automated vehicle (AV) are con-
formed in Figure 2(b).
verted to the current coordinate system. The dynamic
trajectories are all formulated in the moving coordinate
Lane-changing trajectory curve system unless otherwise noted. Therefore, for the start-
ing point (0,0) of each sampling step, there is,
In Yang et al.’s25 research, a time-independent cubic
polynomial equation was introduced to characterize ys (0) = a0 = 0, ð2Þ
the lane-changing trajectory curve, which could loosen 0

the unreal assumptions on velocity and acceleration of ys (xis ) = tan ui , ð3Þ

Luo et al.’s24 model. In addition, compared with the ui indicates the initial course angle of the trajectory.
higher order polynomials, the cubic polynomial equa- (xfs , yfs ) is the destination coordinate of the lane-changing
tion requires fewer parameters, which reduces the com- trajectory. The subject vehicle should satisfy the dynamic
putational burden. Therefore, this paper still uses the constraints of the termination state, which means that
following cubic polynomial equation to represent the the lateral acceleration of the vehicle is 0. Thus,
lane-changing trajectory curve,
yfs = ys (xfs ) = a0 + a1 xfs + a2 xfs 2 + a3 xf3
s , ð4Þ
ys (xs ) = a0 + a1 xs + a2 xs 2 + a3 xs 3 , ð1Þ
0
ys (xfs ) = 0: ð5Þ
where, xs and ys respectively denote the longitudinal
and the lateral positions of the subject vehicle, and a0, Combining the formulas (2)–(5), the following equa-
a1, a2, and a3 are the parameters of the trajectory curve. tion can be obtained,
Furthermore, the moving coordinate system is used to
study this dynamic trajectory planning process. The Q  a = b, ð6Þ
Liu et al. 1811

Figure 3. The scenario before the AV enters the target lane.

Figure 4. The scenario after the AV enters the target lane.

2 3
1 0 0 0 article include collision-avoidance, rollover-avoidance,
60 1 0 0 7 comfort and traffic efficiency.
where Q=64 1 xfs xfs 2 xfs 3 5
7 and
2
0 1 2xfs 3xfs
f T Collision-avoidance. The lane-changing maneuver can be
b = ½0, tanui , ys , 0 . Thus, trajectory parameters a = divided into two stages: before the AV enters the target
½a1 , a2 , a3 , a4 T can be quickly calculated by using for- lane (Figure 3) and after the AV changes into the target
mula (6) and the trajectory equation is, lane (Figure 4). In these two figures, the solid vehicles
indicate the final positions of the vehicles and the
3yfs  2xfs tanui dashed vehicles show the trajectory of lane-changing.
yis (xis ) = tanui xis +  2
xfs The lane-changing line is the long dashed line that
ð7Þ divides the current lane and the target lane. Before
xfs tanui  2yfs changing into the target lane, the AV should prevent
xis 2 + 3
xi3
s :
(xfs ) itself from colliding with the preceding vehicle (PC)
and the following vehicle (FC) in the current lane
The coordinate information of both the AV and the simultaneously, and maintain a safe distance (PT and
surrounding vehicles can be obtained through V2V FT) between the preceding and following vehicles in
communication in each time step. This paper assumes the target lane.
that the final lateral position yfs is the centerline of tar- After changing into the target lane, the safe distance
get lane and is a known parameter like ui . Thus, of PC and FC can be ignored to reduce the calculation
according to equation (7), the trajectory is uniquely amount. Since lane-changing maneuver involves two
determined according to the final longitudinal position lanes and surrounding vehicles, the course angle of
xfs . lane-changing on the highway should not be too large.
Figure 5(a) shows the course angle variation of a vehi-
cle during the lane-changing process, which is simu-
Performance specification of the lane-changing
lated based on the real driving dataset NGSIM.
trajectory As shown in Figure 5(a), the course angle of a vehi-
Safety includes two aspects: (1) avoid collisions with cle is generally small during the lane-changing maneu-
surrounding vehicles and (2) avoid vehicle rollovers. ver. According to the geometric relationship of velocity
On the premise of safety, the comfort of passengers shown in
and the traffic efficiency are also the goals this research Figure 5(b), when course angle ui is relatively small,
needs to consider while planning the lane-changing tra- the vehicle’s longitudinal velocity vn can be approxi-
jectory. Therefore, the performance criteria in this mately equal to the velocity us , Thus,
1812 Proc IMechE Part D: J Automobile Engineering 235(7)

2 2
(uiFT ) (ui )
DiFT = uiFT t +  s , ð12Þ
2bFT 2bs
where uiFC and uiFT respectively denote the initial longi-
tudinal velocity of the FC and FT in each sampling
period, bFC and bFT indicate the maximum decelera-
tions of the FC and FT. It is reasonably assumed that
the surrounding vehicles maintain the uniform linear
motion during the remaining lane-changing process.
Thus the longitudinal positions of the above-mentioned
vehicles during lane-changing maneuver can be
expressed as,

xfj = xij + vij t j 2 fPC, FC, PT, FTg, ð13Þ

where vij represents the longitudinal velocity of the vehi-

cle at time t during the lane-changing process, and xij
indicates the longitudinal position of the vehicle at
beginning of each sampling period.
From equation (7), it can be seen that the lane-
changing trajectory is uniquely determined by the final
longitudinal position xfs . What’s more, Yang et al.
pointed out that as long as the final position xfs of the
lane-changing vehicle dropped in a safe position range,
there would be no collisions with preceding and follow-
ing vehicles.25 This paper uses this collision-avoidance
scheme to prevent potential collisions. Assuming that
the lane-changing trajectory is between the centerline of
Figure 5. Diagram of the course angle variation and the
geometric relation of velocity during lane-changing: (a) the the current lane and the centerline of the adjacent lane,
course angle variation and (b) the geometric relation of velocity. w means the width of the lane, and w/2 denotes the lat-
eral coordinate of the lane-changing line. Combined
with equation (7), the longitudinal position xr can be
vn = cos ui us ’us : ð8Þ expressed as,

For convenience, the longitudinal speed of the sub- ys (xr ) = w=2, ð14Þ
ject vehicle can be replaced by velocity us . At the same
time, to avoid potential collisions, Gipps’ safe distance and tr is the instant when the subject vehicle reaches the
rule (Gipps, 1981) is introduced to calculate the safety lane-changing line, which can be calculated as,
distance. According to this principle, DPC and DPT indi- ð xr qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cate the safe distance between the subject vehicle and tr = ( 1 + y0 s 2 ds)=uis + t: ð15Þ
xis
the preceding vehicle in the current lane and the target
lane, respectively. Therefore, for xr, the collision-avoidance zone is,
2 2  i 
(uis ) (ui ) xFC + uiFC tr + DiFC  xis , xiPC + uiPC tr  DiPC  xis  lPC ,
DiPC = uis t +  PC , ð9Þ
2bs 2bPC ð16Þ
2 2
(uis ) (uiPT ) lPC is the vehicle length of PC. To avoid a collision in
DiPT = uis t +  , ð10Þ
2bs 2bPT the current lane, it is necessary to set a safety position
where t indicates the reaction time of vehicle. uis , uiPC , range where the AV can drive into by adjusting velocity
and uiPT , respectively denote the initial longitudinal or trajectory. By using equations (9), (11), and (16), the
velocity of the AV, PC, and PT in each sampling condition of obtaining the collision-avoidance zone can
period. Meanwhile, bs, bPC, and bPT are the maximum be expected as,
decelerations of the corresponding vehicle. 2 2
Similarly, the safe distance between the AV and the uiFC ui
(uiFC + uis )t +  PC \ (uiPC  uiFC )tr  lPC :
following vehicle in the current lane and the target lane 2bFC 2bPC
can be formulated as follows, ð17Þ

(uiFC )
2
(ui )
2 Similarly, the collision-avoidance zone for the final
DiFC = uiFC t +  s , ð11Þ longitudinal position xfs can also be obtained. The zone
2bFC 2bs
[safe_left, safe_right] can be defined as follows,
Liu et al. 1813

safe left = xfFT + DFT  xis , ð18Þ Combined with equations (18), (19), and (24), two
inequality constraints about uis can be obtained as
safe right = xfPT  DPT  lPT  xis : ð19Þ follows,
2
In which, safe_left and safe_right respectively denote uift
the boundary of the collision-avoidance zone of the tar- uis t + xfft + uift t + \ xfpt  xsi  lpt , ð26Þ
2bft
get lane. lPT is the length of the vehicle PT. Moreover, 2
2
equation (13) is used to calculate xfFT and xfPT . tf is the uis 6yfs uis uipt
ffir \ xfft +
+ uis t + pffiffiffiffiffiffiffiffiffiffi  lpt  xsi : ð27Þ
time when the whole lane-changing process is com- 2bs 6yis al 2bpt
pleted, which can be calculated as follows,
ð xf qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tf = ( 1 + y0 s 2 ds)=uis + t: ð20Þ Comfort. To guarantee a comfortable lane-changing
xis process, acceleration and velocity variation are two fac-
tors that should be considered. Yang et al.25 proposed
Thus, when the subject vehicle’s final longitudinal posi-
that the side acceleration at the final position afl repre-
tion xfs enters the collision-avoidance zone, it can effec-
sented the lane-changing comfort for passengers. The
tively prevent collisions with the PT and FT.
formula of afl can be shown as follows,
 
 
Rollover-avoidance. Vehicles maintain the lateral move-  00
ys 
ment during the lane-changing process, yet moving too afl ¼ ðufs Þ2   
3 : ð28Þ
ð1 þ y0 2 Þ2 
fast may lead to a rollover. To avoid the rollover, the s
maximum side acceleration afl at the final position Using equation (7), The final form of afl is,
should not exceed the threshold of the rollover side  
acceleration arl ,30  f f
f f 2 2xs tanui  6ys 
al = (us )   : ð29Þ
 (xf )
2 
afl 4arl : ð21Þ s

At the same time, in Yang et al.’s research, the relation- Meanwhile, to ensure the comfort for passengers,
ship between the final longitudinal position and the the velocity of vehicle should not change too fast in the
maximum side acceleration can be expressed as follows, process of lane-changing. This research hypothesizes
that the subject vehicle maintains uniform motion dur-
6yfs uis ing each sampling period. In order to adapt to a real-
xfs = qffiffiffiffiffiffiffiffiffiffi
ffi: ð22Þ time driving environment, the variation of velocity is
6yfs afl determined by the initial state of two sampling periods.
Thus, the velocity us should satisfy the constraint con-
Therefore, in order to avoid the rollover, xfs should sat- dition and the cost function of comfort is shown as
isfy the following inequality, follows,
6yfs uis jDus =t j4amax , ð30Þ
xfs ø safe roll = qffiffiffiffiffiffiffiffiffiffi
ffi: ð23Þ
f r  i 
6ys al afl 2  u  ui 
s s1 2
min J1 = v1 ( max ) + v2 ( ) , ð31Þ
al tamax
If the above conditions are satisfied, the subject vehi-
cle can avoid rollover. Combined with the previous where afl denotes the side acceleration at the final posi-
collision-avoidance zone, the final longitudinal position tion, amax
l and amax respectively denote the maximum
of the lane-changing trajectory should enter a safety allowable side acceleration and the travel acceleration
zone to avoid both rollovers and collisions. Therefore, in the lane-changing process, uis and uis1 represent the
a safe lane-changing trajectory’s xfs should satisfy the initial velocity of the current sampling period and the
following two constraints simultaneously, previous sampling period, respectively. v1 and v2 are
 the weight values.
safe left \ xfs \ safe right
: ð24Þ
safe roll \ xfs
Traffic efficiency. In a real-world lane-changing scenario,
To simplify the logical calculation, we order the subject vehicle tends to finish a lane-changing pro-
low_limit=Max(safe_left, safe_roll), up_limit=safe_- cess as soon as possible. Some studies adopt the time
right. Thus, the safe zone for the final longitudinal ratio tfs =tmax to represent this characteristic,25,31 where
position of lane-changing trajectory is [low_limit, up_li- tfs denotes the time used to finish the lane-changing pro-
mit]. The inequality (25) is a necessary and sufficient cess and tmax is the longest allowable time for lane-
condition for the existence of this safe zone, changing maneuver. Traffic efficiency description in
this paper is developed by the above method and the
low limit \ up limit: ð25Þ cost function can be calculated as follows,
1814 Proc IMechE Part D: J Automobile Engineering 235(7)

2 2
tfs xfs  xis optimization problem with the interior-point algorithm,
min J2 = v3 = v3 , ð32Þ the optimal solution xfop and uiop would help to find the
tmax uis tmax
optimal lane-changing trajectory. Real-time informa-
where xis represents the initial longitudinal coordinate tion about the driving environment is obtained through
of the current sampling period and xfs is the final longi- V2V system, which is defined as:
tudinal position. It is reasonably assumed that the AV
keeps a uniform longitudinal motion during the remain- environment(n) =
ing lane-changing process. While the time tfs can be ð36Þ
xik , yik , uik k 2 fpt, pc, ft, fcg, n4N,
approximately obtained by (xfs  xis )=uis . Moreover, v3
is the weight coefficient for traffic efficiency. where xik , yik , and uik , respectively represent the vehicle’s
initial coordinate information and velocity in each sam-
pling period. N is the maximal sampling number, and
The DLTP algorithm
the algorithm runs once in each sampling step n.
By consideration of all performance specifications, this To sum up, Algorithm 1 explains a DLTP algorithm,
paper proposes a cost function to find the optimal lane- and its key thought includes: at every sampling time, if
changing trajectory of CAVs. The function shows as the current speed and the planned trajectory meet the
follows, requirements of driving, the AV would continue the
"   #2 lane-changing without additional operation; when the
 f f
i 2 2xs tanui  6ys  r current vehicle status does not satisfy the condition of
J = v1 (us )  =al
 2  lane-changing, the vehicle needs to adjust the velocity
(xfs ) ð33Þ
 i  or trajectory to adapt to the driving environment; if the
 u  ui  x f
 x i
+ v2 ( s maxs1 )2 + v3 ( si s 2
) , adjustment of velocity or trajectory cannot satisfy the
ta us tmax requirement of safe lane-changing operation, it indi-
cates that the current traffic environment is not suitable
where J denotes the cost function, v1 , v2 , and v3 can
to continue change lanes, and the AV should cease the
be obtained by the mature driver model and actual lane
lane-changing maneuver immediately. In such a situa-
changing data.32 The initial course angle ui is given and
tion, the original lane is designated as the new target
the initial course angle in each subsequent sampling
lane, and the subject vehicle can still use the above
period uip can be calculated as follows,
DLTP algorithm to complete this operation. The lane-
uip = arctan changing safety monitoring algorithm in Algorithm 1
! will be introduced in ‘‘The lane-changing safety moni-
3yfn  2xfn tanui xfn tanui  2yfn 2 toring algorithm’’ section Algorithm 2.
tanui + 2
xip + 3
(xip )
(xfn ) (xfn )
n ø 2: ð34Þ The lane-changing safety monitoring algorithm
In which, (xfn , yfn )
is the final position of planning trajec- The core idea of the lane-changing safety monitoring
tory at each sampling step, and the new coordinate xip (LSM) algorithm is to detect whether the automated
of the target vehicle can be obtained by V2V technology
in each time step.
The cost function is minimized to obtain the optimal Algorithm1: The DLTP algorithm
lane-changing trajectory. In addition, some certain con-
Input: Real-time driving 9 else
straints in safe and real lane-changing maneuver cannot environment information 10 solving optimization
be ignored. Hence, the trajectory planning problem (n); the subject vehicle’s problem (35)
could be converted into the constrained optimization initial velocity uin and the 11 if( Algorithm 2out ==1)
problem which is shown in (35). final longitudinal 12 xnf + 1 = xnf
coordinate xnf . N is the 13 uin + 1 = uin
14 using Formula (7)
min J(xfop , uiop ) maximum sampling
number. to calculate ysn
s:t: Output: The trajectory 15 n = n + 1
8 ysn 16 else
> ð35Þ
< Eq:(30) 1n = 1; 17 interrupt lane-changing and
Eq:(17) 2 while (n4N): turn back to the original lane
>
: 3 perceived 18 end if
Eq:(26, 27) environment{n} 19 end if
4 if ( Algorithm 2 out 20 end while
It should be noted that the constraint condition (17) is ==1)
only applicable before the subject vehicle enters the tar- 5 xnf + 1 = xnf
get lane. The constraints mentioned above ensure the 6 uin + 1 = uin
safety of the subject vehicle in the whole lane-changing 7 using Formula (7) to
process and the cost function combines comfort and calculate ysn
8 n=n+1
efficiency simultaneously. By solving the constrained
Liu et al. 1815

Algorithm 2: The LSM algorithm Algorithm3:The LSD algorithm

Input: real-time driving environment information environment Input: initial driving f

(35) minJ(xop, uop )
(n); the subject vehicle’s initial velocity uin and the final environment information 1
13 if (us = uop )
longitudinal coordinate xnf . environment (1); sampling 14 xstart = xs1 
Output: monitor result out. period T; the initial 15 else if (u1s  uop  \ tamax )
1. Using Formula (9)-(12), (18)-(24) to calculate relevant velocity and longitudinal u1 + u
16 xstart = xs1 + t( s 2 op )
collision-avoidance and rollover-avoidance parameters. coordinate u1s
17 else
2. Calculating the safe position range: [low_limit, right_limit]. xs1 . u 2 u1
2
18 xstart = xs1 + ( op2amaxs )
3. Determine whether Conditions (37)-(38) are valid at the Output: xs1 , xsf , u1s .
19 end if
same time. 1 Calculating parameters
20 end if
If true, out=1; else out=0. S1pc , S1fc
21 xs1 = xstart
4. Return out 2 While (Condition (37) f
22 xsf = xop
is not satisfied): 1
23 us = uop
3 xs1 = xs1 + u1s  T
Return xs1 , xsf , u1s
vehicle’s current speed and the final planned longitudi- 4 update S1pc , S1fc
nal position satisfy the condition of safe driving. 5 end while
6 Calculating parameters
Similar to the above collision-avoidance constraints, low_limit, up_limit
considering that the real-time information of vehicle’s 7 while (Condition (25)
position and velocity can be obtained by V2V system, is not satisfied):
the collision-avoidance conditions between the AV and 8 xs1 = xs1 + u1s  T
the surrounding vehicles are as follows, 9 update low_limit,
up_limit
( 10 end while
xipc  xis ø Dipc 11 if
, ð37Þ
xis  xifc ø Difc (low_limit \ up_limit)
12 Solving optimization
where xipc and xifc are the longitudinal positions at the problem
ith sample time of the vehicle PC and FC, respectively.
At the same time, it is also necessary to consider that
the final longitudinal position of the lane-changing tra- trajectory without additional action. If the information
jectory should fall into the safe position range which of the starting-point is not satisfied with any of the
can avoid both rollovers and collisions. Thus, the safe above conditions, the vehicle needs to adjust the state
constraint of xfs is as follows, of the starting-point. Therefore, this paper proposes a
lane-changing starting-point determination (LSD) algo-
low limit4xfs 4up limit: ð38Þ
rithm to identify whether the current state information
According to the above two safety problems during meets the requirements of a lane-changing maneuver.
the lane-changing progress, this paper proposes an The LSD algorithm is shown in Algorithm 3 and the
LSM algorithm that can real-time monitor whether the update strategies are as follows,
status of CAVs meet the requirement of safe lane-
changing. If the requirement is satisfied, the vehicle has (1) If condition (37) and (25) are not satisfied, the
no need to re-plan the lane-changing trajectory, but vehicle maintains the uniform linear motion until
only keeps the current speed and the initial lane- the next sampling time T, and the position of the
changing trajectory. Otherwise, when the safety condi- starting-point can be updated by the following
tions are not satisfied, it proves that the velocity or the equation,
lane-changing trajectory is not suitable for continuous
changing lanes, and the trajectory needs to be re-
planned. The LSM algorithm is shown in Algorithm 2. x1s = x1s + u1s  T, ð39Þ

where x1s and u1s denote the initial longitudinal position

The lane-changing starting-point determination and the velocity of CAVs, respectively. In addition, the
algorithm algorithm updates x1s and the safe position range, [low_-
Before a lane-changing maneuver begins, the subject limit, right_limit].
vehicle first needs to determine whether the current
location information meets the safety requirements pre- (2) When condition (25) is satisfied, uop and xfop can
sented in the LSM algorithm. Then, the vehicle needs be solved by optimization problem (35). If
to judge whether the current velocity can complete the (u1s = uop ), the AV can change lane in the current
lane-changing maneuver. If the current conditions of state.
the starting-point satisfy the requirements, the subject (3) If u1s can be adjusted to the optimal velocity uop ,
vehicle can get the optimal final longitudinal position the new initial longitudinal position xstart can be
xfs and use Algorithm 1 to plan lane-changing calculated by the following formula,
1816 Proc IMechE Part D: J Automobile Engineering 235(7)

Table 1. Default parameters of the DLTP mode.

Parameter Unit Value

Reaction time of vehicle t s 0.5

Lane width w m 4
Maximum decelerations of vehicle b m/s2 3.5
Maximum allowable side acceleration amaxl m/s2 1.4
Maximum allowable travel acceleration amax m/s2 3
Longest allowable time for lane-changing maneuver tmax s 10
Weight coefficients of the cost function [v1 , v2 , v3 ] [0.5,0.3,0.2]

Figure 6. The safe distance of the surrounding vehicle: (a) sampling time is 0.5s, (b) sampling time is 1s, (c) sampling time is 1.5s
and (d) corresponding velocities variation.

u1s + uop evaluate whether the proposed DLTP model can be

xstart = x1s + t( ): ð40Þ
2 applied in real-world traffic. Meanwhile, the effective-
ness of the LSD Algorithm and the influence of the
(4) If u1s cannot be adjusted to uop during the reaction
sampling time on the final trajectory are analyzed.
time t, xstart would be updated as follows,
Furthermore, MPC algorithm and CarSim-Simulink
2 2
co-simulation platform are performed to evaluate the
uop  u1s
xstart = x1s +( ): ð41Þ traceability of the proposed dynamic trajectory.
2amax

Simulations with real-world lane-changing data

Simulations and analyses
Data introduction. To verify the effectiveness of the pro-
The typical lane-changing scenarios, obtained from the posed model, the traffic data provided by the Federal
real-world driving dataset NGSIM, are used to Highway Administration’s (FHWA) Next Generation
Liu et al. 1817

Figure 7. The comparison between DLTP trajectory and NGSIM trajectory of the vehicle 605: (a) trajectory and (b) velocity.

Figure 8. The safe distance of the surrounding vehicles of the Figure 9. The comparison between DLTP trajectory and
vehicle 605: (a) target lane and (b) current lane. NGSIM trajectory of the vehicle 1042.

Simulation (NGSIM) project (https://data.transporta-

tion.gov) are chosen to extract real-world lane-changing coordinate, velocity acceleration, vehicle gap, etc.32 As a
data. Ten frames of traffic data are obtained per second, high-resolution real-world vehicle trajectory data,
including vehicle ID, lateral coordinate, longitudinal NGSIM is widely used to explore the various CAVs
1818 Proc IMechE Part D: J Automobile Engineering 235(7)

Figure 10. The safe distance of the surrounding vehicles of the Figure 11. The comparison between DLTP trajectory and
vehicle 1042. NGSIM trajectory of the vehicle 205.

scenarios in traffic environment, and many researches trajectories. Figure 6 shows the planning trajectories at
utilize NGSIM to verify the effectiveness of the mod- different sampling times and Figure 6(d) exhibits the
els.33–35 corresponding velocities variation.
In this study, NGSIM data provides a real-world Figure 6 shows that as sampling time goes on, the
lane-changing environment for the automated vehicle. number of re-planned trajectory decreases and the
In simulations, the velocities and the positions of sur- planned trajectory becomes smoother. Similarly, the fluc-
rounding vehicles (PC, FC, PT, and FT) are given by tuation of velocity also becomes more gentle. However,
NGSIM data, while the subject vehicle’s lane-changing it must be noted that an excessive sampling time will
trajectory and speed fluctuation are produced by the miss some unexpected situations, which may result in the
proposed DLTP method. In addition, Yang et al.35 automated vehicle’s failure to respond timely, thus leads
have shown that the interval 5 s before and after the to traffic accidents. Therefore, it is important to select a
time of lane-changing is an appropriate period to ensure suitable sampling time. In most cases, the sampling time
the entire lane-changing process is covered. Thus, this of this paper is determined as 1 s.
paper extracts the vehicle position information in
T 2 ½tr  5s, tr + 5s where tr is the instant when the
vehicle AV reaches the lane-changing line. If not speci- Planned lane-changing trajectory analysis. Figure 7 shows
fied, the default values for parameters in the simulation the results of comparing the proposed DLTP model
model are shown in Table 1, which can be determined with the real acceleration lane-changing trajectory (ID
according to references and the user preferences.25,30,36 of the vehicle = 605 in NGSIM data). As shown in
Figure 7(a), in the early stage of lane-changing, the AV
Sampling time analysis. There is no doubt that the sam- keeps going in a straight line, which proves that the ini-
pling time will have an impact on the planned tial speed or position cannot satisfy the requirements of
Liu et al. 1819

Table 2. Main parameters of the experimental vehicle.

Parameter Value Parameter Value

Front wheel slip ratio 0.2 Rear wheel slip ratio 0.2
Distance between centroid and front axle 1.23 m Distance between centroid and rear axle 1.47 m
Front wheel lateral cornering stiffness 66,900 N/rad Rear wheel lateral cornering stiffness 62,700 N/rad
Front wheel longitudinal cornering stiffness 66,900 N/rad Rear wheel longitudinal cornering stiffness 62,700 N/rad
Vehicle mass 1723 kg Moment of Inertia around z-axis 4175 kgm2

vehicles. Longitudinal dashed lines in Figure 8 indicate

the time when the subject vehicle reaches the lane-
changing line, and the trajectories of surrounding vehi-
cles are extracted from the real-world lane-changing
environment. In Figure 8(a), it can be observed that
before entering the target lane, the AV keeps a safe dis-
tance from the proceeding and following cars of the
current lane. As for the target lane in Figure 8(b), the
AV can also avoid collisions with adjacent vehicles.
The same conclusion can be drawn from Figures 9 and
10 (a deceleration lane-changing scenario of ID = 1042
in NGSIM data).
Figure 11 exhibits an interruption lane-changing sce-
nario of the vehicle 205. The planned trajectory pro-
duced by the proposed DLTP model does not complete
the lane-changing maneuver but turns back to the orig-
inal lane, while NGSIM trajectory finishes the lane-
changing process with a more aggressive operation.
This difference indicates that the DLTP method pays
more attention to security and has the capability to
deal with the scenario where a lane-changing maneuver
is interrupted. Figure 12 also shows that the AV can
maintain a safe distance from surrounding vehicles dur-
ing the whole interruption process, which reflects the
good safety of the system.
Meanwhile, in the above three lane-changing scenar-
ios, the average computation time of the planning algo-
rithm is 0.07 s, which satisfies the dynamic performance
of the model. Therefore, this paper draws a conclusion
that the subject vehicle adopts the proposed DLTP
model to complete the lane-changing maneuver safely
and smoothly, and the planned trajectory is more com-
fortable than NGSIM lane-changing trajectory.

Figure 12. The safe distance of the surrounding vehicles of the

vehicle 205. Simulations with CarSim-Simulink platform
MPC algorithm. MPC is a feedback control algorithm,
which uses a model to predict future output of pro-
the DLTP model. At this time, the LSD algorithm
cesses and it has the ability to deal with multi-objective
adaptively adjusts the driving state of the vehicle to
constraints.37,38 Accordingly, MPC is widely used in
meet the safety requirement. What’s more, similar to
trajectory tracking and obstacle avoidance of intelligent
NGSIM trajectory, the planned trajectory can reach
vehicles.39–41
the centerline of the target lane smoothly, but the var-
Compared with the kinematical vehicle model used
iance of velocity is more moderate than the real sce-
in the previous DLTP model,13,24 this paper applies the
nario in Figure 7(b). It indicates that the planned
vehicle dynamic model to satisfy the requirements of
trajectory is more comfortable. Furthermore, consider-
safe driving in the complex traffic environment, which
ing the safety of vehicle, Figure 8 exhibits the distance
improves the reliability of automated vehicles at high
between the planned trajectory and the surrounding
speed. With limitation in time and space, this paper
1820 Proc IMechE Part D: J Automobile Engineering 235(7)

Figure 13. Simulation results for the acceleration scenario: (a) path, (b) longitudinal velocity, (c) lateral acceleration, (d) roll angle,
(e) front wheel steer angle and (f) slip angle.

does not introduce the dynamic vehicle model in detail. path tracking, the objective function based on MPC
A more detailed description of the model can refer to algorithm can be designed as,
Rajamani.42 The equation of the dynamic vehicle model
can be compactly written as, J(j(t), u(t  1), DU(t))
Np
X
_ = f(j(t), u(t)),
j(t) ð42Þ 2
= Yðt + ijtÞ  Yref (t + ijt) Q
i=1 ð43Þ
where j(t) = ½y,
_ x, _ Y, XT is the vehicle state vec-
_ f, f,
tor, which respectively represent the vehicle velocity in X
N c 1

+ kDu(t + ijt)k2R + re2 ,

y and x direction, the course angle, the course angle I=1
velocity, and the position coordinate of vehicle. d indi-
cates the vehicle’s front steering angle, and u = ½d is where Np and Nc respectively indicate the prediction
the control input vector. and the control horizon. Q and R are weight matrices.
In addition, in order to achieve the optimal control The first part of the equation represents the ability of
effect, the optimal solution obtained by the designing path following, and the second part means the smooth-
objective function and the dynamic constraints can be ness of the control vector. The third part indicates the
used as the control input of the controller. For vehicle relaxation factor. Combined with some constraints in
Liu et al. 1821

Figure 14. Simulation results for the deceleration scenario: (a) path, (b) longitudinal velocity, (c) lateral acceleration, (d) roll angle,
(e) front wheel steer angle and (f) slip angle.

vehicle driving, the final optimization problem can be boundaries of the soft constraint. e means the relaxa-
expressed as, tion factor.
The solution of the above optimization problem is
min J(j(t), u(t  1), DU(t)) completed within each control cycle, and the control
DU, e
incremental sequence in the control time domain can
s:t: DUmin 4DUt 4DUmax
be obtained as follows,
Umin 4ADUt + Ut 4Umax ð44Þ
yh, min 4yh 4yh, max DUt = ½Dut , Dut + 1 ,    Dut + Nc 1 T : ð45Þ
ys, min  e4ys 4ys, max + e
The first element in the control sequence affects the sys-
e . 0, tem as an increment of the actual control input. Thus,
the current control vector is expressed as,
where Umin and Umax are the boundary conditions of
the control vector. Similarly, DUmin and DUmax indicate u(t) = u(t  1) + Dut : ð46Þ
the constraints of the control increment. yh and ys
respectively represent the hard constraint and the soft After entering the next control cycle, the system would
constraint of output. yh,max and yh,min are the con- repeat the above process to achieve the desired trajec-
straints of hard constraint, ys,max and ys,min are the tory tracking control.
1822 Proc IMechE Part D: J Automobile Engineering 235(7)

Figure 15. Simulation results for the interruption scenario: (a) path, (b) longitudinal velocity, (c) lateral acceleration, (d) roll angle,
(e) front wheel steer angle and (f) slip angle.

Simulation based on CarSim-Simulink platform. To validate all originated from the above NGSIM lane-changing
the proposed DLTP model based on real-world traffic scenarios. In addition, to better reflect the tracking
data, this paper establishes a simulation platform accuracy, this paper adopts Root Mean Squared Error
through MATLAB/Simulink and CarSim. CarSim deli- (RMSE) as the evaluation index.
vers accurate, detailed and efficient methods for simu-
lating the performance of vehicles. The main Acceleration lane-changing scenario. Figure 13 shows the
parameters of the experimental vehicle are shown in simulation results of the acceleration lane-changing sce-
Table 2. Other parameters involved are chosen based nario. In this scenario, the subject vehicle completes a
on their default setting in CarSim. The trajectory, long- lane-changing maneuver by acceleration. Figure 13(a)
itudinal speed, lateral acceleration, roll angle, front and (b) respectively exhibit the traceability of path and
wheel steer angle and slip angle of the vehicle are cho- longitudinal velocity, where the blue line represents the
sen as Carsim’s output during the dynamic automated simulation driving information (including longitudinal
lane-changing maneuver. The driving environment speed and trajectory) of the target vehicle, and the red
information of three scenarios, the acceleration lane- line indicates the target path and velocity respectively.
changing scenario, the deceleration lane-changing sce- RMSE values of the lane-changing trajectory following
nario and the interruption lane-changing scenario are and speed tracking are 0.05 and 1.49, respectively,
Liu et al. 1823

which indicates that the planned lane-changing velocity planning strategy and considers more complete driving
and trajectory can be well tracked by the automated environment information. Besides, this model allows
vehicles. Figure 13(c) and (d) respectively illustrate the CAVs to avoid potential collisions resulting from varia-
lateral acceleration variation and the roll angle varia- tions in the states of surrounding vehicles during the
tion in Carsim. The small variation in two angles means lane-changing maneuver. The efficiency and comfort of
that the automated vehicles have good safety and com- lane-changing are also fully considered. Finally, the
fort performance during the lane-changing process. The effectiveness of the proposed model is verified on
vehicle’s front wheel steer angle and slip angle shown in CarSim-Simulink platform by adopting real-world
Figure 13(e) and (f) are both small, which means the lane-changing data and MPC algorithm. Future works
automated vehicles have good driving stability during will concentrate on the performance of V2V communi-
the lane-changing maneuver. cation and establishing an experimental platform for
real vehicles. Strengthening the robustness of the
Deceleration lane-changing scenario. Figure 14 shows method also becomes our next focus.
the simulation results of the deceleration lane-changing
scenario. The subject vehicle completes a lane-changing Declaration of conflicting interests
maneuver mainly by a deceleration in this scenario.
Figure 14(a) and (b) illustrate that the automated vehi- The author(s) declared no potential conflicts of interest
cles can also track the planned speed and trajectory well with respect to the research, authorship, and/or publi-
in deceleration scenario (RMSE for the lane-changing cation of this article.
trajectory tracking is less than 0.1, and the value for the
velocity tracking is less than 1.35, both of which are Funding
small). Figure 14(c) to (f) indicate that the driving The author(s) disclosed receipt of the following finan-
safety, comfort and stability of the automated vehicles cial support for the research, authorship, and/or publi-
perform well in this lane-changing process. cation of this article: This work was supported by
National Natural Science Foundation of China under
Interruption lane-changing scenario. If the surrounding Grant U1664264 and the National Key R&D Program
traffic environment is not suitable for continuing of China under Grant 2019YFB1600500.
changing lane, the lane-changing maneuver should be
interrupted, and vehicle should go back to the original
lane. The simulation results of the interruption lane- ORCID iD
changing scenario are exhibited in Figure 15. Figure Xiao Liu https://orcid.org/0000-0001-8254-3014
15(a) shows that the planned trajectory does not reach
the target lane but turns back to the original lane, References
which shows the proposed DLTP model has the ability
1. Bevly D, Cao X, Gordon M, et al. Lane change and
to handle the interruption lane-changing scenario.
merge maneuvers for connected and automated vehicles:
Besides, the great traceability of the control algorithm
a survey. IEEE Trans Intell Veh 2016; 1: 105–120.
(the path and speed tracking’s RMSE are 0.07 and 2. Ji X, Liu Y, He X, et al. Interactive control paradigm-
1.43, respectively), the positive driving safety, comfort based robust lateral stability controller design for auton-
and stability of the automated vehicles are also exhib- omous automobile path tracking with uncertain distur-
ited in Figure 15. bance: a dynamic game approach. IEEE Trans Veh
In addition, a smaller wheel steer angle that reflected Technol 2018; 67: 6906–6920.
in the above three scenarios can be easily adjusted to 3. Bonnefon J-F, Shariff A and Rahwan I. The social
zero at the end of the lane-changing maneuver, which dilemma of autonomous vehicles. Science 2016; 352:
enables the subject vehicle to make its body parallel 1573–1576.
with the lane line and enter the lane-keeping state more 4. Ramezani M and Ye E. Lane density optimisation of
automated vehicles for highway congestion control.
quickly. This conclusion is also consistent with the
Transportmetrica B-Transp Dyn 2019; 7: 1096–1116.
existing studies.21,25,43
5. Levin MW and Boyles SD. A multiclass cell transmission
model for shared human and autonomous vehicle roads.
Transp Res Part C: Emerg Technol 2016; 62: 103–116.
Conclusion 6. Petit J and Shladover SE. Potential cyberattacks on auto-
In this paper, a dynamic lane-changing trajectory plan- mated vehicles. IEEE Trans Intell Transp Syst 2015; 16:
546–556.
ning method is designed for CAVs in real-world traffic
7. Leng Z and Minor MA. Curvature-based ground vehicle
environments. Firstly, a time-independent polynomial control of trailer path following considering sideslip and
equation is adopted to characterize the lane-changing limited steering actuation. IEEE Trans Intell Transp Syst
trajectory curve. Secondly, the DLTP, LSM and LSD 2017; 18: 332–348.
algorithms are proposed to constitute the dynamic 8. Malikopoulos AA, Hong S, Park BB, et al. Optimal con-
lane-changing trajectory model. The key advantage of trol for speed harmonization of automated vehicles.
this DLTP model is that it proposes a detailed velocity IEEE Trans Intell Transp Syst 2019; 20: 2405–2417.
1824 Proc IMechE Part D: J Automobile Engineering 235(7)

9. Hou Y, Edara P and Sun C. Situation assessment and 25. Yang D, Zheng S, Wen C, et al. A dynamic lane-changing
decision making for lane change assistance using ensem- trajectory planning model for automated vehicles. Transp
ble learning methods. Expert Syst Appl 2015; 42: 3875– Res Part C: Emerg Technol 2018; 95: 228–247.
3882. 26. Jiang H, Shi K, Cai J, et al. Trajectory planning and opti-
10. Zhao D, Lam H, Peng H, et al. Accelerated evaluation of misation method for intelligent vehicle lane changing
automated vehicles safety in lane-change scenarios based emergently. IET Intell Transp Syst 2018; 12(10): 1336–
on importance sampling techniques. IEEE Trans Intell 1344.
Transp Syst 2017; 18: 595–607. 27. Ding Y, Zhuang W, Wang L, et al. Safe and optimal
11. Kim J, Kim J, Jang G-J, et al. Fast learning method for lane-change path planning for automated driving. Proc
convolutional neural networks using extreme learning IMechE, Part D: J Automobile Engineering 2020:
machine and its application to lane detection. Neural Net- 0954407020913735.
works 2017; 87: 109–121. 28. Wang Y, Wei C, Liu E, et al. Lane-changing trajectory
12. Yang K, Guler SI and Menendez M. Isolated intersection planning method for automated vehicles under various
control for various levels of vehicle technology: conven- road line-types. IET Intell Transp Syst 2020; 2(1): 14–23.
tional, connected, and automated vehicles. Transp Res 29. Zhou B, Wang Y, Yu G, et al. A lane-change trajectory
Part C: Emerg Technol 2016; 72: 109–129. model from drivers’ vision view. Transp Res Part C:
13. Zheng H, Zhou J, Shao Q, et al. Investigation of a longi- Emerg 2017; 85: 609–627.
tudinal and lateral lane-changing motion planning model 30. Gunawan FE. Two-vehicle dynamics of the car-following
for intelligent vehicles in dynamical driving environments. models on realistic driving condition. J Transp Syst Eng
IEEE Access 2019; 7: 44783–44802. Inf Technol 2012; 12: 67–75.
14. Hobert L, Festag A, Llatser I, et al. Enhancements of v2x 31. Chen L, Qin D, Xu X, et al. A path and velocity planning
communication in support of cooperative autonomous method for lane changing collision avoidance of intelli-
driving. IEEE Commun Mag 2015; 53: 64–70. gent vehicle based on cubic 3-D Bezier curve. Adv Eng
15. An H and Jung J-I. Design of a cooperative lane change Softw 2019; 132: 65–73.
protocol for a connected and automated vehicle based on 32. Coifman B and Li L. A critical evaluation of the Next
an estimation of the communication delay. Sensors 2018; Generation Simulation (NGSIM) vehicle trajectory data-
18(10): 3499. set. Transp Res Part B-Methodol 2017; 105: 362–377.
16. Choi J, Va V, Gonzalez-Prelcic N, et al. Millimeter-wave 33. Wang X, Jiang R, Li L, et al. Capturing car-following
vehicular communication to support massive automotive behaviors by deep learning. IEEE Trans Intell Transp
sensing. IEEE Commun Mag 2016; 54: 160–167. Syst 2018; 19: 910–920.
17. Li Z, Elefteriadou L and Ranka S. Signal control optimi- 34. Schubert R, Schulze K and Wanielik G. Situation assess-
zation for automated vehicles at isolated signalized inter- ment for automatic lane-change maneuvers. IEEE Trans
sections. Transp Res Part C: Emerg Technol 2014; 49: 1– Intell Transp Syst 2010; 11: 607–616.
18. 35. Yang D, Zhu L, Ran B, et al. Modeling and analysis of
18. Liang X, Guler SI and Gayah VV. Joint optimization of the lane-changing execution in longitudinal direction.
signal phasing and timing and vehicle speed guidance in a IEEE Trans Intell Transp Syst 2016; 17(10): 2984–2992.
connected and autonomous vehicle environment. Transp 36. Ross M, Patel D and Wenzel T. Vehicle design and the
Res Rec 2019; 2673: 70–83. physics of traffic safety. Phys Today 2006; 59(1): 49.
19. Tang J, Liu F, Zhang W, et al. Lane-changes prediction 37. Garcı́a CE, Prett DM and Morari M. Model predictive
based on adaptive fuzzy neural network. Expert Syst Appl control: theory and practice—a survey. Automatica 1989;
2018; 91: 452–463. 25: 335–348.
20. Cao H, Song X, Zhao S, et al. An optimal model-based 38. Mayne DQ, Rawlings JB, Rao CV, et al. Constrained
trajectory following architecture synthesising the lateral model predictive control: stability and optimality. Auto-
adaptive preview strategy and longitudinal velocity plan- matica 2000; 36: 789–814.
ning for highly automated vehicle. Veh Syst Dyn 2017; 39. Raffo GV, Gomes GK, Normey-Rico JE, et al. A predic-
55: 1143–1188. tive controller for autonomous vehicle path tracking.
21. Chu K, Lee M and Sunwoo M. Local path planning for IEEE Trans Intell Transp Syst 2009; 10: 92–102.
off-oad autonomous driving with avoidance of static 40. Ji J, Khajepour A, Melek WW, et al. Path planning and
obstacles. IEEE Trans Intell Transp Syst 2012; 13: 1599– tracking for vehicle collision avoidance based on model
1616. predictive control with multiconstraints. IEEE Trans Veh
22. Chen C, He Y, Bu C, et al. Quartic bezier curve based tra- Technol 2017; 66: 952–964.
jectory generation for autonomous vehicles with curva- 41. Rasekhipour Y, Khajepour A, Chen S-K, et al. A poten-
ture and velocity constraints. In: 2014 IEEE international tial field-based model predictive path-planning controller
conference on robotics and automation. 2014, pp.6108– for autonomous road vehicles. IEEE Trans Intell Transp
6113. IEEE. Syst 2017; 18: 1255–1267.
23. Zhang H and Yang S. Smooth path and velocity plan- 42. Rajamani R. Vehicle dynamics and control. Springer Sci-
ning under 3D path constraints for car-like vehicles. Rob ence & Business Media, 2006.
Auton Syst 2018; 107: 87–99. 43. Zhang S, Deng W, Zhao Q, et al. Dynamic trajectory
24. Luo Y, Xiang Y, Cao K, et al. A dynamic automated lane planning for vehicle autonomous driving. In: IEEE inter-
change maneuver based on vehicle-to-vehicle communica- national conference on systems, man, and cybernetics,
tion. Transp Res Part C: Emerg 2016; 62: 87–102. 2013, pp.4161–4166. IEEE.