You are on page 1of 9

Advances in Engineering Software xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Advances in Engineering Software


journal homepage: www.elsevier.com/locate/advengsoft

A path and velocity planning method for lane changing collision avoidance
of intelligent vehicle based on cubic 3-D Bezier curve
Chen Longa,b, Qin Dongfanga, Xu Xinga,b, , Cai Yingfenga,b, Xie Jua

a
School of Automotive and Traffic Engineering, Jiangsu University, Zhenjiang 212013, China
b
Automotive Engineering Research Institute, Jiangsu University, Zhenjiang 212013, China

ARTICLE INFO ABSTRACT

Keywords: A novel path and velocity planning method for lane changing collision avoidance of intelligent vehicle based on
Intelligent vehicle cubic 3-D Bezier curve is proposed in this paper to enhance the flexibility of path and guarantee the conformity
Bezier curve to the realistic lane changing maneuver. Owing to the characteristic of 3-D Bezier curve, the velocity information
Lane changing is appended to the path to establish a real-time correspondence between them. In the XY plane, the integrated
Path planning
path is divided into a collision avoidance path and a lane changing path, which is combined with two asym-
Velocity planning
metrical Bezier curves. And the velocity planning of two paths is implemented by the same curve in the Z axis.
The objective functions and dynamic constraints are formulated to solve the optimal problems of the path and
velocity planning for satisfying different traffic situations. The simulations are performed in the CarSim-Simulink
co-simulation platform and the results show the performance and effectiveness of the proposed approach.

1. Introduction terminal longitudinal position can be determined. Funke et al. [25]


introduced the clothoid including the relationships between straights
Research on exploring intelligent vehicle [1–3] and electric vehicle and arcs for path profile generation and combined two-elementary path
[4] to improve traffic safety and efficiency is one of the most ex- segments into a bi-elementary path. Chen et al. [26] presented a path
tensively studied topics in the future automotive industry [5–7]. Take planning method by applying the segmented quadratic Bezier curve to
the fact that vehicle lane changing is regarded as one kind of common calculate the lane changing path to meet the maximum curvature
and dangerous driving maneuver into account [8–10], much un- constraint, however the disadvantage of this method is the planning
precedented attention has been concentrated on this microscopic traffic path cannot satisfy more dynamic constraints for the reason that the
behavior. The entire lane changing process could be divided into three number of control points is finite. Il Bae et al. [27] studied a lane
main subparts: lane changing decision [11,12], path planning [13–15], changing path on the basis of the quintic Bezier curve, and the path
and path following [16–18]. Among the three subparts, path planning is could guarantee the requirements of curvature and steering wheel angle
tightly integrated with traffic environment and has become a vital both at the initial and terminal positions which should be zero. In
component in the field of Intelligent Transportation System (ITS). The contrast, the selection of control points is not discussed in detail.
studies of path planning could contribute to make the intelligent vehicle Through investigating and summarizing the previous research
reach the desired destination smoothly and securely in real world traffic achievements, one can find that researchers generally consider that
environment without collisions [19–20]. most lane changing paths based on geometric curve function are sym-
In the existing research, most scholars focus on the context of the metric about the center of lane line, so the shape of the partial curve is
path planning method based on geometric curve function for lane not convenient to change. To comply with the real lane changing
changing maneuver, for instance, slope sine function, polynomial maneuver, the path should be variable in both local and global parts
function, B-spline function and Bezier curve function [21–27]. The lane according to the diverse driver demands and actual driving environ-
changing path generated by above methods possesses the simple and ment.
continuous shape, and accordingly is not complicated to calculate. Yang In addition, most path planning approaches assume that the vehicle
et al. [24] adopted cubic polynomial curve to plan the lane changing drives along the path with a constant velocity, thus little work has been
path and designed the cost function. By settling the cost function, the done to plan the vehicle velocity in view of conforming to the actual


Corresponding author.
E-mail address: xuxing@mail.ujs.edu.cn (X. Xu).

https://doi.org/10.1016/j.advengsoft.2019.03.007
Received 21 November 2018; Received in revised form 12 February 2019; Accepted 12 March 2019
0965-9978/ © 2019 Elsevier Ltd. All rights reserved.

Please cite this article as: Chen Long, et al., Advances in Engineering Software, https://doi.org/10.1016/j.advengsoft.2019.03.007
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

driving situation [28–31]. Chen et al. [28] presented the path planning cubic 3-D Bezier curves, and their expressions can be written as
method based on the quartic Bezier curve, and the velocity curve is
carried out via using linear velocity profile generation method ac- P1 (t ) = P10 (1 t ) + 3P11 (1 t ) 2t + 3P12 (1 t ) t 2 + P13 t 3
cording to the side-slip constraint, which ensures the safety of the path P2 (t ) = P20 (1 t ) + 3P21 (1 t )2t + 3P22 (1 t ) t 2 + P23 t 3 (3)
following. Cao et al. [29] took advantage of the quadruplicate function
where P1(t) represents the first path, which is named as collision
to design the velocity profile in terms of time variable. Then the velo-
avoidance path; P2(t) represents the second path, which is named as
city function is figured out by designing the objective function and
lane changing path. The control points can be expressed as
related kinematic constraints, but the objective function just containing
the relationship between the velocity and longitudinal jerk is not P1i = [ x1i y1i z1i ]T , P2i = [ x2i y2i z2i ]T (4)
comprehensive enough.
To remedy the deficiency of lane changing path and velocity plan- where i = (0,1,2,3).
ning in the aforementioned studies, a novel path and velocity planning The X and Y axis coordinates of control points are used for planning
method of lane changing collision avoidance based on cubic 3-D Bezier the lane changing collision avoidance path, and the Z axis coordinates
curve for intelligent vehicle is proposed in this paper. The reason of are applied to design the velocity curve of intelligent vehicle. Hence,
using this method is that this planning method could make the path and the 3-D Bezier curve simultaneously includes the information of path
velocity curves more flexible and easily adapt to realistic lane changing and velocity, and the corresponding velocity at any position could be
maneuver. Besides, the velocity is added to the position of the planned obtained on the basis of this curve.
path, and then a real-time correspondence between path and velocity is To satisfy the velocity and acceleration constraints of intelligent
established with the property of cubic 3-D Bezier curve. First of all, the vehicle at the initial and terminal points of path, the path must possess
integrated path of lane changing maneuver is parameterized as two the property of more than second-order derivation. Therefore, the cubic
asymmetric cubic Bezier curves in the XY plane. The specific co- Bezier curve is chosen to satisfy the request by considering the higher
ordinates of path control points are selected by considering the corre- order curve has more calculation cost. The lane changing collision
sponding dynamic constraints and different driving requirements. avoidance path and velocity curve based on cubic 3-D Bezier curve is
Then, the velocity planning is designed with the coordinates of the Z shown in Fig. 1. The X and Y coordinates represent longitudinal and
axis and formulated as a nonlinear optimization problem with precisely lateral displacements respectively and the Z coordinate represents ve-
different dynamic constraints and traffic situations. At last, the model locity of subject vehicle. The entire path curve is shown in Fig. 2 which
prediction control (MPC) algorithm is utilized to track the reference is projected into the XY plane by Fig. 1 and the integrated velocity
path and velocity. curve is shown in Fig. 3 which is projected into the XZ plane by Fig. 1.
The rest of this paper is organized as follows: The path and velocity
planning are designed in Sections 2 and 3, respectively. The simulation 2.2. Collision avoidance path planning
results are provided in Section 4 and followed by the conclusive re-
marks in Section 5. In this section, when the subject vehicle detects the obstacle vehicle
in front of it, the lane changing maneuver is triggered to avoid collision
2. Lane changing collision avoidance path planning for safety. Then, the first collision avoidance path is planned in the XY
plane of cubic Bezier curve. The purpose of this path is to avoid colli-
2.1. Cubic 3-D Bezier curve sion between the subject vehicle and front vehicle and guarantee the
safety of lane changing maneuver at the beginning. The collision
Bezier curve is manipulated by connecting each control point to avoidance path is shown in Fig. 4.
form bounding polygon. Both ends of the curve are determined by the As we can see in Fig. 4, there are four control points and their co-
initial and terminal control points, and the remaining control points can ordinates are P10, P11, P12 and P13 respectively. The origin of coordinate
change the shape of curve. Bezier Curve has useful properties for path is fixed on the center of mass of subject vehicle at the initial position
planning: during the process of collision avoidance. The X and Y axis denote the
longitudinal and lateral displacements, D is the initial distance to the
a) It always starts at P0 and stops at Pn. leading vehicle in the original lane, d represents the reserved safe dis-
b) It is always tangent to P0 P1 and Pn 1 Pn at P0 and Pn respectively. tance to the forward vehicle to avoid collision, r is the radius of safety
c) It always lies within the convex hull consisting of its control points. circle which is designed to ensure the security when the subject vehicle
diverts to another lane. It is used to consider the influence of crosswind
Therefore, the position and shape of Bezier curve could be easily and the characteristic of understeer, and its value could be set around
changed by adjusting control points. Besides, the constraint conditions 0.5 m–1 m, w is the width of vehicle, and its value is usually 1.5 m–2 m.
can be satisfied with the tangent lines of the start and end points and
the properties of convex hull.
We define the locations of n + 1 control points Pi (i = 0,1,2, …, n) in
the space, and the n-order Bezier curve can be expressed as
n
P (t ) = Pi Bi, n (t ), t [0, 1]
i=0 (1)

where Pi and t are the control points and the parameter of Bezier curve
respectively. Bi,n(t) is n-order Bernstein function and can be presented
as
n!
Bi, n (t ) = Cni t i (1 t )n i = t i (1 t )n i
(n i ) ! n! (2)

where i = (0,1,2, …, n).


The lane changing collision avoidance path and velocity planning
method presented in this paper is combined with two asymmetrical Fig. 1. Cubic 3-D Bezier curve.

2
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

The second derivative of Eq. (5) can be deduced as

X¨1 (t ) = 6(x13 3x12 + 3x11 x10) t + 6(x12 2x11 + x10)


Y¨1 (t ) = 6(y13 3y12 + 3y11 y10 ) t + 6(y12 2y11 + y10 ) (7)
During the process of path planning, the front vehicle is assumed
that it just drives in the current lane straightly with the constant ve-
locity and will not take steering behavior, the velocity of subject vehicle
is supposed as constant because the velocity planning is conducted in
the next section. The initial longitudinal velocity is denoted as vx10, and
the longitudinal and lateral accelerations at initial state are equal to
zero. So the constraints of subject vehicle at initial state can be ex-
pressed as

X1 (0) = x10 = 0, X1 (0) = vx10, X¨1 (0) = 0


Y1 (0) = y10 = 0, Y1 (0) = 0, Y¨1 (0) = 0 (8)
Fig. 2. Path curve in the XY plane.
Then, substitute Eqs. (5)–(7) into Eq. (8), the results are as follows
v x10 2v x10
x10 = 0, x11 = 3
, x12 = 3
y10 = 0, y11 = 0, y12 = 0 (9)
The coordinates of the first three control points of collision avoid-
ance path have been determined by the above Eq. (9) and the issue of
specifying the coordinate of the fourth control point needs to be further
settled.
With respect to the parametric equation of collision avoidance path,
the curvature of path could be expressed as
X1 (t ) Y¨1 (t ) X¨1 (t ) Y1 (t )
K1 = 2 2
(X1 (t ) + Y1 (t )) 3/2
(10)
In Fig. 4, when the displacement in the Y axis is determined, the
curvature of path can be influence by the coordinate of x13. In a bid to
improve driving safety and comfort, it is important to reduce the cur-
Fig. 3. Velocity curve in the XZ plane.
vature as much as possible. Combining with Eq. (10), one can find that
the positive and negative properties of X¨1 (t ) will greatly impact the
value of K1 because the values of X1 (t ), Y1 (t ) and Y¨1 (t ) are not negative.
Since the coordinates of the first three control points have been de-
termined, the coordinates of the fourth control point can be serve to
make the curvature minish, which means the value of X¨1 (t ) should be
positive. By utilizing Eq. (7), the expression of Ẍ1 (1) can be written as
X¨1 (1) = 6x13 12x12 6x11 > 0 (11)
Thus

Fig. 4. Collision avoidance path.


x13 > 2x12 x11 = vx10 (12)

vx10 < x13 < D + la d (13)


Considering the security, the maximum of w is chosen as vehicle width.
The reserved safety distance d can be designed as
W is the width of lane and its value is general 3.5 m. la represents the
distance between the center of mass and the head of vehicle. d = kd (vx10 vf ) tc + d 0, kd [0, 1] (14)
The vital point of collision avoidance path planning based on Bezier
where kd is the dangerous coefficient and increases with the speed
curve is selecting the coordinates of control points because they decide
difference, vf is the velocity of front vehicle, tc is the time of collision
the position and shape of the path. The coordinates of control points are
and its value is D/(vx10 − vf), d0 is the minimum reserved safety dis-
chosen with respect to the initial state and dynamic constraints of
tance.
subject vehicle, the distance to the leading vehicle and geometric re-
From Fig. 4, the origin coordinate of safety circle can be expressed
lationship. The parametric equation of collision avoidance path can be
as (D + la d, w/2) via determining the safety distance d. Then, a
expressed as
tangent line is made to the edge of safety circle from the control point
X1 (t ) = x10 (1 t )3 + 3x11 (1 t ) 2t + 3x12 (1 t ) t 2 + x13 t 3 P12 and the point of tangency is control point P13. Finally, the co-
Y1 (t ) = y10 (1 t )3 + 3y11 (1 t )2t + 3y12 (1 t ) t 2 + y13 t 3 ordinate of P13 can be obtained by solving Eq. (15) which could be built
(5)
by using geometrical relationship.
The first derivative of Eq. (5) can be deduced as

X1 (t ) = 3(x13 3x12 + 3x11 x10 )t2


(x13 (D + la d )) 2 + y13 ( 2)
w 2
= r2

+ 6(x12 2x11 + x10) t + 3(x11 x10) (x13 x12)2 + y132 + r 2 = (D x12 )2 + ()


w 2
2 (15)
Y1 (t ) = 3(y13 3y12 + 3y11 y10 ) t 2
Through utilizing above approaches to obtain the control point P13,
+ 6(y12 2y11 + y10 ) t + 3(y11 y10 ) (6) the safety of subject vehicle while avoiding collision can be assured.

3
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

W y13
x21 = x13 + y13
(x13 x12)
y21 = W (19)
Due to the uncertainty of the end point and time of lane changing, to
get the coordinate of control point P23, the lane changing time T and
terminal coordinate x23 are chosen as variables and the related objec-
tive function is design by taking advantage of genetic algorithm (GA).
The objective function expression can be designed as
2 2 2
1 K2 T S
J1 = w11 dt + w12 + w13
0 Km Tm Sm (20)
where w11, w12, w13 are weight coefficients and they can be obtained by
the mature driver model and actual lane changing data, K2 is the cur-
Fig. 5. Various collision avoidance paths with diverse initial distance D.
vature of lane changing path, T is the time of lane changing, S is the
length of lane changing path, Km, Tm, Sm are the maximums of curva-
ture, time and length respectively. The first part of the function re-
Moreover, the performance of stability and safety when the subject presents the curvature of lane changing path, the second part indicates
vehicle avoids collision can also be improved by adjusting the co- the traveling time of lane changing process and the third part represents
ordinate of P13. Once the coordinates of four control points have been the length of lane changing path.
selected, the collision avoidance path is also determined. When the For the sake of obtaining the appropriate optimal solution on the
state and constraint of subject vehicle or the conditions of leading ve- basis of genetic algorithm (GA), the constraint relationship between T
hicle change, we can adjust the coordinates of control points to adapt to and x23 needs to be defined which can be deduced from the path curve.
the updated and dynamic environment. The various collision avoidance The lane changing path curve is shown in Fig. 7.
paths with diverse initial distance D are presented in Fig. 5.
In Fig. 7, AD represents the lane changing path and its length is S,
AB and BD are the tangent lines at the initial and terminal of the path.
2.3. Lane changing path planning Their geometrical relationships can be derived as
ED < S < AB + BD < AE + ED (21)
In realistic traffic environment, the end point of lane changing
maneuver is uncertain and could be influenced by some important
x23 x20 < vx10 T < W y20 + x23 x20 (22)
factors, such as path curvature, lane changing time and traffic effi- The genetic algorithm (GA) is introduced to receive the most op-
ciency. To handle the uncertainty of end point and the time of lane timal value of T and x23, which means to solve the minimum of ob-
changing process, this paper applies genetic algorithm (GA) to design jective function J1 within constraints. So the issue of path planning
the lane changing path based on cubic Bezier curve. The lane changing converts into a nonlinear optimization problem with disparate con-
path is shown in Fig. 6. straints.
The parameter equation of this path can be expressed as
min:J1 = (T , x23)
X2 (t ) = x20 (1 t )3 + 3x21 (1 t )2t + 3x22 (1 t ) t 2 + x23 t 3 s . t . x23 x20 < vx10 T < W y20 + x23 x20
Y2 (t ) = y20 (1 t )3 + 3y21 (1 t )2t + 3y22 (1 t ) t 2 + y23 t 3 0 ay a y max
(16)
0 K2 Km
To ensure that the two paths are continuous in shape at the con- Smin S Smax
nection point, the control point P20 should coincide with P13. Tmin T Tmax (23)
Furthermore, to assure intelligent vehicle can track the path success-
The obtained optimal solutions are T* and x 23
* , so the coordinate of
fully, the curve of yaw angle should be continuous all the time. The
control point P23 is (x 23
* , W ).
function expression of yaw angle curve can be obtained as
When the subject vehicle arrives at the destination of path, it needs
Y (t ) to satisfy the terminal state dynamic constraint which the longitudinal
= actan and lateral accelerations at final state should also be equal to zero.
X (t ) (17)
Then, the constraints can be given as
Hence, the slope of path at the connection point should be the same, X2 (1) = x23, X2 (1) = vx10, X¨2 (1) = 0
which means that control points P12, P13, P20 and P21 must be in a
Y2 (1) = y23 = W , Y2 (1) = 0, Y¨2 (1) = 0 (24)
straight line. As a result, the coordinate of control point P21 can be
figured out according to Eq. (18).

K P12, p13 = K P20, P21 = tan (18)

Fig. 6. Lane changing path. Fig. 7. Lane changing path curve.

4
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

vehicle needs to satisfy other dynamic constraints during driving. These


dynamic constraints are:

1) Lateral acceleration constraint can be given as

a y = K1 Z12 min{µg, a y max } (30)

where aymax is the maximum allowed lateral acceleration.

2) In order to guarantee a pure rolling motion to avoid the wheel


skidding phenomenon, it is necessary to verify that the force
transmitted to the ground is smaller than the ground friction force.
Fig. 8. Different lane changing paths with disparate terminal coordinates. For simplicity, the total tyre force is expressed as

Ft = Fx2 + Fy2 = (mZ1)2 + (mK1 Z12) 2 µmg (31)

According to Eq. (24), the coordinate of control point P22 can be


obtained as
x21 + x23 Thus, the longitudinal acceleration constraint can be computed as
x22 = 2
y22 = W (25) ax = Z1 min{ µ2 g 2 K12 Z14 , ax max } (32)

Up to now, four control points of lane changing path have been where axmax is the maximum allowed longitudinal acceleration.
determined. Facing with different lane changing time and distance re-
quirements, disparate destinations of lane changing path can be ob- 3) In a bid to assure that the planned velocity does not exceed the
tained by varying the weight coefficients of each item in the objective maximum velocity that vehicle can reach, the maximum velocity
function, and it is more in line with the driver demand and actual lane constraint should be set as
changing environment, so the designed lane changing path has varia-
Z1 vmax (33)
bility. The different lane changing paths with disparate terminal co-
ordinates x23 are displayed in Fig. 8.
Through the aforementioned two sections, the entire lane changing
collision avoidance path planning has been completed and the whole where vmax is the maximum allowed velocity.
path curve is shown in Fig. 2. The velocity planning needs to not only satisfy different kinds of
dynamic constraints, but also meet the various requirements of human
3. Velocity planning driver and actual driving conditions. Thus, the objective function is
used to get different results and its expression is designed as
For the aim to ensure the feasibility, comfort and safety of the
( ) dt
2
planned path when the intelligent vehicle follows the path, it is im-
perative to plan the velocity of subject vehicle. In the above section, the
J2 = w21 0
1
(1 Z1 2
vmax )
dt + w22 0
1 K1 Z12
ay max

( )
2
planed lane changing collision avoidance path is piecewise and asym- + w23
1 Z1
dt
0 ax max (34)
metrical, so the velocity planning also needs to be accordingly piece-
wise, which is divided into two subparts: a collision avoidance velocity where w21, w22, w23 are weight coefficients and they can be obtained by
planning and a lane changing velocity planning. the mature driver model and actual lane changing data, Z1 and Z1 re-
present velocity and acceleration of collision avoidance path, respec-
3.1. Collision avoidance velocity planning tively, K1 is the curvature of collision avoidance path. The first part of
the function denotes the velocity of subject vehicle during the collision
In this section, the Z axis coordinate of cubic 3-D Bezier curve is avoidance process, and the second part and the third part represent the
applied to plan the velocity. The parameter function of collision lateral and longitudinal accelerations respectively.
avoidance velocity planning can be designed as To specify the above coordinates, Sequential Quadratic
Programming (SQP) algorithm is chosen to work out the minimum of
vx1 (t ) = Z1 (t ) = z10 (1 t )3 + 3z11 (1 t )2t
the objective function J2. So the velocity planning issue is transformed
+ 3z12 (1 t ) t 2 + z13 t 3 (26) into a nonlinear constrained optimization problem with several dy-
The parameter function of acceleration can be deduced as namic constraints.

a x1 (t ) = Z1 (t ) = 3(z13 3z12 + 3z11 z10) t 2 min:J2 (z12 , z13)


+ 6(z 12 2z11 + z10) t + 3(z11 z10) (27) s . t . Z1 vmax
K1 Z12 min{µg , a y max}
Take the initial condition of velocity and acceleration into account.
Z1 min{ µ2g 2 K12 Z14 , a x max } (35)
Z1 (0) = vx10, Z1 (0) = 0 (28)
Substitute Eq. (28) into Eqs. (26) and (27), the results can be ob- In a word, the planning method based on cubic 3-D Bezier curve and
tained as SQP algorithm are combined to specify the coordinates of control points
of collision avoidance velocity curve within vehicle dynamic con-
z10 = vx10 , z11 = vx10 (29)
straints. By modifying the values of w21, w22 and w23, variable velocity
Therefore, if the coordinates of z12 and z13 are confirmed, the col- curves could be achieved to meet different demands of human driver
lision avoidance velocity planning will be accomplished. Firstly, the and driving conditions.

5
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

3.2. Lane changing velocity planning

The approach of lane changing velocity planning is similar to the


method in the last section, and the parameter equation of lane changing
velocity planning can be designed as

vx2 (t ) = Z2 (t ) = z20 (1 t )3 + 3z21 (1 t ) 2t


+ 3z22 (1 t ) t 2 + z23 t 3 (36)

The parameter function of acceleration can be deduced as

a x2 (t ) = Z2 (t ) = 3(z23 3z22 + 3z21 z20) t 2


+ 6(z 22 2z21 + z20) t + 3(z21 z20) (37)

To keep the continuity of the velocity and acceleration curve of the


two paths at the connection point, the starting point of the lane chan- Fig. 9. Disparate lane changing collision avoidance velocity curves.
ging velocity should coincide with the end point of the collision
avoidance velocity and the slopes of two curves at jointed point should 4. Simulation results
be the same. What's more, the velocity planning must meet the velocity
and acceleration constraints at the terminal state. In this section, to validate the feasibility of the path and velocity
Z2 (0) = Z1 (1), K P12, p13 = K P20, P21 planning method approached in this work, the simulations use the
model predictive control (MPC) algorithm to track the reference path in
Z2 (1) = vx 23, Z2 (1) = 0 (38)
a high-fidelity CarSim-Simulink co-simulation platform.
where vx23 and ax23 are the final longitudinal velocity and acceleration
when the subject vehicle arrives at the target position. 4.1. MPC tracking method
Substitute Eq. (38) into Eqs. (36) and (37), the results can be ob-
tained as Model predictive control (MPC) algorithm is an appealing technique
z20 = z13 for tracking path due to the forecast of the future path and the ability to
z21 = z20 +
x21 x20
(z z12) deal with multi-objective constraints. Accordingly, it can effectively
x12 13
x 13
improve the following precision and address different driving condi-
z22 = z23 = vx 23 (39) tions for intelligent vehicle.
Since the velocity of vehicle at the end of the path is generally By employing the MPC method, the dynamic vehicle model is
unknown, it is necessary to determine the coordinate of control point needed firstly. The equation of the vehicle model is compactly written
z23 to achieve the whole velocity curve. So, to obtain the velocity of as
vehicle final state, considering that the velocity should be alterable with
(t ) = f ( (t ), u (t )) (42)
the driver demand and actual driving circumstance, another objective
function is used and its expression is designed as where (t ) = [y , x , , , Y , X ,]T is
the vehicle state vector, u = [δ] is
2 2 control input vector.
1 Z2 1 K2 Z22
J3 = w31 1 dt + w32 dt Secondly, the optimal solution pursued in the model predictive
0 vmax 0 a y max controller control process is to obtain the best path tracking effect with
1 Z2
2
designing the objective function and setting the dynamic constraints.
+ w33 dt The objective function for the path following based on the MPC method
0 ax max (40)
can be designed as
where w31, w32, w33 are weight coefficients and they can be obtained by
the mature driver model and actual lane changing data, Z2 and Z2 re- Jmin 1 ( (t ), u (t 1), U (t ))
present velocity and acceleration of lane changing path, respectively, =
Np
i=1
Y (t + i t ) Yref (t + i t ) 2
+
Nc 1
i=1
u (t + i t ) 2
R + 2
Q
K2 is the curvature of lane changing path. The first part of the function
is the velocity of subject vehicle during the lane changing process, the (43)
second part and the third part illustrate the lateral and longitudinal where Np and Nc are the prediction horizon and control horizon, re-
accelerations respectively. spectively. Q and R are the weight matrices, ρ is the weight factor, ε is
Similarly, the SQP algorithm is also implemented to solve the the relaxation factor. The first part of the equation means the ability of
minimum value of the above objective function J3 and the velocity path following, the second part represents the requirement for smooth
planning problem is accordingly transformed into a nonlinear con- change of control vector and the third part prevents the no solution of
strained optimization problem. equation.
Thirdly, the constraints should also be set and the optimization
min:J3 (z23)
problem is expressed as
s . t . Z2 vmax
K2 Z22 min{µg , a y max } min J ( (t ), u (t 1), U (t ))
U,
Z2 min{ µ2 g 2 K22 Z24 , ax max } (41) s. t . Umin Ut Umax
Umin A Ut + Ut Umax (44)
As we can see, the above methods can obtain the coordinates of the
control point of lane changing velocity planning. Therefore, the in- where Umin and Umaxare the boundary conditions of control vector,
tegrated velocity planning curve is formed by the collision avoidance ΔUminand ΔUmaxare the constraints of control increment.
and lane changing velocity curve, which is shown in Fig. 3. The dis- The solution process of the above optimization problem is com-
parate lane changing collision avoidance velocity curves with different pleted within each control cycle, and the optimal control incremental
kinds of driving situations are exhibited in Fig. 9. sequence in the control time domain is as follows

6
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

Ut* = [ut*, ut*+ 1, ...,ut*+ Nc ]T


1 (45)

4.2. Cases of several driving conditions

In the MPC controller, some key parameters need be determined.


The prediction horizon Np is 20, the control horizon Nc is 5 and the
preview time τ is 0.02 s. To analyze the effectiveness of planned path
and velocity with different subject vehicle velocities, two simulation
cases including the constant and variable velocity driving conditions
which the initial velocities are 72 km/h and 36 km/h respectively are
involved. In these cases, the initial distance between subject and
leading vehicle is set as 30 m, the front vehicle velocity is 30 km/h. To
further analyze the results of various conditions of the front vehicle,
different initial distances and velocities simulation cases of the front
vehicle is also included.
Case 1: 72 km/h driving condition
This case represents the high speed lane changing condition and the

Fig. 10. (continued)

simulation results in case 1 are shown in Fig. 10(a–e). Fig. 10(a) and (b)
represent the path following results about displacement Y in the X axis
and in the driving time t axis respectively. Fig. 10(c) depicts the si-
mulation results of lateral acceleration including constant and variable
velocity driving conditions. Fig. 10(d) and (e) illustrate the simulation
results of the vehicle velocity and longitudinal acceleration under the
variable velocity driving condition.
As displayed in Fig. 10(a), the real tracking path of vehicle can
follow the referenced planned path in the overall trend with high pre-
cision and the path following deviation of variable velocity is slightly
smaller than the result of constant velocity. In Fig. 10(b), the lane
changing time of variable velocity is longer than the result of constant
velocity because of deceleration. According to Fig. 10(c), the lateral
acceleration of variable velocity is significantly reduced at the peaks on
account of the velocity decreasing at the position of large curvature, so
the planned velocity result can enhance the driving stability and im-
prove the riding comfort although prolong the lane changing time.
Fig. 10(d) shows the curves of the reference velocity and real velocity,
and the following error is generally small. Then Fig. 10(e) shows the
curve of longitudinal acceleration. It can be seen that the velocity and
longitudinal acceleration of subject vehicle in this case satisfy the dy-
namic constraints and are all within the settled constraint ranges.
Case 2: 36 km/h driving condition
In the case 2, to further verify the adaptation of proposed planning
method, the low speed lane changing condition is considered in this
work. The simulation results are shown in Fig. 11(a–e). Fig. 11(a) and
(b) demonstrate the path following results about displacement Y in the
X axis and in the driving time t axis, respectively. Fig. 11(c) indicates
the simulation results of lateral acceleration including constant and
variable velocity driving conditions. Fig. 11(d) and (e) describe the
results of the vehicle velocity and the longitudinal acceleration under
the variable velocity driving condition.
As shown in Fig. 11(a), the real tracking path of vehicle can also
Fig. 10. Simulation results of 72 km/h driving condition.

7
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

follow the reference planned path in the overall trend with high pre-
cision and the path following deviation of variable velocity is slightly
larger than the result of constant velocity. In Fig. 11(b), the lane
changing time of variable velocity is obviously shorter because of ac-
celeration. In Fig. 11(c), the lateral acceleration of variable velocity is
greatly augmented due to the velocity increase during the process, thus
the planned velocity result can shorten lane changing time and improve
traffic efficiency while sacrificing some driving comfort. Fig. 11(d)
shows the curves of the reference velocity and real velocity, and the
following error is small. Then Fig. 11(e) displays the curve of long-
itudinal acceleration. It can be seen that the velocity and longitudinal
acceleration of subject vehicle in this case satisfy the dynamic con-
straints and are also within the settled constraint ranges.
Case 3: Different initial distances and velocities of the front vehicle
In this case, the velocity of subject vehicle is set as 72 km/h. The
simulation results of different initial distances and velocities of the front
vehicle are shown in Figs. 12 and 13, respectively.
In Fig. 12, the various initial distances are set as 25 m, 30 m, 35 m

Fig. 11. (continued)

Fig. 12. Path following results with different initial distances.

Fig. 11. Simulation results of 36 km/h driving condition. Fig. 13. Path following results with different velocities of the front vehicle.

8
L. Chen, et al. Advances in Engineering Software xxx (xxxx) xxx–xxx

and the velocity of the leading vehicle is 30 km/h. The simulation re- cluttered environments. Adv Eng Software 2016;92:40–7.
sults manifest that the whole paths planned by this method could [7] Broggi A, Buzzoni M, Debattisti S, Grisleri P. Extensive tests of autonomous driving
technologies. IEEE Trans Intell Transp Syst 2013;14(3):1403–15.
change correspondingly with different initial distances. Each reference [8] Lian YF, Wang XY, Tian YT, Liu K. Lateral collision avoidance robust control of
path can be tracked during the process and the following deviations are electric vehicles combining a lane-changing model based on vehicle edge turning
acceptable. trajectory and a vehicle semi-uncertainty dynamic model. Int J Autom Technol
2018;19(2):331–43.
In Fig. 13, the diverse velocities of the front vehicle are designed as [9] Wang Z, Deng WW, Zhang SM, Shi JP. Vehicle automatic lane changing based on
30 km/h, 40 km/h, 50 km/h and the initial distance is 30 m. The si- model predictive control. SAE Int J Passenger Cars-Electron Electr Syst
mulation results indicate that the integrated paths designed by this 2016;9(1):231–6.
[10] Petrov P, Nashashibi F. Modeling and nonlinear adaptive control for autonomous
method could change correspondingly with different velocities of the vehicle overtaking. IEEE Trans Intell Transp Syst 2014;15(4):1643–56.
leading vehicle. Every reference path can be followed with high accu- [11] Suh JS, Chae HS, Yi KS. Stochastic model-predictive control for lane change deci-
racy. Therefore, the adaptation and feasibility of the presented planning sion of automated driving vehicles. IEEE Trans Vehicular Technol
2018;67(6):4771–82.
method are validated.
[12] Mehdi KE, Victor LK, Winnie D. Categorization of the lane change decision process
on freeways. Transp Res Part C 2016;69:515–26.
5. Conclusion [13] Fassbender D., Mueller A., Wuensche H.J. Trajectory planning for car-like robots in
unknown, unstructured environments. Conference on intelligent robots and systems
(IROS). IEEE 2014: 3630–3635.
In this paper, the lane changing collision avoidance path and velo- [14] Oh JS, Lim KI, Kim HJ. A research of obstacle detection and path planning for lane
city planning method based on cubic 3-D Bezier curve is proposed. change of autonomous vehicle in urban environment. J Instit Control Rob Syst
Firstly, the planning method by exploiting piecewise cubic 3-D Bezier 2015;21(2):115–20.
[15] Yang I, Kim HJ, Jeon WH, Kim H. Development of realistic shortest path algorithm
curve could enhance the security of lane changing process without considering lane changes. J Adv Transp 2016;50(4):541–51.
collision and improve the flexibility of path and velocity to accom- [16] Leng Z, Minor MA. Curvature-based ground vehicle control of trailer path following
modate different traffic circumstances. Moreover, the planned results considering sideslip and limited steering actuation. IEEE Trans Intell Transp Syst
2017;18(2):332–48.
can satisfy multiple dynamic constraints. Finally, the path following [17] Ji XW, Liu YL, He XK, Yang KM, Na XX, Lv C, Liu YH. Interactive control paradigm
simulation experiments including different driving conditions are car- based robust lateral stability controller design for autonomous automobile path
ried out through applying MPC algorithm, the performances of various tracking with uncertain disturbance: a dynamic game approach. IEEE Trans
Vehicular Technol 2018;67(8):6906–20.
velocities and diverse initial distances are analyzed. The results verify [18] He XK, Liu YL, Lv C, Ji XW, Liu YH. Emergency steering control of autonomous
the effectiveness of the proposed planning approach and illustrate this vehicle for collision avoidance and stabilization. Vehicle Syst Dyn 2018. https://
method can meet various kinds of driving situations. In the future work, doi.org/10.1080/00423114.2018.1537494.
[19] Funke J, Brown M, Erlien SM, Gerdes JC. Collision avoidance and stabilization for
the dynamic control of intelligent vehicle during path following will be
autonomous vehicles in emergency scenarios. IEEE Trans Control Syst Technol
researched in detail to improve the driving stability. 2017;25(4):1204–16.
[20] Itoh M, Inagaki T. Design and evaluation of steering protection for avoiding colli-
Acknowledgments sions during a lane change. Ergonomics 2014;57(3):361–73.
[21] Jolly KG, Sreerama KR, Vijayakumar R. A Bezier curve based path planning in a
multi-agent robot soccer system without violating the acceleration limits. Rob
The authors are grateful for the support provided by the National Auton Syst 2009;57:23–33.
Natural Science Foundation of China (Grant No. U1564201), and the [22] Li AJ, Li SM, Li DR. Research on key technologies in intelligent vehicle trajectory
planning. Mech Sci Technol Aerosp Eng 2013;07:1022–6.
Primary Research & Development Plan of Jiangsu Province (Grant No. [23] Maekawa T, Noda T, Tamura S, Ozaki T, Machida K. Curvature continuous path
BE2016149). generation for autonomous vehicle using b-spline curves. Comput-Aided Design
2010;42(4):350–9.
[24] Yang D, Zheng SY, Wen C, Peter J, Bin R. A dynamic lane-changing trajectory
Conflict of interest planning model for automated vehicles. Transp Res Part C 2018;95:228–47.
[25] Funke J, Gerdes JC. Simple clothoid lane change trajectories for automated vehicles
The authors declare that there is no conflict of interests regarding incorporating friction constraints. J Dyn Syst Measur Control 2016;138(2):1–9.
[26] Chen JJ, Zhao P, Mei T, Liang HW. Lane change path planning based on piecewise
the publication of this paper.
bezier curve for autonomous vehicle. IEEE international conference on vehicular
electronics and safety (ICVES). 2013. p. 17–22.
References [27] Bae I., Moon J., Park H., Kim J.H., Kim S. Path generation and tracking based on a
bezier curve for a steering rate controller of autonomous vehicles. IEEE conference
on intelligent transportation systems 2013: 436–441.
[1] Zhang H, Wang J. Active steering actuator fault detection for an automatically- [28] Chen C, He Y, Bu C, Han J, Zhang X. Quartic Bézier curve based trajectory gen-
steered electric ground vehicle. IEEE Trans Vehicular Technol eration for autonomous vehicles with curvature and velocity constraints. IEEE in-
2017;66(5):3685–702. ternational conference on robotics and automation. 2014. p. 6108–13.
[2] Cai YF, Liu Z, Wang H, Sun XQ. Saliency-based pedestrian detection in far infrared [29] Cao HT, Song XL, Zhao S, Bao S, Huang Z. An optimal model-based trajectory
images. IEEE Access 2017;5:5013–9. following architecture synthesising the lateral adaptive preview strategy and
[3] Wang H, Dai L, Cai YF, Sun XQ, Chen L. Salient object detection based on multi- longitudinal velocity planning for highly automated vehicle. Vehicle Syst Dyn
scale contrast. Neural Netw 2018;101:47–56. 2017;55(8):1143–88.
[4] Shuai Z, Zhang H, Wang J, Li J, Ouyang M. Combined AFS and DYC control of four- [30] Zhang H, Yang SW. Smooth path and velocity planning under 3D path constraints
wheel-independent-drive electric vehicles over CAN network with time-varying for car-like vehicles. Rob Auton Syst 2018;107:87–99.
delays. IEEE Trans Vehicular Technol 2014;63(2):591–602. [31] Jorge V, Vicente M, Joshué P, Jorge G. Smooth path and speed planning for an
[5] Yu Y, Kamel AE, Gong GH. Modeling and simulation of overtaking behavior in- automated public transport vehicle. Rob Auton Syst 2012;60:252–65.
volving environment. Adv Eng Software 2014;67:10–21.
[6] Li B, Shao ZJ. Precise trajectory optimization for articulated wheeled vehicles in

You might also like