Research Article

International Journal of Advanced

Robotic Systems
November-December 2016: 1–13
Data-driven automatic parking ª The Author(s) 2016
DOI: 10.1177/1729881416663667
constrained control for four-wheeled arx.sagepub.com

mobile vehicles

Wenxu Yan1, Jing Deng1, and Dezhi Xu1

Abstract
In this article, a novel data-driven constrained control scheme is proposed for automatic parking systems. The design of
the proposed scheme only depends on the steering angle and the orientation angle of the car, and it does not involve any
model information of the car. Therefore, the proposed scheme-based automatic parking system is applicable to different
kinds of cars. In order to further reduce the desired trajectory coordinate tracking errors, a coordinates compensation
algorithm is also proposed. In the design procedure of the controller, a novel dynamic anti-windup compensator is used to
deal with the change magnitude and rate saturations of automatic parking control input. It is theoretically proven that all
the signals in the closed-loop system are uniformly ultimately bounded based on Lyapunov stability analysis method.
Finally, a simulation comparison among the proposed scheme with coordinates compensation and Proportion Integration
Differentiation (PID) control algorithm is given. It is shown that the proposed scheme with coordinates compensation has
smaller tracking errors and more rapid responses than PID scheme.

Keywords
Data-driven constrained control, automatic car parking, dynamic anti-windup, coordinates compensation

Date received: 29 April 2016; accepted: 21 June 2016

Topic: Special Issue - Manipulators and Mobile Robots

Topic Editor: Michal Kelemen

Introduction include parallel parking, vertical parking, and oblique park-

ing, as shown in Figure 1. Parallel parking is most common
Nowadays, with the fast development of science and tech-
in our daily life, so the parallel automatic parking system
nology and increase of people’s living standard, the num-
has become a hot spot of current research.3,4 Therefore, the
ber of cars is increasing very fast. The space of car parking
work of this article is to study the parallel parking.
significantly reduced, and traffic accidents are particu-
Parallel parking consists of three procedures5: (1) park-
larly prone to happen when backing the car for these driv-
ing space detection, (2) path planning, and (3) path track-
ers who have less experience. The university of Michigan ing. Once drivers choose the automatic parking mode, the
study shows that manual parking is one of the important
width and length of the parking space are detected by
reasons why the driver is easy to cause the traffic acci-
dent.1 Thus, the automatic parking problem has become a
hot research topic. 1
School of Internet of Things Engineering, Institute of Electrical Engineering
Automatic parking system can avoid parking accident. It and Intelligent Equipment, Jiangnan University, Wuxi, China
uses the sensor technology, computer technology, and auto-
matic control technology to accurately perceive the parking Corresponding author:
Dezhi Xu, School of Internet of Things Engineering, Institute of Electrical
environment and planning a optimal parking path and then Engineering and Intelligent Equipment, Jiangnan University, Wuxi 214122,
lead the vehicle to the target by controlling the vehicle China.
automatically complete the path tracking.2 Parking types Email: xudezhi@jiangnan.edu.cn

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License
(http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 International Journal of Advanced Robotic Systems

artificial neural network. The authors realized the virtual

process of automatic parking in the simulation software.
But there are many types of cars on the market. For each
new type of the car, the parameters of PID controller need
to be adjusted, the fuzzy control rules should be reformu-
lated, and the neural network control need to be retrained.
Hence, the automatic parking system established by the
above algorithm is poor in portability.
The data-driven control methods have been applied in
several fields.13–16 Model-free adaptive control (MFAC),
one of the data-driven control methods, was put forward
by Hou. In MFAC, combined with a innovative concept
Figure 1. Classification of parking: (a) parallel parking, (b) vertical named pseudo partial derivative (PPD), an equivalent
parking, and oblique parking. dynamic linearization data model is developed. First, it takes
the place of the general discrete time nonlinear system at
ultrasonic sensors mounted on the side of the car. When the each current operating point. Next, the estimation of PPD is
drivers stop a car at the beginning position, environmental only conducted online through the input/output (I/O) data
geometry and target position are calculated automatically from the controlled system. Eventually, the controller is
from the sensor information. Then, a target parking trajec- designed. At present, the MFAC method has been success-
tory is created by the automatic parking controller and the fully applied in the fields of smart grid, chemical industry,
car is guided into the target parking space by controlling the welding process, wind power generation, urban expressway,
electronic power steering.6 artificial heart rate regulation, and so on.17–23 Theoretical
Four-wheeled vehicle is the most commonly used in our analysis, simulation, and practical application show that the
daily life. Its dynamics model can be transformated into MFAC method is simple and practical, small computational
nonholonomic chained system,7 and the corresponding burden, and strong robustness; it can deal with the control
control algorithm can be designed based on the nonholo- problem of unknown nonlinear time-varying systems.24
nomic chained system. But the automatic parking process In this article, we propose a novel data-driven con-
has the characteristics of nonlinearity, time variation, mul- strained control scheme with coordinates compensation for
tivariable, and so on. It is difficult to establish an accurate automatic parking system. The proposed scheme consists
model of the nonholonomic chained system. Even if the of constrained control algorithm, parameter estimation
model is established, the corresponding control algorithm algorithm, parameter reset algorithm, and coordinates com-
is very complex and the computation is very large. So it is pensation algorithm. The design of the control scheme is
difficult to popularize this method. In the study by Chen only based on the I/O data of the automatic parking system,
et al.,8 the fuzzy-Proportion Integration Differentiation which does not include the vehicle model information.
(PID) controller is designed for automatic car parking sys- Therefore, the proposed scheme is applicable to different
tem, the front wheel angle is the output of the controller, vehicle types. The proposed scheme gets a better control
and the speed error is the input of the controller. In the effect by adopting the coordinates compensation algorithm.
study by Liang et al.,9 the automatic parking system is Because the control input is subject to change magnitude
designed based on the fuzzy controller by adopting fuzzy and rate constraints, a novel dynamic anti-windup compen-
control algorithm. The vehicle coordinate position error sator is used to deal with the saturation problem. The simu-
and the orientation angle of the car error are the output of lation results have verified the superiority of the proposed
the controller, and the front wheel angle is the input of the constrained control scheme.
controller. Different fuzzy logic controllers are designed
for different parking situations in the study by Li et al.,10
and the experimental verification is done on the car robot Problem formulation for automatic
that on the basis of field programmable gate array. In the parking process
study by Baturone et al.,11 the fuzzy controller is designed
by adopting the position coordinate of the car, the direction Parking space detection
of the vehicle body, vehicle speed, and trajectory curvature The diagram of parking space detection is shown in
as inputs, and the marching direction, vehicle speed, and Figure 2. There are ultrasonic probes on the vehicle. The
the target curvature size as outputs. The experiment is done ultrasonic probes in the front and rear of the vehicle are
on a car robot. In the study by Heinen et al.,12 the authors used to ensure parking safety. The ultrasonic probes
proposed a control method based on neural network by mounted on the side of the vehicle are used to detect the
adopting the position coordinate of the car, the orientation parking space. Start the automatic parking system when
angle of the car as the input of artificial neural network, and the vehicle is ready to pull over, and the system detects the
the vehicle speed and the front wheel angle as the output of distance information by the ultrasonic probes mounted on
Yan et al. 3

Figure 4. Vehicle model.

P3 P 2 which goes through the beginning location and is
parallel to the x-axis. Then make a circle O 3 that is tangen-
Figure 2. Diagram of parking space detection. tial to the straight lines P0 P1 and P3 P2 . Denote the point of
tangency of this circle O 3 and line P3 P2 as P2 and the point
of tangency of this circle O 3 and line P0 P1 as P 1 . Then the
path planning of parallel parking is illustrated as the curve
in Figure 3.
From this diagram, the angle  of line P0 P1 and the
coordinates of P 0 and P2 are got through some geometric
calculations as
0:5Wc þ R2 cos   R1 ð1  cos Þ
 ¼ tan1 (2)
Lp  R 1 sin   R2 sin   S
 
Figure 3. Path planning of automatic parking. P0 : R1 sin ; R1 ð1  cos Þ (3)
0 1
the side of the vehicle . If there is a vehicle parked at the
R 1 =cos   R 1 þ d þ 0:5Wc
roadside, the distance detected by the ultrasonic probe P2 : @ ; d þ 0:5Wc A (4)
mounted on the right side of the vehicle is d1 , otherwise, tan 
the detection distance is d2 . If there is a parking space, we
can get the dashed-dotted line pulse as shown in Figure 2. where d is the lateral distance between the two vehicles as
According to the vehicle speed and pulse duration, the shown in Figure 3.
length of parking spaces Lp can be obtained. If Lp is greater Based on Figure 3, the minimal length of parking space
than the minimal available length of parking space Lp min , Lp min is given when Lp 0 p 1 ¼ Lc
the parking space is identified as available. Otherwise, the Lp min ¼ ðR 1 þ R2 Þ sin  0 þ Lc cos  0 (5)
vehicle will continue to move forward to find other avail-
able parking spaces. After finding the available parking where Lp0 p 1 denotes the length of line P0 P1 , and for the  0 ,
spaces, the systems enter the stage of path planning. we consider the following equation with
Lc sin  0 ¼ ðR1 þ R2 Þcos  0 þ 0:5Wc  R1 (6)
Path planning
Path tracking
The path planning of parallel parking consists of the fol-
lowing procedures. The given parameters are the vehicle Let us consider the four-wheeled car shown in Figure 4.
width Wc , the vehicle length Lc , and the length of parking Obey the assumption that sideslip does not happen, the
space Lp (refer to Figure 3). Make a circle O 1 at the center state space equation of four-wheeled car25,26 is obtained as
of (0, R 1 ) and with a radius of R1 . This circle is tangent to 8
> x_ ¼ v cos 
the x-axis at the origin O. Make a circle O 2 at the center of >
>
>
< y_ ¼ v sin 
the corner O2 and with a radius of (7)
>
> v tan 
> _
:¼ L
>
R2 ¼ 0:5Wc þ S (1)
where S is the safe distance between the two cars. where (x, y) is the coordinate of the center of the rear
Draw the common tangent P0 P 1 of both circles O1 and wheel axle, L denotes the wheel base,  is the orientation
O 2 ; the point of tangency with O1 is denoted as P0 and the angle of the car, and v and  are the vehicle speed and the
point of tangency with O 2 is denoted as P1 . Draw a line steering angle, respectively.
4 International Journal of Advanced Robotic Systems

In the practical application, equation (7) is usually trans-

formed into the following discrete model e 0 ðk þ 1Þ ¼ T ðkÞ~
ðkÞ þ Fe 0 ðkÞ (13)
8 where ~ðkÞ ¼ ðkÞ  ^ðkÞ denotes the error of parameter
> xðk þ 1Þ ¼ xðkÞ þ Tv cosððkÞÞ
>
> estimation. The adaptive update algorithm for the estimated
>
< yðk þ 1Þ ¼ yðkÞ þ Tv sinððkÞÞ
parameters ^ðkÞ is chosen as
(8)
>
> Tv tanððkÞÞ  
>
> ^ðk þ 1Þ ¼ ^ðkÞ þ ðkÞðkÞ e 0 ðk þ 1Þ  Fe 0 ðkÞ (14)
: ðk þ 1Þ ¼ ðkÞ þ L
The gain ðkÞ is chosen as:
where T denotes sampling time.  1
In addition, the input ðkÞ of the vehicle cannot change too ðkÞ ¼ 2 jjðkÞjj 2 þ  (15)
fast within a small time interval. So the control input is subject
to change magnitude and rate constraints as following where  is a positive constant; hence, ðkÞ is positive defi-
nite for all the k. Noticed that, with the assumption that
 min  ðkÞ   max _
_ min  ðkÞ  _ max (9) jjðkÞjj  g, ðkÞ can be lower bounded as
Based on the state space equation of four-wheeled car, 2
jjðkÞjj  ¼ >0
the coordinate of the car ðx; yÞ is determined by the orienta- g2 þ 
tion angle  of the car with little change in the parking
In view of equations (13) and (14) and condition
speed. Therefore, ensure the coincidence of the coordinate
ðk þ 1Þ  ðkÞ, which denotes the slow time-varying sys-
of the car and target trajectory by controlling the steering
tem, the estimation dynamic error is obtained as
angle  to control the orientation angle  of the car is the
control objective of automatic parking system. e 0 ðk þ 1Þ ¼ T ðkÞ~
ðkÞ þ Fe 0 ðkÞ
(16)
~ðk þ 1Þ ¼ H ~ðkÞ
Main results
where H is given by
In this section, we will put forward a novel data-driven con-
H ¼ I2  ðkÞðkÞT ðkÞ
strained control scheme with coordinates compensation for
automatic car parking system. Major contributions in the fol- and I 2 denotes the ð2  2Þ identity matrix.
lowing works are (1) an unknown PPD estimation algorithm,
(2) proposed a data-driven constrained control algorithm via Theorem 1. The equilibrium ½e 0 ; ~T T ¼ ½0; 0T 21 T of
anti-windup scheme, and (3) the coordinates compensation the system (16) is globally uniformly stable. Furthermore,
algorithm of the target orientation angle of the car. the estimation error eo ðkÞ converges asymptotically to 0.

PPD parameter estimation algorithm Proof. Consider the Lyapunov function

First of all, the nonlinear system of automatic parking can T ðkÞ~
V1 ðkÞ ¼ Pe 0 2 ðkÞ þ ~ ðkÞ (17)
be transformed into the following dynamic linearization
data model where and P are positive constants and P is calculated by
P  F 2 P ¼ Q with Q is a positive constant. By taking into
ðk þ 1Þ ¼ ’1 ðkÞðkÞ þ ’ 2 ðkÞðkÞ (10) equation (16), we have
where  and  are defined as the orientation angle of the car V1 ðkÞ ¼ V1 ðk þ 1Þ  V1 ðkÞ
and the steering angle, respectively. Parameters ’1 and ’ 2 are ¼ PT ðkÞ~ T ðkÞðkÞ  2PFT ðkÞ~
ðkÞ~ ðkÞe 0 ðkÞ
called PPD. ðk þ 1Þ ¼ ðk þ 1Þ  ðkÞ and ðkÞ ¼
ðkÞ  ðk  1Þ. The model (10) can be redescribed as þ PF 2 eo2 ðkÞ þ ~T ðkÞð H T H  Þ~
ðkÞþ Pe 20 ðkÞ
¼ Qe 20 ðkÞ  T ðkÞ½ T ðkÞðkÞ  PðkÞ
ðk þ 1Þ ¼ ðkÞ þ T ðkÞðkÞ (11)
 2PFe 0 ðkÞðkÞ
where ðkÞ ¼ ½ðkÞ; ðkÞT and ðkÞ ¼ ½’ 1 ðkÞ; ’2 ðkÞT .
The proposed parameter identification observer has the  Qje 0 ðkÞj 2  ½ T ðkÞðkÞ  PjjðkÞjj 2
following structure: þ 2PFjje 0 ðkÞjjðkÞjj
^ðk þ 1Þ ¼ ^ðkÞ þ T ðkÞ^ (12)  c 1 je 0 ðkÞj 2  c 2 jjðkÞjj 2
ðkÞ þ Ke 0 ðkÞ
where e 0 ðkÞ ¼ ðkÞ  ^ðkÞ is the estimation error of out- where ðkÞ ¼ T ðkÞ~ ðkÞ, c 1 ¼ Q  ð1=
Þ, and c 2 ¼
put, ^ðkÞ ¼ ½^ ^ 2 ðkÞT , and we choose the gain K as
’ 1 ðkÞ; ’   2  P 
P2 F 2 . Hence, V1 ðkÞ  0 provided that
making F ¼ 1  K in the unit circle.
, Q, and satisfy the following inequalities
Therefore, by taking into account equations (11) and (12), 1
Q> ;   2  P 
P2 F 2 > 0
the estimation dynamic error of output can be given by

Yan et al. 5

Noticed that V1 ðkÞ is negative definite in the variables Observer tracking error is defined as eðkÞ ¼  ðkÞ
eo ðkÞ and ðkÞ. Since V 1 ðkÞ in a decreasing and nonnegative ^ðkÞ  ðkÞ, under the dynamic constraints (9), and com-
function, it converges to a constant value V1 1  0, as bining with (21), thus
k ! 1; hence, V 1 ðkÞ ! 0. This means that both eo ðkÞ and
~ðkÞ remain bounded for all the k, and limk!1 eo ðkÞ ¼ 0. eðk þ 1Þ ¼  ðk þ 1Þ  ^ðk þ 1Þ  ðk þ 1Þ
In order to make the parameter estimation, law (14) have ¼  ðk þ 1Þ  ^ðkÞ  ’
^ ðkÞðkÞ 1
a stronger capability in tracking time-varying parameter, a ’ ^ 2 ðkÞðkÞ  Ke 0 ðkÞ  ðkÞ
reset algorithm which13 should be considered is as follows
’ ^ 2 ðkÞð 0 ðkÞ  ðkÞÞ
½^ ^ 2 ðkÞT ¼ ½^
’1 ðkÞ; ’ ’1 ð1Þ; ’ ^ 2 ð1ÞT ¼  ðk þ 1Þ  ^ðkÞ  ðkÞ  ’

^ 1 ðkÞðkÞ
" #  
^ 1 ðkÞ
’  Ke 0 ðkÞ  ’ ^ 2 ðkÞ  0 ðkÞ  ðk þ 1Þ
if ½^
’1 ðkÞ; ’
^ 2 ðkÞ ’2 ðkÞÞ 6¼ sgnð^
 &; or sgnð^ ’2 ð1ÞÞ
^ 2 ðkÞ
’ (23)
(18) Substituting equation (22) into equation (23), we obtain
where & is a small positive constant and ’ ^ 1 ð1Þ and ’
^ 2 ð1Þ  
eðk þ 1Þ ¼ 2   ðk þ 1Þ  ^ðkÞ  ðkÞ
^ 1 ðkÞ and ’
are the initial value of ’ ^ 2 ðkÞ, respectively. ^ 2 ðkÞ þ 
’

’ ^ 1 ðkÞðkÞ  Ke 0 ðkÞ
A data-driven constrained control algorithm
(24)
and stability analysis
The convergence and tracking performance analysis for
On the basis of the observer (12), the data-driven uncon-
data-driven constrained control law (20) and (22) are
strained control algorithm is computed as
shown in theorem 2.
 0 ðkÞ ¼ ðk  1Þ
 
^ 2 ðkÞ  ðk þ 1Þ  ^ðkÞ  Ke0 ðkÞ  ’
’ ^ 1 ðkÞðkÞ Theorem 2. For given j ðkÞ   ðk  1Þj   , by using
þ the data-driven constrained control law (20) and (22), the
^ 22 ðkÞ þ 
’ solution of close-loop observer error system (24) is uni-
(19) formly ultimately bounded (UUB)27 for all the k with ulti-
mate bound limk!1 jeðkÞj  a2 =ð1  a1 Þ.
where  ðkÞ is the target orientation angle of the car.  is a
where  is a given positive constant
given small finite positive number. By taking into account

the input constraints (9), the adaptive constrained controller a1 ¼
is described as the following ^ 22 ðkÞ þ 
’
 
ðtÞ ¼ Sat ðk  1Þ þ Satfð 0 ðkÞ  ðk  1ÞÞ; a2 ¼ j þ ð1  ÞðkÞ  Ke 0 ðkÞ  ’
^ 1 ðkÞðkÞj
^ 22 ðkÞ þ 
’
 
T _ min ; T _ max ;  min ;  max (20)
where T denotes the sampling time, and Satð Þ function is
Proof. Considering the absolute value of equation (24), it
defined as
8 turns into
<b a  b
> 
Satða; b; cÞ ¼ a b < a < c jeðk þ 1Þj ¼ j ðk þ 1Þ  ^ðkÞ  ðkÞ
> ^ 22 ðkÞ
’ þ
:
c ac  Ke 0 ðkÞ  ’^ 1 ðkÞðkÞj
Because of the dynamic constraints in the close-loop 
¼ 2 j ðk þ 1Þ   ðkÞ þ eðkÞ
control system, an anti-windup compensator is designed ^ 2 ðkÞ þ 
’
to adapt to the target orientation angle of the car  ðkÞ. The þ ð1  ÞðkÞ  Ke 0 ðkÞ  ’ ^ 1 ðkÞðkÞj
compensation signal ðkÞ is designed as follows  
 2 jeðkÞj þ 2 j
ðk þ 1Þ ¼ ðkÞ þ ’
^ 2 ðkÞð 0 ðkÞ  ðkÞÞ (21) ^ 2 ðkÞ þ 
’ ^ 2 ðkÞ þ 
’
where lies in unit circle. Therefore, the controller (19) þ ð1  ÞðkÞ  Ke 0 ðkÞ  ’
^ 1 ðkÞðkÞj
can be redescribed as ¼ a1 jeðkÞj þ a2
^ ðkÞ
’ (25)
 0 ðkÞ ¼ ðk  1Þ þ 2 2
’^ 2 ðkÞ þ  Considering a Lyapunov function as V2 ðkÞ ¼ jeðkÞj,
 
  ðk  1Þ  ^ðkÞ  ðkÞ  Ke0 ðkÞ  ’
^ 1 ðkÞðkÞ from equation (25), one has
(22) V2 ðk þ 1Þ ¼ jeðk þ 1Þj  jeðkÞj ¼ ð1  a1 ÞV2 ðkÞ þ a2
6 International Journal of Advanced Robotic Systems

Since 0  a1 < 1 and a2 is bounded, based on the

lemma in,27 by using the constrained control law (20) and
(22), the result of close-loop observer system (24) is UUB
for all the k with ultimate bound limk!1 jeðkÞj 
a2 =ð1  a1 Þ.

Corollary. Based on the controller (20) and (22), together

with the observer (12) and adaptive laws (14), we can
ensure that the system tracking error eðkÞ ¼  ðkÞ  ðkÞ
is UUB with ultimate bound limk!1 j eðkÞj  a2 =ð1  a1 Þ.

Proof. Since
Figure 5. The compensation theory of the target orientation
eðkÞ ¼ eðkÞ  e 0 ðkÞ (26) angle of the car.
Considering the absolute value and limiting on both
sides of eqaution (26), we obtain to point F. The solid curve EF denotes the target trajectory
a2 and the dashed-dotted line EP(k) denotes the actual parking
lim j
eðkÞj  lim je 0 ðkÞj þ lim jeðkÞj  (27) trajectory. Point P(k) and point P ðk þ 1Þ are defined as the
k!1 k!1 k!1 1  a1
current location of the car and the target location of next
Hence the tracking error eðkÞ is UUB for all the k with moment, respectively.  ðk þ 1Þ and gðkÞ are defined as the
ultimate bound limk!1 jeðkÞj  a2 =ð1  a 1 Þ. target orientation angle of the car of next moment and the
The controller (22) can be divided into two parts: feed- orientation angle from point P ðk þ 1Þ to point PðkÞ,
back control and feedforward control, which is described as respectively.
From Figure 5, we know that the actual position of car
 0 ðkÞ ¼  01 ðkÞ þ  02 ðkÞ has separated from the target trajectory. Hence, we propose
where the compensation algorithm of the target orientation angle
Feedback control: of the car as the following
 
^ 2 ðkÞ
’ ~ðk þ 1Þ ¼  ðk þ 1Þ þ gðkÞ   ðk þ 1Þ (28)
 01 ðkÞ ¼ ðk  1Þ þ 2
^ 2 ðkÞ þ
’ 
 
 ^
  ðk þ 1Þ  ðkÞ  ðkÞ  Ke 0 ðkÞ where is a positive adjustable parameter, used to adjust
the intensity of compensation. ~ðk þ 1Þ denotes the com-
Feedforward control: pensated target orientation angle of the car. For the gðkÞ,
^ 2 ðkÞ^
’ ’1 ðkÞðkÞ we consider the following equation with
 02 ðkÞ ¼  2 0 1
^ 2 ðkÞ þ 
’ 
yðkÞ  y ðk þ 1Þ
Hence, we obtain the novel data-driven constrained con- gðkÞ ¼ arctan@ A (29)
xðkÞ  x ðk þ 1Þ
trol scheme of automatic parking system that consists of
eqautions (14), (18), (20), and (22). where ðxðkÞ; yðkÞÞ denotesthe current coordinate of the car
and x ðk þ 1Þ; y ðk þ 1Þ denotes the target coordinate
of the car of next moment.
The compensation algorithm of the target orientation Hence, we obtain the novel data-driven constrained con-
angle of the car trol algorithm with coordinates compensation as the
In the actual control process, there often exists tracking following
error between the actual orientation angle of the car and  
^ðk þ 1Þ ¼ ^ðkÞ þ ðkÞðkÞ e 0 ðk þ 1Þ  Fe 0 ðkÞ (30)
the target orientation angle of the car. Hence, there will
exist steady-state error between the actual parking trajec-
½^ ^ 2 ðkÞT ¼ ½^
’ 1 ðkÞ; ’ ’ ð1Þ; ’^ ð1ÞT
tory and the target trajectory. In order to eliminate the " 1 #2
steady-state error, we will propose a coordinates compen- ^ 1 ðkÞ
’
if ½^’ 1 ðkÞ; ’^ 2 ðkÞ   &; or sgnð^
’2 ðkÞÞ 6¼ sgnð^
’ 2 ð1ÞÞ
sation algorithm for automatic car parking systems. The ^ 2 ðkÞ
’
data-driven constrained controller corrects the steering (31)
angle according to the difference between the target orien-
tation angle of the car and the current orientation angle of 0 1

the car. yðkÞ  y ðk þ 1Þ
gðkÞ ¼ arctan@ A (32)
The compensation theory of the target orientation angle xðkÞ  x ðk þ 1Þ
of the car is shown in Figure 5. Backing a car from point E
Yan et al. 7

Figure 6. Flowchart of the novel data-driven constrained control system design procedure for automatic parking.

Table 1. Basic parameters of FAW-VW CC.  

~ðk þ 1Þ ¼  ðk þ 1Þ þ gðkÞ   ðk þ 1Þ (33)
Parameter Representation Value
Lc Vehicle length 4.799 m
 0 ðkÞ ¼ ðk  1Þ
Wc Vehicle width 1.855 m  
L Wheel base 2.712 m ^ 2 ðkÞ ~ðk þ 1Þ  ^ðkÞ  ðkÞ  Keo ðkÞ  ’
’ ^ 1 ðkÞðkÞ
Lp Length of parking space 5.600 m þ
^ 22 ðkÞ þ 
’
 max Maximal steering angle 42.00

S Safe distance 0.500 m (34)

50
θ*
PID algorithm
Orientation angle of the car θ (°)

40 proposed algorithm
42.5 43.5
42 43
41.5
42.5
30 41
40.5 42
41.5
3.2 3.4 3.6 3. 8
16.2 16.4 16.6
20
16.5
12.5
10 16
12
2.7 2.8 2.9
11.5
0
11
18.8 19 19.2
−10
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 7. The orientation angle of the car of FAW-VW CC automatic parking.

20
proposed algorithm.
PID
15
Steering angle β (°)

10

−5
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 8. The steering angle of FAW-VW CC automatic parking.

8 International Journal of Advanced Robotic Systems

0.1
proposed algorithm
PID
0.05

0
x−Axis error (m)

−0.05

−0.1

−0.15

−0.2

−0.25
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 9. Tracking error of x-axis.

0.6
proposed algorithm
PID
0.5

0.4
y−Axis error (m)

0.3

0.2

0.1

−0.1
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 10. Tracking error of y-axis.

4
proposed algorithm
PID
3

2
Error of θ (°)

−1

−2
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 11. Tracking error of orientation angle of the car.

 Simulation results
ðtÞ ¼ Sat ðk  1Þ þ Satfð 0 ðkÞ  ðk  1ÞÞ;
  In order to verify the superiority of the novel data-driven
_
T  min ; T _ max g ;  min ;  max (35)
constrained control scheme with coordinates compensa-
In order to give a clear idea of the comprehensive pro- tion, a simulation comparison among the proposed con-
posed novel data-driven constrained control system design strained control scheme with coordinates compensation
procedure for automatic parking, we provide a flowchart as and PID control scheme is given for different two cars with
shown in Figure 6. constant parking speed. PID controller adopts the position
control algorithm as the following
Yan et al. 9

−3
8 x 10
6

Compensation signal ρ(k)

4
2
0
−2
−4
−6
−8
−10
−12
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 12. Compensation signal.

1
PPD1

1
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 13. PPD1 parameter estimation. PPD: pseudo partial derivative.

0.05

0.05

0.05

0.05
PPD2

0.05

0.05

0.05

0.05
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 14. PPD2 parameter estimation. PPD: pseudo partial derivative.

Table 2. Basic parameters of Audi A6L.

X
k
Parameter Representation Value ðkÞ ¼ Kp eðkÞ þ Ki eðjÞ þ Kd ðeðkÞ  eðk  1ÞÞ (36)
j¼1
Lc Vehicle length 5.015 m
Wc Vehicle width 1.874 m where eðkÞ ¼  ðkÞ  ðkÞ denotes the error of the orien-
L Wheel base 3.012 m
Lp Length of parking space 5.600 m
tation angle of the car. Kp , Ki , and Kd denote the propor-
 max Maximal steering angle 42.00
tion parameter, integral parameter, and differential
S Safe distance 0.500 m parameter, respectively.
10 International Journal of Advanced Robotic Systems

50 *
θ
Orientation angle of the car θ (°) PID algorithm
40 44.5 proposed algorithm
41 44
16 40.5
43.5
30 40
15.5 43
3.2 3.4 3.6
2.7 2.8 2.9 16 16.2 16.4 16.6
20

13
10 12.8
12.6
12.4
0 18.9 19 19.1

−10
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 15. The orientation angle of the car of Audi A6L automatic parking.

20
proposed algorithm
PID
15
Steering angle β (°)

10

−5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (s)

Figure 16. The steering angle of Audi A6L automatic parking.

0.1
proposed algorithm.
0.05 PID

0
x−Axis error (m)

−0.05
−0.1
−0.15
−0.2
−0.25
−0.3
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (s)

Figure 17. Tracking error of x-axis.

Here, we choose the optimal PID parameters by tak- ¼ 0:01,  ¼ 10, K ¼ 0.6, ¼ 0.96,
ing into account the two car models, after repeated  min ¼ 42 deg,  max ¼ 42 deg, _ min ¼ 20 deg= s,
simulation test, and considering the principle of rapid- and _ max ¼ 20 deg= s.
ity and small overshoot. Parameter settings of the two The simulation consists of two parts: (1) automatic park-
algorithms are Kp ¼ 4:0, Ki ¼ 0:018, Kd ¼ 0:08, ing simulation of FAW Volkswagen CC (FAW-VW CC)
^ 1 ð1Þ ¼ 1, ’
’ ^ 2 ð1Þ = 0:05, & ¼ 104 ,  ¼ 0:003, and (2) automatic parking simulation of Audi A6L.
Yan et al. 11

0.6
proposed algorithm.
PID
0.5

0.4
y−Axis error (m)
0.3

0.2

0.1

−0.1
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (s)

Figure 18. Tracking error of y-axis.

5
proposed algorithm
PID
4

3
Error of θ (°)

−1

−2
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (s)

Figure 19. Tracking error of orientation angle of the car.

−4
× 10
0

−0.2
Compensation signal ρ(k)

−0.4

−0.6

−0.8

−1

−1.2
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Figure 20. Compensation signal.

Automatic parking simulation results of FAW-VW CC L

R1 ¼ (37)
The basic parameters of FAW-VW CC and parking space tanð max =1:1Þ
are shown in Table 1. We should try to avoid the car
steering wheel at the maximum rotation angle to protect The automatic parking simulation results of FAW-VW
the vehicle performance, therefore, the R1 is chosen as CC with constant parking speed (v ¼ 1.5 km/h) are shown
follows in Figures 7 to 14.
12 International Journal of Advanced Robotic Systems

1
PPD1

1
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (s)

Figure 21. PPD1 parameter estimation. PPD: pseudo partial derivative.

0.05
0.05
0.05
0.05
PPD2

0.05
0.05
0.05
0.05
0.05
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (s)

Figure 22. PPD2 parameter estimation. PPD: pseudo partial derivative.

From Figure 7, we observe that the two schemes can Automatic parking simulation results of Audi A6L
both achieve the tracking of the target orientation angle
The basic parameters of Audi A6L and parking space are
of the car in the automatic parking process. By the local
shown in Table 2. R1 is same as equation (37).
amplification as shown in Figure 7, it can be seen that the
The automatic parking simulation results of Audi A6L
tracking effect of the novel data-driven constrained control
with constant parking speed (v ¼ 1.5 km/h) are shown
scheme with coordinates compensation is obviously better
in Figures 15 to 22. From the simulation comparisons, we
than PID control scheme.
observe that the novel data-driven constrained control
From Figure 8, we observe that the steering angle of
scheme with coordinates compensation has smaller track-
the two schemes is consistent with the change of the
ing errors and faster response speed than PID scheme.
orientation angle of the car in the automatic parking
process. The front wheels of the two schemes are not
swinging back and forth, which makes the parking pro-
cess smooth. Conclusion
The parking error comparisons of FAW-VW are A novel data-driven constrained control scheme is proposed
shown in Figures 9 to 11. From these figures, we can for automatic car parking systems. The design of the pro-
see clearly that the tracking error of the novel data- posed scheme only depends on the steering angle and the
driven constrained control scheme with coordinates orientation angle of the car, and it does not involve any
compensation is smaller than PID control scheme. The model information of the car. A coordinates compensation
parking error comparisons verified the validity of the algorithm is also proposed, which reduces the tracking errors
coordinates compensation algorithm. The anti-windup of proposed scheme. A dynamic constraints unit with anti-
compensation signal ðkÞ is shown in Figure 12. In the windup compensator is designed to accommodate the refer-
whole tracking process, tuning PPD parameters are ence trajectory  ðkÞ. The proposed constrained control
shown in Figures 13 and 14. algorithm has the real-time implementation advantages of
Yan et al. 13

not requiring any iterative calculations. The simulation control on an FPGA-based car-like mobile robot. IEEE Trans
results have verified the superiority of the proposed scheme. Ind Electron 2003; 50(5): 867–880.
11. Baturone I, Moreno-Velo FJ, Sanchez-Solano S, et al. Auto-
Acknowledgements matic design of fuzzy controllers for car-like autonomous
The authors sincerely thank the editor and all the anonymous robots. IEEE Trans Fuzzy Syst 2004; 12(4): 447–465.
reviewers for their valuable comments and suggestions. 12. Heinen MR, Osorio FS, Heinen FJ, et al. SEVA3D: using
articial neural networks to autonomous vehicle parking con-
Declaration of conflicting interests trol. In: IEEE Conference on Neural Networks, Vancouver,
The author(s) declared no potential conflicts of interest with respect Canada, 16–21 July 2006, pp. 4704–4711.
to the research, authorship, and/or publication of this article. 13. Hou Z and Jin S. A novel data-driven control approach for a
class of discrete-time nonlinear systems. IEEE Trans Control
Funding Systems Technol 2011; 19(6): 1549–1558.
The author(s) disclosed receipt of the following financial support 14. Hou Z and Jin S. Data driven model-free adaptive control for
for the research, authorship, and/or publication of this article: This a class of MIMO nonlinear discrete-time systems. IEEE
work is supported by the National Natural Science Foundation of Trans Neural Netw 2011; 22(12): 2173–2188.
China (61503156, 61403161, and 51405198) and the Fundamen- 15. Hou Z and Wang Z. From model-based control to data-driven
tal Research Funds for the Central Universities (JUSRP11562 and control: survey, classification and perspective. Inf Sci 2013;
NJ20150011). 235: 3–35.
16. Xu JX and Hou ZS. Notes on data-driven system approaches,
References Acta Autom Sinica 2009; 35(6): 668–675.
1. Green P. Parking crashes and parking assistance system 17. Zhao Y, Lu C, and Han YD. MIMO model free adaptive wide
design: evidence from crash databases, the literature and area power system stabilizer design. J Tsinghua Univ 2013;
insurance agent interviews. In: 2006 SAE world congress, 53(4): 453–458,464.
Detroit, MI, 3–6 April 2006. 18. Xu D, Jiang B, and Shi P. A novel model free adaptive control
2. Sazawa M and Ohishi K. Fast continuous path tracking con- design for multivariable industrial processes. IEEE Trans Ind
trol considering high precision and torque saturation. In: Electron 2014; 61(11): 6391–6398.
IEEE Conference on Industrial Electronics, Porto, 3–5 19. Xu D, Jiang B, and Shi P. Adaptive observer based data-
November 2009, pp. 3089–3094. driven control for nonlinear discrete-time processes. IEEE
3. Cheng K, Zhang Y and Chen H. Planning and control for a Trans Autom Sci Eng 2014; 11(4): 1037–1045.
fully-automatic parallel parking assist system in narrow park- 20. Lv FL, Chen HB, Fan CJ, et al. A novel control algorithm for
ing spaces. In: IEEE Conference on Intelligent Vehicles Sym- weld pool control. Ind Robot 2010; 37(1); 89–96.
posium, Gold Coast, QLD, 23–26 June 2013, pp. 1440–1445. 21. Lu XP, Li W, and Lin YG. Load control of wind turbine based
4. Run RS, Yen JC and Chen C. A study of automatic parallel on model-free adaptive controller. Trans Chin Soc Agric
parking system-from the viewpoints of user and manufac- Machinery 2011; 42(2): 109–114,129.
turer. In: IEEE Conference on Circuits, Communications and 22. Chi RH and Hou Z. A model-free periodic adaptive control
System, Wuhan, China, 17–18 July 2011, pp. 1–5. for freeway traffic density via ramp metering. Acta Autom
5. Jung HG, Lee YH, and Kim J. Uniform user interface for Sinica 2010; 36(7): 1029–1032.
semiautomatic parking slot marking recognition. IEEE Trans 23. Gao B, Gu KY, Zeng Y, et al. An anti-suction control for an
Veh Technol 2010; 59(2): 616–626. intra-aorta pump using blood assistant index: a numerical
6. Kim D, Chung W, and Park S. Practical motion planning for simulation. Artif Organs 2012; 36(3): 275–285.
car-parking control in narrow environment. IET Control The- 24. Hou Z and Xu JX. On data-driven control theory: the state of
ory Appl 2010; 4(1): 129–139. the art and perspective. Acta Autom Sinica 2009; 35(6);
7. Liu K-Z and Horii M. An experimental comparison of non- 650–667.
holonomic control methods: automatic parking benchmark. 25. Liu KZ, Dao MQ, and Inoue T. An exponentially-convergent
In: IEEE Conference on Control Conference, Kos, 2–5 July control algorithm for chained systems and its application to
2007, pp. 2039–2044. automatic parking systems. IEEE Trans Control Syst Technol
8. Chen CH, Hsu CW, and Yao CC. A novel design for full 2006; 14(6): 1113–1126.
automatic parking system. Taipei, China, 2012. 26. Muller B, Deutscher J and Grodde S. Continuous curvature
9. Liang Z, Zheng G, and Li J. Application of fuzzy control trajectory design and feedforward control for parking a car.
strategy in automatic parking path planning. Nanchang, IEEE Trans Control Syst Technol 2007; 15(3): 541–553.
China, 2012. 27. Spooner J, Maggiore M, and Ordonez R. Stable adaptive
10. Li THS, Chang S-J, and Chen Y-X. Implementation of control and estimation for nonlinear systems. New York:
human-like driving skills by autonomous fuzzy behavior Wiley, 2002.