Mechanical Systems and Signal Processing 159 (2021) 107715

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

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

Heading tracking of 6WID/4WIS unmanned ground vehicles

with variable wheelbase based on model free adaptive control
Yue Jiang, Xiaojun Xu ⇑, Lei Zhang
College of Intelligent Science, National University of Defense Technology, Changsha, 410073, China

a r t i c l e i n f o a b s t r a c t

Article history: The six-wheel independent drive and four-wheel independent steering unmanned ground
Received 7 November 2020 vehicle (6WID/4WIS UGV) with variable wheelbase has unique advantages in environmen-
Received in revised form 6 January 2021 tal adaptability and maneuverability. However, due to its strong nonlinearity and uncer-
Accepted 27 January 2021
tainty, it has the great challenges in control. Based on model free adaptive control
Available online 17 March 2021
(MFAC) method and particle swarm optimization (PSO) algorithm, the heading tracking
problem of 6WID/4WIS UGV under the influence of uncertainties is studied in this paper.
Keywords:
First, to improve the convergence characteristics, the improved MFAC scheme is proposed,
Unmanned ground vehicle
Variable wheelbase
which can effectively overcome the influence of time delay and wheelbase. Then, based on
Heading control the desired steering angle and yaw moment obtained by MFAC method and the single-
Model free adaptive control track model of 6WID/4WIS UGV, the torque of six electric wheels is distributed by PSO
Particle swarm optimization algorithm to achieve the coordinated control of driving, braking and steering. Finally, the
Coordinated control feasibility and effectiveness of the proposed control strategy are verified by simulation
results.
Ó 2021 Published by Elsevier Ltd.

1. Introduction

Unmanned ground vehicles (UGVs) are widely used in military and civil areas. China’s complex terrain type puts forward
higher requirements for the structure and control performance of the UGV’s driving system. In-wheel motor is a hot topic.
Distributed drive system can simplify vehicle structure, realize complex control, and show great development potential in
improving vehicle mobility and maneuverability [1–5]. Therefore, to meet the operational requirements of crossing barriers
and ditches, six-wheel independent driving and four-wheel independent steering unmanned ground vehicle (6WID/4WIS
UGV) with variable wheelbase is proposed in [6]. Precise heading control is the precondition for UGV to complete the task.
But due to its strong nonlinearity, uncertainty of operation environment and difficulty in establishing an accurate mathemat-
ical model, it faces great challenges in heading control.
Path following system is an important part of UGV, which can minimize the lateral offset and heading error, and has been
the focus of attention [7–11]. On the basis of the current vehicles’ status and road information, path following is generally
completed through the lateral and longitudinal control of vehicles [11]. There are many control algorithms for path follow-
ing, including model predictive contr4ol [8,12–14], sliding mode control [15–16] and robust control [17–18]. With the devel-
opment of advanced control technology, researchers put forward many novel methods to improve the performance of
autonomous vehicles under uncertainty and external interference. In [19], a model predictive path-planning controller is

⇑ Corresponding author.
E-mail address: xuxiaojunmail@sina.com (X. Xu).

https://doi.org/10.1016/j.ymssp.2021.107715
0888-3270/Ó 2021 Published by Elsevier Ltd.
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

introduced that its objective includes potential functions along with the vehicle dynamics terms. To slove underactuated
problem with multi-objectives, Liu et al proposes an efficient model predictive control for path following and roll stabiliza-
tion of marine vessels [20]. Based on an exponential-like sliding-mode fuzzy type-2 neural network approach, Taghavifar et al
presents a novel robust adaptive indirect control method for enhanced path tracking performance of UGV [21]. The above
contributions show that the researchers gradually tend to consider how to ensure the stability of the UGV under the influ-
ence of uncertainties [22–23] in path following, and achieved satisfactory results. Besides, some research results are based on
accurate mathematical models (including unknown vehicle parameters and uncertainties) for controller design. Sometimes,
these methods are not universal. However, there are great challenges in the practical application of complex control algo-
rithms and structures [14]. There are few study on path following of distributed drive UGV with variable wheelbase. Mean-
while, it is rarely reported that the model free adaptive control (MFAC) method is used to track the desired path of multi axle
vehicles with variable wheelbase.
Tracking the desired path or heading on the premise of ensuring vehicle stability is of great significance to controller
design [12]. At present, PID has achieved good results in longitudinal motion control [9]. Thus, the lateral stability has been
widely concerned [7], such as active front wheel steering (AFS) [24] and direct yaw moment control (DYC) [25–26]. To
achieve the coordinated control of vehicle motion and hybrid powertrain, Shuai et al proposed a hierarchical control strategy
constructed by tracking controller and allocation algorithm [3]. In [4], a hierarchy control strategy is presented to coordinate
the braking and steering performance for in-wheel motor drive electric vehicles. In [27], a novel lateral stability controller
design for vehicle path tracking is developed to provide more stability for vehicle path tracking control. kim et al proposed an
optimal traction, braking and steering coordination to improve vehicle lateral stability and manoeuvrability of a 6WD/6WS
vehicle [28]. Although those control strategies have achieved satisfactory results, it can not be directly applied to 6WID/4WIS
UGV. On the one hand, they did not consider the influence on the performance of steering system and driving system when
the structural parameters such as wheelbase were changed; on the other hand, 6WID/4WIS UGV equips the electro-
hydraulic steering system (EHSS) instead of the AFS system, and it is difficult to control its lateral velocity. In addition, it
is important for 6WID/4WIS UGV to have abilities of strong flexibility, maneuverability and high-frequency heavy-duty
steering, which largely depends on the performance of the EHSS and power system. Thus, it is of great significance to develop
the coordinated control strategy to ensure the lateral stability of 6WID/4WIS UGV in path following. There are few studies on
the lateral stability in path following of multi axis UGV with variable wheelbase.
MFAC method only uses the input and output data of the controlled system, does not rely on the precise mathematical
model of the controlled system. It has the characteristics of overcoming time delay, strong adaptive and high robustness. If
the nonlinear system can satisfy certain assumptions, it can be applied to solve the control problem. MFAC method has
achieved good results in heading control of unmanned surface vehicle and other research areas [29–31]. Heading tracking
is a means of path following. By constructing the function of desired heading angle, the complex path tracking problem
can be simplified as a simple heading tracking problem [19]. Inspired by this, the heading tracking problem of 6WID/4WIS
UGV under the influence of uncertainties is studied in this paper. Firstly, the MFAC method is used to track the desired head-
ing and overcome the influence of uncertainties; secondly, to improve the convergence characteristics, the improved MFAC
scheme is proposed, which can effectively overcome the influence of time delay and variable wheelbase. Then, based on PSO
algorithm, the coordinated control of the power system and steering system in heading tracking is realized by the torque
distribution. Finally, the feasibility and effectiveness of the control strategy are validated by simulation results.
The main work of this paper is as follows. In Section 2, the structure of 6WID/4WIS UGV is introduced. The mathematical
models of multi axle vehicle, EHSS and in-wheel motor are established, which are used to generate input and output data and
serve as the basis of control strategy. In Section 3, the improved MFAC control scheme and control strategy based on PSO
algorithm are proposed. In section 4, the heading tracking performance under the influence of uncertainties is carried
out. The MFAC-PSO control strategy is validated by simulation. At last, Section 5 concludes the paper and outlines areas
for future research.

2. System modeling and analysis

In order to perform special tasks in special environment and meet the harsh power requirements of off-road, obstacle
crossing, mobility and silence, the powertrain of UGV gradually develops from traditional fuel power to hybrid power and
pure electric power. The 6WID/4WIS UGV adopts a series hybrid system to meet the requirements of high-speed mobility
and high-intensity driving. 6WID/4WIS UGV is equipped with an engine, generator, in-wheel motors, electric hydraulic
steering system, etc. The structure diagram is shown in Fig. 1. In terms of structure and operation environment, 6WID/4WIS
UGV has the following characteristics: (i) in order to improve the mobility and environmental adaptability, the wheelbase [6]
can be adjusted through the hydraulic system; (ii) it is greatly affected by the disturbance and uncertainties when perform-
ing tasks in the field. 6WID/4WIS UGV has a variety of operation modes to meet different requirements. At high speed, front
wheel steering is adopted to ensure the driving stability of the vehicle. On the condition of low speed, four wheel steering is
utilized to improve the steering sensitivity. The pure electric drive can be utilized to reduce noise. The braking modes include
regenerative braking, compound braking and hydraulic friction braking. It is very important for 6WID/4WIS UGV to operate
in a condition with low noise. Therefore, this paper mainly discusses how to realize the heading control of 6WID/4WIS UGV
through four wheel steering at low speed. The pure electric drive and regenerative braking can meet the demand.
2
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 1. Structure diagram of 6WID/4WIS UGV with variable wheelbase.

2.1. Vehicle model

The single-track linear model has been widely utilized to analyze and calculate the required longitudinal force and exter-
nal yaw moment [28]. The single-track model of 6WID/4WIS UGV is shown in Fig. 2. The longitudinal motion, lateral motion
and yaw motion are coupled and nonlinear. The longitudinal motion can be expressed as [3]
qC d Af dv x
Fx
v 2x
mgsinu
F f ¼m ð1Þ
2 dt

X
rr
Fx ¼ F xi ð2Þ
i¼fl

where Fx denotes the total longitudinal forces; q, Af and Cd denote the air density, vehicle frontal area and air resistance
coefficient, respectively; vx represents the longitudinal speed; m, g and u denote the mass, gravity acceleration and slope,
respectively; Ff is the total rolling resistance of all wheels; Fxi is the longitudinal force of each wheel, i = [fl,fr,ml,mr,rl,rr].
6WID/4WIS UGV needs to be satisfied with an Ackerman steering geometry relation between the front and the rear steer-
ing angles. Assuming that the front wheel angle is very small, the relationship between the front wheels angle and the rear
wheels angle is as follows

Fig. 2. The single-track model of 6WID/4WIS UGV.

3
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

lf þ lm
df ¼ ð3Þ
qs

lm
lr lm
lr
dr ¼ ¼ df ð4Þ
qs lf þ lm

lm
lr
dmin 6 dr ¼ df 6 dmax ð5Þ
lf þ lm
Where df and dr denote the front wheels angle and rear wheels angle respectively; lf,lm and lr denote the horizontal dis-
tances from the front axle, the central axle and the rear axle to the center of gravity of the vehicle. It is worth noting that
the value of lm can be adjusted through actively control; dmin and dmax are the minimum and maximum angles of the wheels,
which are limited by the steering structure.
6WID/4WIS UGV can directly control the torque of six electric wheels. External yaw torque is generated to track the
desired yaw rate, and ensure the vehicle stability. Assuming roll motion is negligible, according to Newton’s law

mv x ðb_ þ wÞ
_ ¼ F yfl þ F yfr þ F yml þ F ymr þ F yrl þ F yrr ð6Þ
 
€ ¼ lf F yfl þ F yfr
lm ðF yml þ F ymr Þ
lr ðF yrl þ F yrr Þ þ Mz
Iz w ð7Þ
where vx represents the longitudinal speed; w represents the yaw angle; b represents the side slip angle; Fyfl, Fyfr, Fyml, Fymr,
Fyrl and Fyrr denote the lateral forces of each wheel respectively; Iz is the inertia moment; Mz is the external yaw moment
generated by the longitudinal force, i.e
   
M z ¼ F xfl
df cosdf =2 þ lf sindf þ F xfr df cosdf =2 þ lf sindf
F xml dm =2 þ F xmr dm =2
ð8Þ
þF xrl ð
dr cosdr =2 þ lr sindr Þ þ F xrr ðdr cosdr =2 þ lr sindr Þ
where Fxfl, Fxfr, Fyxml, Fxmr, Fxrl and Fxrr denote the longitudinal forces of each wheel respectively; df, dm and dr denote the
track width of the front axle, the middle axle and the rear axle respectively.
When the heading angle reaches the desired value, the side slip angle should be zero. The heading angle can be described
as
h¼bþw ð9Þ
where h is heading angle.
The tire cornering stiffness of the front wheel, middle wheel and rear wheel is defined as Kf , Km and Kr respectively, and
the vehicle motion be expressed as follows.
2 2 3
2_3 2 3 ð
K f þ K r llr
l m
Þ 2 3
h h mv x f þlm 0
6 7
6 w_ 7 6w7 6 0 7 607
6 7 6 7 6 7 6 7
6 7 ¼ A6 7 þ 6 2 7df þ 6 7Mz ð10Þ
4 b_ 5 4 b 5 6 mv x ð
K f þ K r llfr
l
þlm 7
m
Þ 405
4 5
c_ c
ðK f lf þ K r lr
2 lr
lm
Þ
Iz
1
Iz
lf þlm

y ¼ ½ 1 0 0 0 x ð11Þ
2 2ðK f þK m þK r Þ 2ðK f lf
K m lm
K r lr Þ
3
0 0 mv x mv 2x
6 7
60 7
6 0 0 1 7
A¼6
60 2ðK f þK m þK r Þ 2ðK f lf
K m lm
K r lr Þ 7
7 ð12Þ
6 0 mv x mv x2
1 7
4 5
2ðK f lf
K m lm
K r lr Þ 2 2 2
2ðK f lf þK m lm þK r lr Þ
0 0 Iz Iz v x

According to the vehicle model, the steady-state yaw rate can be obtained [28,32]
A1 C 2
A2 C 1
css ¼ df ð13Þ
A2 B1
A1 B2
2ðK f þK m þK r Þ 2ðK f lf
K m lm
K r lr Þ
where A1 ¼ mv x
, B1 ¼ mv 2x

1, C 1 ¼ m2v x ð
K f þ K r llr
l m
Þ
f þlm

  2 2 2
2 K f lf
K m lm
K r lr 2ðK f lf þ K m lm þ K r lr Þ 2 lr
lm
A2 ¼ B2 ¼ C 2 ¼
ðK f lf þ K r lr Þ
Iz Iz v x Iz lf þ lm
The yaw rate is limited by road adhesion conditions [3–4,28,32]

4
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

X 
 
jcj 6  ui F zi =mv x ð14Þ

where ui is the friction coefficient of each wheel; F zi represent tire normal force of each wheel.
Therefore, the desired yaw rate can be expressed as
 
Xrr 
cdes ¼ minf jcss j ;  
ui F zi =mv x gsgnðdf Þ ð15Þ
 i¼fl 

where sgn(x) is a symbolic function.

2.2. Electro-hydraulic steering mechanism

The structure diagram of independent steering mechanism is shown in Fig. 3. As shown in Fig. 3(c), according to cosine
theorem, when the hydraulic cylinder is in the retraction limit position, Eq. (16) can be obtained.
2 2 2
lR þ lb
ljmin
umin ¼ arccos ð16Þ
2lh lb
When the hydraulic cylinder is at the extension limit position, Eq. (17) can be described as
2 2 2
lR þ lb
ljmax
umax ¼ arccos ð17Þ
2lh lb
When the hydraulic cylinder is in any position, Eq. (18) can be obtained.
2 2 2
lh þ lb
lj
uj ¼ arccos ð18Þ
2lh lb
where lj ðj ¼ fl; fr; rl; rrÞrepresents the length of the hydraulic cylinder in working condition.
The steering angle can be described as
(
lf þ lm
ðuj
umin Þ; uj < umax þ2 umin
df ¼ dr ¼ ð19Þ
lm
lr uj
umax þ2 umin ; uj > umax þ2 umin
The schematic diagram of the steering system is shown in Fig. 3(b). The pressure of return oil is considered to be zero, the
flow through the electro-hydraulic proportional valve is
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
< qj1 ¼ C q jj1 xv j q2 ½ð1 þ sgnðxv j ÞÞpp =2
sgnðxv j Þpj1 
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð20Þ
>
: qj2 ¼ C q jj2 xv j 2 ½ð1
sgnðxv j ÞÞpp =2 þ sgnðxv j Þpj2 
q

Fig. 3. Schematic diagram of steering mechanism.

5
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

where C q is the flow coefficient; jj1 and jj2 ðj ¼ fl; fr; rl; rrÞ are the area gradient of the corresponding proportional valve; xvj
is the displacement of the proportional valve spool, which is assumed to be positive when the rod free cavity is extended; pp
is the outlet pressure of the hydraulic pump; pj1 and pj2 are the pressure of the rod chamber and the rodless chamber respec-
tively; q is the oil density.
The movement of the hydraulic cylinder can be expressed as
(
x_ j1 ¼ qj1 gcv =Aj1
ð21Þ
x_ j2 ¼ qj2 gcv =Aj2

where Aj1 and Aj2 represent the effective area of rod chamber and rodless chamber respectively; xj1 and xj2 represent the
displacement of extension and retraction of hydraulic cylinder; gcv represents volumetric efficiency of hydraulic cylinder.
The continuous equation of hydraulic cylinder flow is [33]
(
qj1 ¼ Aj1 x_ j þ C t pj1 þ p_ j1 V j1 =be
ð22Þ
qj2 ¼ Aj2 x_ j þ C t pj2 þ p_ j2 V j2 =be

Where V j1 and V j2 represent the effective initial volume of rod chamber and rodless chamber respectively; Ct represents
leakage coefficient.
When the rodless chamber is in the extended state, keep the displacement of the hydraulic pump constant, and define
l ¼ xv j , x ¼ ½x1 ; x2 ; x3  ¼ ½xj ; x_ j ; pj1 T , y ¼ xj , the state equation of the EHSS is obtained as follows.
8
>
> x_ 1 ¼ x2
>
< qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
C j A
x_ 2 ¼ q gj1 j1
q ðpp
x3 Þu
2
ð23Þ
cv
>
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
: x_ 3 ¼
be Aj1 x2
be C t x3 þ C q jj1 Aj1 2 ðpp
x3 Þu
V j1 V j1 V j1 q

Eq. (23) indicates that the EHSS has strong nonlinearity, and the controller design based on the mathematical model may
make the control unstable. Therefore, only the input and output of the steering control are determined by using the above
formulas, which lays the foundation for the subsequent controller design.

2.3. In-wheel motor model

Without considering the hydraulic braking, the motion of in-wheel motor can be expressed as [3–4]

dxxi
F xi ¼ ðT ei
J Þ=r i ð24Þ
dt

T emin 6 T ei 6 T emax ð25Þ

where Tei is the output torque of the in-wheel motor; J is the moment of inertia of the wheel; ri is the wheel radius; Temax
and Temin are the maximum and minimum torque of the in-wheel motors, respectively.
The power of the in-wheel motors is one of the bases to study the energy consumption and torque distribution, it can be
expressed as

Pei ¼ T ei xxi ge
sgnðT ei Þ ð26Þ

where Pei is the power and ge is the efficiency of in-wheel motors.

3. Controller design

The wheelbase of 6WID/4WIS UGV is uncertain which will be adjusted by the hydraulic system according to the actual
operating conditions, such as obstacle crossing. Thus, there are unmodeled dynamics and other uncertainties in the modeling
process. When the accurate mathematical model of the vehicle cannot be obtained, the data-driven control method only uses
the process data to design the controller [34], which provides a new idea for heading control of 6WID/4WIS UGV. This section
will design the heading controller of 6WID/4WIS UGV based on the MFAC method and the single-track model. The architec-
ture of heading control is presented in Fig. 4.
According to the desired heading and longitudinal speed generated by upper controller, the desired steering angle, the
yaw moment and the total longitudinal force are obtained by the MFAC controller, PID controller and fuzzy logic controller,
respectively. The calculated longitudinal force is distributed to six electric wheels by PSO algorithm. By constructing the
desired yaw angle function, the complex path tracking problem can be simplified to a simple yaw angle tracking problem.
When the actual yaw angle is close to the desired yaw angle, the lateral offset of UGV converges to zero [19].
Based on MFAC control theory, only the input and output data are required for heading tracking of the 6WID/4WIS UGV
with variable wheelbase. So, the design idea of the controller has certain generality. The wheelbase, yaw angle and steering
6
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 4. The architecture of heading control.

angle required in the lower controller can be directly measured or indirectly calculated by GPS/INS and vehicular sensors.
Compared with the model-based control algorithm, precise modeling is not required. For the time-varying and nonlinear
state parameters, such as tire force, slip angle, etc., it is difficult to obtain. The structure of the controller may be increased
if the state observer is established to get these parameters. Although MFAC method needs to adjust the key parameters
through multiple adjustments to obtain good control performance, but its control parameters have more strong robustness.
Therefore, the proposed heading tracking method is feasible in practical application.

3.1. MFAC controller design

The MFAC method uses a new dynamic linearization method and a new concept called pseudo partial derivative (PPD) for
discrete-time nonlinear systems. An equivalent dynamic linearization data model is established at each dynamic working
point of the closed-loop system. Based on the virtual equivalent data model, the controller is designed and the control sys-
tem is analyzed to realize the adaptive control of the nonlinear system.

3.1.1. Compact-form dynamic linearization based on MFAC scheme

The following assumptions are made for single input and single output (SISO) discrete-time nonlinear systems [29]

Assumption 1.. Except for finite time points, the partial derivatives of f ð  Þ with respect to the (ny + 2)th variable is continuous.

Assumption 2.. Nonlinear system satisfies the generalized Lipschitz condition, that is
jyðk1 þ 1Þ
yðk2 þ 1Þj; bk lðk1 Þ
lðk2 Þ k ð27Þ
where nonlinear system yðki þ 1Þ ¼ f ðyðki Þ;   ; yðki
ny Þ; lðki Þ;   ; lðki
nu ÞÞ, i = 1,2; b is a positive constant.
Assumption 1 is a typical constraint on general nonlinear systems in control system design. According to [26], theorem 1
can be obtained:

Theorem 1.. If a SISO discrete-time nonlinear system satisfies assumptions 1 and 2, when k DlðkÞ k–0, there must exist a time-
varying parameter /c ðkÞ 2 Rm named the PPD that causes the system to transform into the compact-form dynamic linearization
(CFDL) model of Eq. (28):

yðk þ 1Þ ¼ yðkÞ þ /Tc ðkÞDlðkÞ ð28Þ

where /c ðkÞ is bounded at any time, and DlðkÞ ¼ lðkÞ
lðk
1Þ, Dyðk þ 1Þ ¼ yðk þ 1Þ
yðkÞ.
Compact-form dynamic linearization model free adaptive control (CFDL-MFAC) of SISO discrete-time nonlinear systems is
as follows
^
/ ^ ðk
1Þ þ gDlðk
1ÞðDyðkÞ
/c ðk
1ÞDlðk
1ÞÞ
^ ðkÞ ¼ / ð29Þ
c c 2
u þ k Dlðk
1Þ k
 
^  ^ ðkÞÞ–signð/
^ ð1ÞÞ/
^ ðkÞ ¼ /
^ ð1Þ
/c ðkÞ < ejDlðk
1Þj 6 esignð/c c c c ð30Þ

q/^ c ðkÞðy ðk þ 1Þ
yðkÞÞ

lðkÞ ¼ lðk
1Þ þ ð31Þ
^ ðkÞ k2
kþk/ c

7
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

where / ^ ðkÞis the estimation of the PPD / ðkÞ;/

^ ð1Þis the initial value of the PPD, i ¼ 1;    ; m;k > 0and l > 0are weight
c c i
coefficients to adjust the change rates of the PPD and controller output, respectively; g 2 ð0; 2 and q 2 ð0; 1are the step
factors.
To ensure the stability of the closed-loop system, the controlled system must satisfy the following assumptions [26], and
theorem 2 is obtained:

Assumption 3.. For a given bounded expected output signal y ðk þ 1Þ, a bounded control input signal l ðkÞ that causes the system
output be equal to y ðk þ 1Þ always exists.

Assumption 4.. For any moment k and DlðkÞ–0, the PPD sign of the system remains unchanged.

Theorem 2.. Under assumptions 1–4 , the CFDL-MFAC scheme can be utilized. When y  ðk þ 1Þ ¼ const, a positive number
kmin > 0 exists; when k P kmin , the following conclusions can be obtained.

(i) the system output tracking error is monotonically convergent;

(ii) the closed-loop system is BIBO stable, i.e., the output sequence fyðkÞg and input sequence flðkÞg are bounded.

Assumption 3 is a necessary condition that system is output controllable. Assumption 4 is intended to show that the out-
put of the corresponding controlled system should not decrease when the control input is increased, which can be consid-
ered as a ‘‘quasi linear” characteristic of the system.

3.1.2. Application of MFAC in 6WID/4WIS UGV heading control

When the front wheel angles of 6WID/4WIS UGV increases within the allowable range, the heading angle also increases
accordingly. It can be seen from formula (16) to formula (19) and Fig. 3(c) that the front wheel angle monotonically increases
with the stroke of the hydraulic cylinder in EHSS. Meanwhile, when the spool displacement of proportional valve (propor-
tional to the input current) increases, the stroke of hydraulic cylinder will not decrease. Therefore, 6WID/4WIS UGV satisfies
the characteristics of theorem 2. However, the key problem is that it is very difficult to make the front and rear wheels angle
satisfy Eq. (4) only by controlling the spool displacement of proportional valve (through input current), because there is
strong nonlinearity between input current of proportional valve and front wheels angle. Therefore, the MFAC method is pro-
posed to calculate the desired steering angle and track desired heading, and then PID control is used to track the desired
angle. Architecture of MFAC controller is illustrated in Fig. 5. hðkÞdenotes the actual heading angle at time k; h ðk þ 1Þis
the desired heading at time (k + 1); dfdes ðkÞ represents the desired steering angle of the front wheel at time k; df ðkÞand
df ðk
1Þ denote the actual steering angle of front wheel at time k and k-1, respectively; dr ðkÞand dr ðk
1Þare the actual steer-
ing angle of rear wheel at time k and k-1.
Because of the limitation of the mechanical structure, the steering angle is bounded. In this paper, the output of the con-
troller is always bounded. This is a data-driven control problem for a class of nonlinear systems with output saturation [34].
If the initial tracking error is too large, the output of the controller may converge slowly, resulting in excessive steering and
low accuracy of heading control. In order to improve the performance of heading tracking and make the original control algo-
rithm more general, the coefficient 0＜ j1  1 is introduced on the basis of the original control scheme. Meanwhile, the yaw

Fig. 5. Architecture of MFAC controller.

8
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

rate is added to the control law and 0  j2, which makes the controller design more general and prevents excessive change
of steering angle and ensure vehicle stability. Eq. (32) is obtained as follows.

q/^ c ðkÞðy ðk þ 1Þ
yðkÞÞ

lðkÞ ¼ j1 lðk
1Þ
j2 c þ ð32Þ
^ ðkÞ k
kþk/
2
c

According to Eq. (32), it is obtained as follows.

^ - 1Þðy ðkÞ
yðk
1ÞÞ ^ 
c ðk c ðkÞðy ðk þ 1Þ
yðkÞÞ
lðkÞ ¼ j21 lðk
2Þ
j1 j2 c þ j1 q/ ^ 2
j2 c þ q/ ^ 2
kþk/ c ðk - 1Þk kþk/ c ðkÞk
^
ð33Þ
c ð1Þðy ð2Þ
yð1ÞÞ
 q/^ c ðkÞðy ðk þ 1Þ
yðkÞÞ
¼ jk1 - 1 lð1Þ
jk
2
1 j2 c
  
j2 c þ jk1 - 2 q/ kþk ^ ð1Þk2
þ  þ ^ 2
/ c kþk/ c ðkÞk

When the desired heading is achieved, the yaw rate should be zero. When 0＜j1＜1, Eq. (33) shows that the influence of
the the previous output on the system gradually approaches to 0, and the convergence characteristic of the system is
improved.
Stability analysis is given as follows. Define the tracking error of the system:
eðk þ 1Þ ¼ y 
yðk þ 1Þ ð34Þ
According to the theorem 1:
jeðk þ 1Þj ¼ jy 
yðkÞ
/c ðkÞDuðkÞj
q/c ðkÞ/^ c ðkÞ ð35Þ
6 jð1 - kþj/
^ ðkÞj ÞeðkÞ - /c ðkÞðj1
1Þuðk
1Þ þ /c ðkÞj2 cj
c

^ ðkÞ P 0from assumption 4 and Eq. (29). According to the literature [29], /
It can be obtained that /c ðkÞ/ ^ ðkÞ is bounded. It
c c



is known from theorem 1 that /c ðkÞ is bounded. j/c ðkÞj 6 b and b is a constant. Without loss of generality, only the situation


of /c ðkÞ > e > 0 is discussed. When choosing k > kmin ¼ b2 =4, then there must exist a constant 0 < M 1 < 1 that enables for-
mula (35) to hold.




q/c ðkÞ/^ c ðkÞ b/^ ðkÞ b/^ ðkÞ b
0 < M1 6  2 6 
c
2 6 ^
c
pffiffiffi < pffiffiffiffiffiffiffiffiffi 1 ð36Þ
^  ^  2 / ðkÞ k 2 kmin
k þ /c ðkÞ k þ /c ðkÞ c

According to formula (35), and k > kmin , there must be a constant d2 < 1 exists that enables formula (37) to hold.

q/c ðkÞ/^ ðkÞ

0<1-  c 
^  6 1 - qM1 ¼ d2 < 1 ð37Þ
k þ / c ðkÞ

It is obtained that jeðk þ 1Þj 6 jd2 eðkÞ - /c ðkÞðj
1Þuðk
1Þ þ /c ðkÞj2 cj. Because j/c ðkÞj 6 b,dmin 6 uðk - 1Þ 6 dmax and
care bounded, there must be a constant 0 ＜ j1  1 and 0  j2 that enables
jeðk þ 1Þj 6 jd2 eðkÞ - /c ðkÞðj1
1Þuðk
1Þ þ /c ðkÞj2 cj
6 d2 jeðkÞj þ j/c ðkÞð1
j1 Þuðk
1Þj þ j/c ðkÞj2 cj



ð38Þ
6 d2 jeðkÞj þ ð1
j1 Þ c þj2 d


k ð1
j1 Þ
j2 d
6    6 d2 jeð1Þj þ 1
d2
c þ 1
d 2

where c and dare constants. y* is a constant and the convergence of the tracking error means that the output y(k) of the
system is bounded.
The output and desired output of 6WID/4WIS UGV are
yðkÞ ¼ hðkÞ ð39Þ

y ðk þ 1Þ ¼ h ðk þ 1Þ ð40Þ

3.2. Yaw rate and vehicle speed controller design

How to calculate the desired yaw moment is of great significance to the ensure lateral stability of vehicles. The yaw
moment control system is a nonlinear system [35]. Fuzzy controller does not rely on the mathematical model, but based
on the fuzzy rules, which is very suitable for yaw moment controller. Fuzzy controller has achieved good results in the cal-
culation of desired yaw moment [36], and provides a good reference. Therefore, when the accurate mathematical model of
6WID/4WIS UGV cannot be obtained, fuzzy logic control is utilized to estimate the desired external yaw moment, while the
desired total longitudinal force is calculated according to the desired longitudinal velocity. The block diagram of yaw rate
9
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

and longitudinal speed controller is shown in Fig. 6.cdes ðkÞis the desired yaw rate at time k; ca ðk
1Þdenotes the actual yaw
rate at time k-1; ec ¼ cdes ðkÞ
ca ðk
1Þrepresents the yaw rate error; v xdes ðkÞand v x ðk
1Þare the desired longitudinal veloc-
ity at time k and the actual longitudinal velocity at time k-1; M zdes ðkÞand F xdes ðkÞdenote the desired external yaw moment and
total longitudinal force at time k respectively.

3.3. Torque distribution based on PSO algorithm

Torque distribution of electric wheels is the key to lateral stability of 6WID/4WIS UGV. PSO algorithm has the advantages
of convenient implementation, strong global search ability and fast convergence, which is appropriate for optimization prob-
lems in complex engineering applications [37–38]. In this paper, PSO is used to optimize the torque distribution of six elec-
tric wheels.

(1) Optimization objective and constraints

The following optimization objectives are considered in this paper. The first optimization objective is

J 1 ¼ k BlPSO
yPSO k2 ð41Þ

Where lPSO ¼ ½M zdes ; F xdes T ,

 
1 1 1 1 1 1
B¼ . Wheelbase and other

df cosdf þ lf sindf df cosdf þ lf sindf
dm dm
dr cosdr þ lr sindr dr cosdr þ lr sindr
parameters can be collected and fed back by vehicular sensors.
In order to improve the driving distance, energy consumption and recovery should be taken into acount. It should be
noted that the driving torque and regenerative braking torque are in the opposite direction, so the second optimization
objective can be uniformly expressed as
X
J2 ¼ P ei ð42Þ

To prevent the output torque of the six electric wheels from excessive, the third optimization objective is defined as
X
J3 ¼ F 2xi ð43Þ

Combined with the optimal distribution constraints and the same characteristics of the six in-wheel motors, the compre-
hensive optimization model can be obtained as follows.
8
> min W 1 J 1 ðyPSO Þ þ W 2 J 2 ðyPSO Þ þ W 3 J 3 ðyPSO Þ
<  T
where yPSO ¼ F xfl ; F xfr ; F xml ; F xmr ; F xrl ; F xrr ð44Þ
>
:
s:t: T emin
r
6 F xi 6 T emax
r
; v x 6 v xmax ; lmmin 6 lm 6 lmmax ; dimin 6 di 6 dimax
Where W1, W2 and W3 are weight coefficients.

(2) Torque distribution

Step 1: The particle swarm is randomly initialized. The initial population is N. The random position of the m-th particle
can be expressed as
T emin T emax
T emin
yPSO ð : ; : ; mÞ ¼ þ randð : ; : ; mÞ; 0 < m 6 N ð45Þ
r r
Where rand(: , : ,m)is a random matrix uniformly distributed in the (0,1) interval.

Fig. 6. Block diagram of yaw rate and longitudinal speed controller.

10
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

The velocity of m-th particle is expressed as

v PSO ð : ; : ; mÞ ¼ v min þ ðv max
v min Þrandð : ; : ; mÞ ð46Þ
Step 2: Calculate the fitness of each particle. According to the fitness function in the comprehensive optimization model,
the fitness of m-th particle is calculated.
J fitness ðyPSO ð : ; : ; mÞÞ ¼ W 1 J 1 ðyPSO ð : ; : ; mÞÞ þ W 2 J 2 ðyPSO ð : ; : ; mÞÞ þ W 3 J 3 ðyPSO ð : ; : ; mÞÞ ð47Þ

Step 3: Update the historical optimal position of individual particles. If the fitness of the m-th particle in the c-th calcu-
lation satisfies
J fitness;c ðyPSO;c ð : ; : ; mÞÞ < PbestJ fitness ðyPSO ð : ; : ; mÞÞ ð48Þ

Where PbestJfitness ðyPSO ð : ; : ; mÞÞdenotes the best fitness of the mth particle in the c-th calculation, the historical optimal
fitness and position of the individual particle are updated as follows.
PbestJ fitness ðyPSO ð : ; : ; mÞÞ ¼ J fitness;c ðyPSO ð : ; : ; mÞÞ ð49Þ

PbestyPSO ¼ yPSO;c ð : ; : ; mÞ ð50Þ

Step 4: Update the historical optimal position of all particles. If the fitness of the m-th particle in the c-th calculation
satisfies
minfJ fitness;c ðyPSO ð : ; : ; 1ÞÞ;    ; J fitness;c ðyPSO ð : ; : ; mÞÞ;    ; J fitness;c ðyPSO ð : ; : ; NÞÞg ¼ J fitness;c ðyPSO ð : ; : ; mÞÞ ð51Þ

then
GbestJfitness ðyPSO Þ ¼ J fitness;c ðyPSO ð : ; : ; mÞÞ ð52Þ

GbestyPSO ¼ yPSO ð : ; : ; mÞ ð53Þ

where GbestJfitness ðyPSO Þand GbestyPSO represent the best fitness and the best position in the first c iterations of the particle
swarm.
Step 5: Update the velocity and position of the particles. The position of the next generation particle swarm is determined
by the current position and velocity. The velocity of the (c + 1)-th generation particle can be expressed as
v PSO;cþ1 ð : ; : ; mÞ ¼ W 1  v PSO;c ð : ; : ; mÞ þ c1  randð1Þ  ðPbestyPSO
yPSO;c ð : ; : ; mÞÞ
ð54Þ
þc2  randð1Þ  ðGbestyPSO
yPSO;c ð : ; : ; mÞÞ

Meanwhile, the velocity and position of the particle m should satisfy the constraint conditions.
v min 6 v PSO;cþ1 ð : ; : ; mÞ 6 v max ð55Þ

ymin 6 yPSO;cþ1 ð : ; : ; mÞ 6 ymax ð56Þ

Step 6: If the termination conditions are not met, go to step 2. If the number of the optimization iteration reaches the pre-
defined threshold, the rolling optimization of PSO algorithm is stopped and the optimization results are output to six in-
wheel motor controllers.
yPSO ¼ GbestyPSO ð57Þ
According to step1 ~ step 6, the optimal distribution flow chart based on PSO algorithm is obtained as shown in Fig. 7.

4. Simulation results and analysis

In this section, the controller and vehicle model are established in Matlab/Simulink and Trucksim, respectively. Some
related researches have achieved satisfactory results by using PID to control the EHSS, and the error is controlled within
2% [39–40].

4.1. The performance of heading control

In order to verify that the j-MFAC method has the characteristics of overcoming time delay and better robustness for
heading control, the following situations are discussed and analyzed. Due to the limitation of the steering mechanism,
the steering angle output by the controller is limited to [-20 deg, 20 deg]. The structural parameters of 6WID/4WIS UGV
are shown in Table 1.
Case1: Compared with other controllers
The performance of heading tracking of different controllers is simulated and analyzed, and several control methods are
selected for comparative analysis. During the simulation process, the vehicle speed is kept at 35 km/h, and the sampling time
11
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 7. Optimal distribution flow chart of torque based on PSO algorithm.

is 0.001 s. The execution delay of electro-hydraulic steering mechanism is also considered. According to the theoretical anal-
ysis in Section 2.1, when the wheelbase changes, the relationship between the front and rear wheels angles will also change
to meet the formula (4). Since the purpose of this paper is not to rely on the accurate mathematical model for controller
design, the classical PID control method and the original control scheme of MFAC are selected for comparison. To ensure
the fairness of the comparison, the optimization parameters are found through multiple adjustments. The control parame-
ters of the control methods are as follows. j-MFAC:j1 = 0.8, j2 = 1.5, q = 1 l = 10, g = 1, k = 0.01, e = 0.1; MFAC:q = 1, l = 1,
g = 1, k = 0.2, e = 10^-5; PID: Kp = 2, KI = 0.01. KD = 0.001. The friction coefficient is 0.8 and wheelbase is 1.1 m. The desired
heading refers to the double line change condition in the literature [15]. The simulation results are shown in Fig. 8.
Fig. 8 (a) and (b) show the change of heading angle and the steering angle respectively. It can be seen that the heading
tracking performance of j-MFAC is better than PID and the original scheme of MFAC. The root mean square errors are shown
in Table 2. Due to the limitation of the steering mechanism, the outputs of the three controllers are in a limited range, and the
output of the j-MFAC controller is more stable than the other two controllers. The control law of PID is mainly affected by
the current error, but the vehicle speed has a significant impact on the change of heading angle, and the electro-hydraulic
steering mechanism has time delay, so the control performance is poor. The control law of MFAC essentially contains only
one variable integral, which leads to poor convergence characteristics and unsatisfactory control performance. In this paper,
the improved j-MFAC scheme adds the coefficient j1 and the differential j2c to the control law, which effectively improves
the convergence characteristics and suppresses the excessive change of the steering angle. Therefore, the control perfor-
mance is the best. In order to ensure the stability of the steering process, particle swarm optimization is used in the torque
distribution. Based on the above analysis, the j - MFAC controller has the best control performance under the premise of
ensuring vehicle stability.
Case2: Robustness to the wheelbase
The performance of heading tracking under different wheelbase is simulated and analyzed, and original control scheme of
MFAC is selected for comparative analysis. The simulation results are shown in Figs. 9 and 10. The control parameters of the
control methods are as follows. j-MFAC:j1 = 0.4, j2 = 0, q = 0.2, l = 10, g = 1, k = 15, e = 0.1; MFAC:q = 0.01, l = 100, g = 1,
k = 200, e = 0.1. The friction coefficient is 0.8.
As shown in Figs. 9 and 10, under ideal conditions, the j-MFAC method achieves good results, but the effect of the original
MFAC scheme is not very well. Combining with the principle of the control method, the causes of the results are analyzed. In
the case of no interference, according to Eqs. (29–31), it can be seen that the output of MFAC method is not only related to

Table 1
The main parameters of 6WID/4WIS UGV.

Parameter Value
Mass (kg) 2500
Horizontal distance between center of gravity and front axle (m) 1.6
Horizontal distance between front axle and middle axle (m) 1.1 ~ 2.1
Horizontal distance between rear axle and middle axle (m) 1.1 ~ 2.1
Height of center of gravity (m) 1.25
Tire diameter (m) 0.996
Tire width (m) 0.309
Maximum off-road speed (km/h) 35
Power of in-wheel motor (kW) 65

12
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 8. The comparison of heading tracking performance of different controllers.

Table 2
Root mean square error of heading angle.

Control method j-MFAC MFAC PID

Heading tracking error (deg) 1.2 10.6 5.2

Fig. 9. The heading control performance of MFAC.

the error of the current time, but also affected by the previous tracking error. In order to reduce this influence, we can only
greatly improve the key control parameter k in the control scheme, and reduce the step factor q. But the influence of the
previous error always exists, and the control output convergence speed becomes slow. Therefore, the actual heading angle
needs a long enough time to converge to the desired value. It should be noted that the forward movement of the middle axle
will lead to higher steering sensitivity and shorter adjustment time. However, the forward movement of the middle axle
makes the stability in the steering process relatively weak [6]. For the original MFAC scheme, due to poor convergence char-
acteristic, the control performance of the same group of control parameters is better when the middle axle is behind the cen-
ter of mass, but the vehicle will lose stability when it is in front of the center of mass, and the robustness of the control
parameters is poor. In this paper, a coefficient of 0 ＜ j1  1 is added to the original scheme. When 0＜ j1 ＜1 is selected,
the influence of the previous error in the j-MFAC control scheme will be gradually eliminated, and the proportion of the
error close to the current time is increasing. Therefore, the output of the controller will converge to zero together with
the tracking error. The simulation results also prove the correctness and effectiveness of the improved scheme, which has
good robustness to the change of vehicle wheelbase.
Case 3: Robustness to the time delay in the execution of EHSS

13
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 10. The heading control performance of j-MFAC.

In the actual control process, there is time delay when the EHSS executes the control command, and there is also a
response time of the hydraulic cylinder. So, the simulation analysis is conducted under the assumption that the EHSS exhibit
a 1 s time delay. The simulation results are illustrated in Fig. 11. The friction coefficient is 0.8. The control parameters of j-
MFAC are consistent with Case 2. MFAC:q = 0.1, l = 100, g = 1, e = 10-5.
Fig. 11(a) is the simulation results of the improved scheme considering a 1 s time delay under different wheelbase; Fig. 11
(b) is the simulation results of the original MFAC scheme under different key control parameters when the wheelbase is
1.1 m. As shown in Fig. 11(b), the system can not achieve ideal control, and the time delay has a very important impact
on its control performance. The control parameters of j-MFAC have certain robustness, but the output of MFAC control
scheme is also related to the tracking error before the system, which will gradually increase with time accumulation. Even
if the key control parameters are reduced, the adjustment time will become slow and it is difficult to obtain satisfactory
results. The influence of different degrees of time delay on the improved MFAC scheme is shown in Fig. 12.
Fig. 12 shows that the control parameters of the improved MFAC scheme have strong robustness to a certain extent. How-
ever, with the increase of time delay, the robustness gradually weakens. In the future research, we can improve the param-
eter adaptation to solve this problem.
Case 4: Robustness to adhesion condition
In order to further discuss the influence of uncertainties on the controller, different friction coefficients are selected for
simulation analysis. The wheelbase is 1.6 m. The simulation results are shown in Fig. 13. The control parameters of j-MFAC
are consistent with Case 2.
It can be seen from Fig. 13 that the smaller the friction coefficient is, the larger the sideslip angle is. It can be seen from Eq.
(9) that the heading angle consists of vehicle yaw angle and side slip angle. When the expected target is constant and the

Fig. 11. The performance of heading tracking when EHSS exhibit a 1 s time delay.

14
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 12. Robustness to time delay.

actual heading angle converges to the desired heading angle, the sideslip angle should be zero. Therefore, the larger the side-
slip angle is, the slower the convergence speed of the controller output is. For different friction coefficient, the improved
MFAC scheme still has strong robustness.
Case 5: Robustness to noise
Considering the measurement error of the sensors, Gaussian white noise with mean value of zero and variance of 10 deg
is added to the feedback heading angle signal. The simulation results are shown in Fig. 14. The control parameters are con-
sistent with case1. The wheelbase of 6WID/4WIS UGV is 1.6 m, and the friction coefficient is 0.2. It can be seen from Fig. 14
that although the noise will reduce the accuracy of heading tracking (root mean square error: noise: 3.56 deg, ideal: 1.1 deg),
the improved MFAC scheme can still track the desired value. Moreover, the vehicle has a small sideslip angle, which shows
that the application of the improved MFAC control scheme in 6WID/4WIS UGV is feasible and effective.

4.2. Effectiveness of torque distribution strategy

To further illustrate the effectiveness of torque distribution strategy, the effects of different wheelbase, variable speed and
desired heading are considered in the simulation. In this paper, the condition of variable heading angle and speed is selected
to analyze, which can further illustrate the effectiveness of the control strategy. The simulation results are shown in Fig. 15,
Fig. 16 and Fig. 17. The root mean square error of heading and yaw rate under different wheelbase is illustrated in Table 3.
According to Fig. 16 WID/4WIS UGV with variable wheelbase can track the desired yaw rate effectively through torque
distribution strategy. As presented in Figs. 15 and 16, when 6WID/4WIS UGV is accelerating or decelerating while heading
tracking, the sudden change of steering angle has a strong influence on the yaw rate. Since the vehicle needs a response time,
at the beginning of acceleration and the end of deceleration, the yaw rate tracking error is relatively larger than the stable
period. In the stable period of heading tracking, the actual yaw rate of 6WID/4WIS UGV is relatively smooth, and the tracking
error is small. Fig. 17 is the torque of six in-wheel motors. Among period of 5 s ~ 20 s, traction and steering system work at
the same time, and 20–25 s is the coordinated action period of regenerative braking and steering. The difference of the

Fig. 13. Curves under different adhesion conditions.

15
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 14. Curves with noise.

Fig. 15. Heading angle and vehicle speed.

Fig. 16. Yaw rate.

output torque provides the yaw moment required during the steering process. The simulation results illustrate that the
coordinated control strategy based on MFAC method and PSO algorithm is effective and feasible.
The MFAC theory has universality to design the controller. By using this method, it is unnecessary to build an accurate
mathematical model. The improved MFAC method proposed in this paper can be applied to nonlinear systems with output
saturation. It also can be applied to control vehicles which are difficult to establish accurate models, working on the change-
able and complex road conditions and have practical control requirements, such as amphibious transport vehicles and
tracked vehicles. The torque distribution strategy based on MFAC and PSO algorithm also provides a reference for differential
steering control of distributed drive vehicles.

16
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

Fig. 17. Torque of in-wheel motors.

Table 3
Root mean square error of heading and yaw rate under different
wheelbase.

Wheelbase(m) 2.1 1.6 1.1

Heading tracking error(deg) 3.2 2.6 3.5
Yaw rate tracking error(rad/s) 0.03 0.02 0.03

The key parameters have a great impact on the control performance. In the simulation process, the optimal parameters
are found after multiple adjustments, which is the limitation of the method used in this paper. With the development of
intelligent control technology, the control method which integrates the advantages of various algorithms has been widely
concerned. In the future research, genetic algorithm, deep reinforcement learning and other methods will be combined to
realize the adaptive adjustment of control parameters. In addition, PSO algorithm may have a large computational complex-
ity. In practical application, high-performance controller is needed, and how to simplify the time complexity is also a prob-
lem worth studying.

5. Conclusions

The heading control of 6WID/4WIS UGV with variable wheelbase under low speed and noise condition is studied. The
simulation results show that the improved MFAC control scheme has better comprehensive control performance, and has
strong robustness to uncertainties, such as the time delay and variable wheelbase of 6WID/4WIS UGV. Further, in order
to realize the coordinated control of driving, regenerative braking and steering, the torque distribution is carried out based
on PSO algorithm. The simulation results of variable speed show that the torque distribution strategy can track the desired
yaw rate, ensure lateral stability and achieve coordinated control no matter where the wheelbase position is. The simulation
results show the effectiveness of the proposed control strategy and lay a foundation for the research of path tracking. The
research results can not only be applied to heading control of multi axle vehicles with variable structure, but also have ref-
erence for other special vehicle with complex dynamic characteristics, such as amphibious transport vehicles and tracked
vehicles. In the future research work, we will continue to focus on how to ensure the heading tracking accuracy of
6WID/4WIS UGV at high speed through the front wheel steering and improve the adaptability of MFAC control parameters
to ensure the stability of 6WID/4WIS UGV under various operating conditions.

CRediT authorship contribution statement

Yue Jiang: Conceptualization, Software, Validation, Writing - review & editing. Xiaojun Xu: Project administration,
Supervision, Investigation. Lei Zhang: Visualization, Investigation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.

Acknowledgements

This work was supported by the National Natural Science Foundation of China under Grant Nos. 51705524 and Science
Foundation for Young Scientists of Hunan Provience under Grant Nos. 2018JJ3605.
17
Y. Jiang, X. Xu and L. Zhang Mechanical Systems and Signal Processing 159 (2021) 107715

References

[1] C. Lv, J. Zhang, Y. Li, et al, Directional-stability-aware brake blending control synthesis for over-actuated electric vehicles during straight-line
deceleration, Mechatronics 38 (2016) 121–131.
[2] W. Zhenpo W. Yachao Z. Lei et al. Vehicle stability enhancement through hierarchical control for a four–wheel–independently- actuated electric
vehicle Energies 10(7):947, 2017.
[3] Z. Shuai, C. Li, J. Gai, et al, Coordinated motion and powertrain control of a series-parallel hybrid 8  8 vehicle with electric wheels, Mech. Syst. Sig.
Process. 120 (2019) 560–583.
[4] Z. Junjun, W. Zhenpo, Z. Lei, et al, Braking/steering coordination control for in-wheel motor drive electric vehicles based on nonlinear model predictive
control, Mech. Mach. Theory 142 (2019) 103586.
[5] G. Chen, M. Hua, C. Zong, et al, Comprehensive chassis control strategy of FWIC-EV based on sliding mode control, IET Intel. Transport Syst. 13 (4)
(2019) 703.
[6] W. Wang, H. Xu, X. Xu, et al, Enhancing the passing ability of unmanned vehicles using a variable-wheelbase driving system, IEEE Access 7 (2019)
115871–115885.
[7] H. Guo, J. Liu, D. Cao, et al, Dual-envelop-oriented moving horizon path tracking control for fully automated vehicles, Mechatronics 50 (2017) 422–433.
[8] J. Guo, Y. Luo, K. Li, et al, Coordinated path-following and direct yaw-moment control of autonomous electric vehicles with sideslip angle estimation,
Mech. Syst. Sig. Process. 105 (2018) 183–199.
[9] T. Chen, L. Chen, X. Xu, et al, Passive actuator-fault-tolerant path following control of autonomous ground electric vehicle with in-wheel motors, Adv.
Eng. Softw. 134 (2019) 22–30.
[10] C. Zhang, D. Chu, S. Liu, et al, Trajectory planning and tracking for autonomous vehicle based on state lattice and model predictive control, IEEE Intell.
Transp. Syst. Mag. 11 (2) (2019) 29–40.
[11] T. Chen, L. Chen, X. Xu, et al, Passive fault-tolerant path following control of autonomous distributed drive electric vehicle considering steering system
fault, Mech. Syst. Sig. Process. 123 (2019) 298–315.
[12] H. Taghavifar, Neural network autoregressive with exogenous input assisted multi-constraint nonlinear predictive control of autonomous vehicles,
IEEE Trans. Veh. Technol. 68 (7) (2019) 6293–6304.
[13] S. Huang, Y. Lin, V. Chinde, et al, Simulation-based performance evaluation of model predictive control for building energy systems, Appl. Energy 281
(2021) 116027.
[14] G. Hongyan, C. Dongpu, C. Hong, et al, Model predictive path following control for autonomous cars considering a measurable disturbance:
Implementation, testing, and verification, Mech. Syst. Sig. Process. 118 (2019) 41–60.
[15] Y. Wu, L. Wang, J. Zhang, et al, Path following control of autonomous ground vehicle based on nonsingular terminal sliding mode and active
disturbance rejection control, IEEE Trans. Veh. Technol. 68 (7) (2019) 6379–6390.
[16] C. Hwang, C. Yang, J.Y. Hung, Path Tracking of an autonomous ground vehicle with different payloads by hierarchical improved fuzzy dynamic sliding-
mode control, IEEE Trans. Fuzzy Syst. 26 (2) (2018) 899–914.
[17] C. Hu, H. Jing, R. Wang, et al, Robust H1 output-feedback control for path following of autonomous ground vehicles, Mech. Syst. Sig. Process. 70 (2016)
414–427.
[18] V. Saša, Raković, Robust control Minkowski-Lyapunov functions, Automatica 125 (2021) 109437.
[19] B. Li, H. Du, W. Li, A potential field approach-based trajectory control for autonomous electric vehicles with in-wheel motors, IEEE Trans. Intell. Transp.
Syst. 18 (8) (2017) 2044–2055.
[20] C. Liu, D. Wang, Y. Zhang, et al, Model predictive control for path following and roll stabilization of marine vessels based on neurodynamic
optimization, Ocean Eng. 217 (2020) 107524.
[21] H. Taghavifar, S. Rakheja, Path-tracking of autonomous vehicles using a novel adaptive robust exponential-like-sliding-mode fuzzy type-2 neural
network controller, Mech. Syst. Sig. Process. 130 (2019) 41–55.
[22] D. Zhang, Z. Xu, G. Feng, et al, Asynchronous resilient output consensus of switched heterogeneous linear multivehicle systems with communication
delay, IEEE/ASME Trans. Mechatron. 24 (6) (2019) 2627–2640.
[23] Z. Xu, H. Ni, H.R. Karimi, et al, A Markovian jump system approach to consensus of heterogeneous multiagent systems with partially unknown and
uncertain attack strategies, Int. J. Robust Nonlinear Control 30 (7) (2020) 3039–3053.
[24] H. Zhang, J. Wang, Active Steering actuator fault detection for an automatically-steered electric ground vehicle, IEEE Trans. Veh. Technol. 66 (5) (2017)
3685–3702.
[25] N. Ahmadian, A. Khosravi, P. Sarhadi, Integrated model reference adaptive control to coordinate active front steering and direct yaw moment control,
ISA Trans. 106 (2020) 85–96.
[26] Chen K, Pei X, Sun D, et al. Active steering control for autonomous vehicles based on a driver-in-the-loop platform: A case study of collision avoidance.
Proc. Inst. Mech. Eng. Part D J. Autom. Eng., 2019, 233(10): 1422-1437.
[27] X. Ji, Y. Liu, X. He, et al, Interactive control paradigm-based robust lateral stability controller design for autonomous automobile path tracking with
uncertain disturbance: a dynamic game approach, IEEE Trans. Veh. Technol. 67 (8) (2018) 6906–6920.
[28] W. Kim, K. Yi, J. Lee, An optimal traction, braking, and steering coordination strategy for stability and manoeuvrability of a six-wheel drive and six-
wheel steer vehicle, Proc. Inst. Mech. Eng. Part D J. Autom. Eng. 226 (1) (2012) 3–22.
[29] Z. Hou S. Jin Model free adaptive control; theory and applications. Science Press Beijing, 2013.
[30] Q. Jiang, Y. Li, Y. Liao, et al, Information fusion model-free adaptive control algorithm and unmanned surface vehicle heading control, Appl. Ocean Res.
90 (2019) 101851.
[31] Y. Liao, Q. Jiang, T. Du, et al, Redefined output model-free adaptive control method and unmanned surface vehicle heading control, IEEE J. Ocean. Eng.
45 (3) (2019) 714–723.
[32] E.D. Williams, Generalised multi-axle vehicle handling, Veh. Syst. Dyn. 50 (1) (2012) 149–166.
[33] Yao J , Yang G , Jiao Z . High dynamic feedback linearization control of hydraulic actuators with backstepping. Proc. Inst. Mechan. Eng. Part I J. Syst.
Control Eng, 2015, 229(8):728-737.
[34] X. Bu, Q. Wang, Z. Hou, et al, Data driven control for a class of nonlinear systems with output saturation, ISA Trans. 81 (2018) 1–7.
[35] Z. Li, H. Rufei, S. Tianmin, et al, Continuous steering stability control based on an energy-saving torque distribution algorithm for a four in-wheel-motor
independent-drive electric vehicle, Energies 11 (2) (2018) 350.
[36] L. Zhai, T. Sun, J. Wang, Electronic stability control based on motor driving and braking torque distribution for a four in-wheel motor drive electric
vehicle, IEEE Trans. Veh. Technol. 65 (6) (2016) 4726–4739.
[37] W. Zhao, Z. Luan, C. Wang, Parametric optimization of novel electric–hydraulic hybrid steering system based on a shuffled particle swarm optimization
algorithm, J. Clean. Prod. 186 (2018) 865–876.
[38] W. Zhao, X. Zhou, C. Wang, et al, Energy analysis and optimization design of vehicle electro-hydraulic compound steering system, Appl. Energy 255
(2019) 113713.
[39] H. Du J. Wei J. Fang The design, simulation, and experiment of high-accuracy multi-axle electro-hydraulic control servo steering system, Adv. Mech.
Eng. 8 10 2016 1687814016674383
[40] H. Du, Q. Zhang, S. Chen, et al, Modeling, simulation, and experimental validation of electro-hydraulic power steering system in multi-axle vehicles,
Proc. Inst. Mech. Eng. Part D J. Autom. Eng. 233 (2) (2019) 317–332.

18