actuators
Article
Dynamic Modeling and Control of a 4-Wheel Narrow
Tilting Vehicle
Sunyeop Lee 1 , Hyeonseok Cho 2 and Kanghyun Nam 1, *
Citation: Lee, S.; Cho, H.; Nam, K.
Dynamic Modeling and Control of a
4-Wheel Narrow Tilting Vehicle.
Actuators 2024, 13, 210. https://
doi.org/10.3390/act13060210
Academic Editor: Keigo Watanabe
Received: 26 April 2024
Revised: 28 May 2024
Accepted: 30 May 2024
Published: 4 June 2024
Copyright: © 2024 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Actuators 2024, 13, 210. https://doi.org/10.3390/act13060210
1 School of Mechanical Engineering, Yeungnam University, Gyeongsan 38541, Republic of Korea;
tjsduqz@yu.ac.kr
2 Hyundai (Kia Namyang) Research and Development Center, Hwaseong 18280, Republic of Korea;
chohs@hyundai.com
* Correspondence: khnam@yu.ac.kr; Tel.: +82-53-810-2455
Abstract: The automotive industries currently face challenges such as emission limits, traffic con-
gestion, and limited parking, which have prompted shifts in consumer preferences and modern
passenger vehicle requirements towards compact vehicles. However, given the inherent limited
width of compact vehicles, the potential risk of vehicle rollover is greater than that of regular vehicles.
This paper addresses the safety concerns associated with vehicle rollover, focusing on narrow tilting
vehicles (NTVs). Quantifying stability involves numerical indicators such as the lateral load transfer
ratio (LTR). Additionally, a unique approach is taken by applying ZMP (zero moment point), com-
monly used in the robotics field, as an indicator of vehicle stability. Effective roll control requires a
detailed analysis of the vehicle’s characteristic model and the derivation of lateral and roll dynamics.
The paper presents the detailed roll dynamics of an NTV with a MacPherson strut-type suspension.
A stability-enhancing method is proposed using a cascade structure based on the internal robust
position controller and outer roll stability controller, addressing challenges posed by disturbances.
Experimental verification using Simscape Multibody and CarSim validates the dynamic model
and controller’s effectiveness, ensuring the reliability of the proposed tilting control for NTVs in
practical scenarios.
Keywords: narrow tilting vehicle; lateral load transfer ratio; zero moment point; robust controller;
cascade control
1. Introduction
As global populations continue to surge alongside rapid economic development,
demand for automobiles is increasing, and transportation problems such as lack of parking
spaces and air pollution are also increasing. In response to this, there has been a heightened
focus on compact vehicles as a potential solution to mitigate the challenges associated with
increased vehicular density [1,2]. However, because compact vehicles are generally narrow,
the potential risk of vehicle rollover is greater than that of regular vehicles. Various studies
have been conducted to supplement the safety of such a roll [3], and many studies suggest
methods for stable driving by adding a structure that makes the vehicle tilt [4,5]. Commonly,
a narrow vehicle with a structure that allows the vehicle to be tilted is collectively referred
to as a narrow tilting vehicle (NTV).
NTV’s ability to actively create roll angles is similar to that of two-wheeled vehicles,
especially motorcycles. Similar to two-wheeled vehicles, which achieve stability during
cornering by tilting inwards relative to the turn radius, the NTV employs a lean towards
the turn’s inner radius to enhance stability. However, relying solely on a tilting angle
is insufficient for numerically determining stability, necessitating the use of indicators
based on a numerical model of the vehicle. One such indicator for evaluating the roll
stability of the vehicle is the lateral load transfer ratio (LTR) [6–9]. This value, determined
through the vertical load difference between the left and right wheels, provides quantitative
https://www.mdpi.com/journal/actuators
Actuators 2024, 13, 210 2 of 15

criteria for roll stability. Additionally, there is another stability criterion, which is named
zero moment point (ZMP) [10,11], based on the lateral force generated at the center of
mass in the pedestrian robot and the rotational moment generated when viewed from
the ground. Although this criterion is commonly used in robotics, our study marks the
pioneering application of this metric in the context of vehicles, showing a novel integration
of established robotic principles into the automotive domain. ZMP, the proposed roll
stability index of the vehicle in this paper, can easily show from a visual perspective how
a vehicle’s roll occurs by observing the point at which the turning moment becomes zero
relative to the ground.
Effective roll control is not possible merely by defining and calculating vehicle stability
criteria. Control logic fundamentally relies on the physical characteristics and dynamic
equations of a vehicle. Therefore, a prerequisite for roll control involves a detailed anal-
ysis of the vehicle’s characteristic model and the derivation of lateral and roll dynamics.
Formulas describing the lateral dynamics during cornering can be obtained through the
basic vehicle dynamics equations [12]. To analyze the characteristics related to roll, deriva-
tion must be performed, and equations must include the effects of lateral motions during
cornering situations. Finally, incorporating the influence of tilting actuators allows for the
derivation of the roll dynamics equation. Several studies have conducted a mathematical
approach to NTV, which has a structure in which an actuator attached to an unsprung mass
directly applies torque to the vehicle body [3,13]. Based on the basic lateral dynamics of
the vehicle, this paper provides detailed roll dynamics of a tilting structure mechanically
connected to a MacPherson strut suspension rather than a structure that directly transmits
drive torque.
Understanding the roll dynamics of tilting vehicles provides a foundation for applying
various roll stability control strategies. While some studies focus on securing roll stability
solely through the tilting actuator torque generation [14–17], some others assume that
steering intervention is possible and present integrated control between steering and tilting
mechanisms [18,19]. Although many studies propose torque inputs based on the vehicle
model, this approach may struggle to ensure stability in the presence of disturbances
like crosswinds. In this paper, a stability-enhancing method is proposed, which adopts a
cascade structure based on an internal robust position controller, mitigating the effects of
disturbances and model uncertainty.
Experimental verification is necessary to confirm the validity of the previously derived
dynamic model and the controller based on it. In this study, experiments were conducted
using a simulation model based on an existing physical vehicle model. Simulation models
were implemented using Simscape Multibody, and parameter matching was performed
to align them closely with the real vehicle through a comparison with the physically
constructed prototype. The designed model was then integrated with the CarSim driving
simulator to implement driving scenarios and validate the effectiveness of the proposed
tilting control. In addition, the stability results were analyzed in detail based on the
suggested stability criteria.

2. System Modeling
The narrow tilting vehicle (NTV) aims to prevent rollovers by allowing rotation of the
vehicle body. The tilting structures proposed in many other studies are based on a double
wishbone-type suspension. One of the models considered in this study is a system with a
tilting structure at the top of a vehicle equipped with a double wishbone-type suspension,
as shown in Figure 1a. However, considering the pros and cons of two types of suspension
systems, as shown in Figure 1, the tilting mechanism concept implemented within the
MacPherson strut-type suspension structure is proposed in this study. The mechanism
achieves vehicle inclination through the controlled rotation of the upper strut mount. The
upper plate of the suspension is attached to the axis of the tilting actuator motor, and
the motor is mounted on the vehicle cabin. Applying the MacPherson strut-type method
Actuators
Actuators2024,
Actuators2024,13,
2024,13,210
13,xxFOR
FORPEER
PEERREVIEW
REVIEW 3 3ofof
of1515
15

has
typethe
type advantage
method
method of simplifying
has the the
advantage of
of structure of
simplifying
simplifying the
the
the tilting mechanism
structure
structure ofthe
of thetiltingand
tilting reducingand
mechanism
mechanism the
and
vehicle
reducingweight.
reducing the
the vehicle weight.

(a)
(a) (b)
(b)
Figure1.1.
Figure
Figure 1. The
The tilting
tilting structures
structures
structures with
withthe
with thesuspensions:
the suspensions:(a)
suspensions: (a)double
(a) wishbone-type
double
double suspension
wishbone-type
wishbone-type and
suspension
suspension and(b)
and
(b)
MacPherson
MacPherson
(b) MacPherson strut-type
strut-type suspension.
suspension.
strut-type suspension.

Throughthe
Through
Through theimplementation
the implementationof
implementation of
ofaaasimple
simple tilting
simpletilting structure
structurein
tiltingstructure inthe
in thehardware-in-the-loop
the hardware-in-the-loop
hardware-in-the-loop
system,
system,
system,as as shown
asshown
shownin in Figure
inFigure 2, it was
Figure2,2,ititwas observed
wasobserved
observedthat that vehicles
thatvehicles
vehicleswithwith
withtilting
tiltingexhibited
tilting exhibitedaaalower
exhibited lower
lower
risk
risk of rollovers.
riskofofrollovers. However,
rollovers.However,
However,due due
due to
toto the
the challenges
challenges
the challenges in
in in analyzing
analyzing
analyzing the
thethe behavior
behavior
behavior of ofoftilting
tilting vehi-
vehicles
tilting vehi-
cles
through through simplistic
simplistic simulations,
simulations, it is it is
necessarynecessary
to deriveto derive
an
cles through simplistic simulations, it is necessary to derive an accurate mathematical an
accurate accurate mathematical
mathematical model
model
for for
suchfor
model such vehicles.
vehicles.
such vehicles.
Lateral Lateral
dynamics
Lateral dynamics of
of the
of the narrow
dynamics thenarrow
tilting tilting
narrow tiltingvehicle
vehicle mechanism
mechanism
vehicle mechanism signif-
significantly
signif-
icantlyfrom
differs differs from conventional
conventional vehicles, vehicles,
primarilyprimarily
due due
to thetoroll
the motion.
roll motion.The The derivation
derivation of
icantly differs from conventional vehicles, primarily due to the roll motion. The derivation
of dynamic
dynamic equations
equations of
of thethe NTV
NTV is
is is necessary
necessary due
due to the
totothe motion generated by the actuator.
of dynamic equations of the NTV necessary due themotion
motiongenerated
generatedby bythetheactuator.
actuator.
The firststep
The step in deriving
deriving thethemodel
modelequations
equationsisisis totocalculate thethedynamic equations for the
Thefirst
first step inin deriving the model equations calculate
to calculate dynamic
the dynamic equations
equations forfor
the
lateral
the lateralandandroll movements
roll movements that
thatoccur
occur inincornering
cornering situations.
situations. AsAs aanext
next step,
step, dynamic
dynamic
lateral and roll movements that occur in cornering situations. As a next step, dynamic
equations needtotobe
equations be derivedforfor thelateral
lateral motion when when the central actuator of the vehicle
equationsneed need to bederived
derived forthe the lateralmotion
motion whenthe thecentral
centralactuator
actuatorof ofthe
thevehicle
vehicle
transmits
transmits force
force to to
thethe main
main bodybody frame.
frame. Subsequently,
Subsequently, the the
dynamic dynamic
model model
is is derived
derived based
transmits force to the main body frame. Subsequently, the dynamic model is derived
onbased
the on thetransmitted
forces forces transmitted
to the to the mainframe
mainframe and and suspension
suspension as the as the
motor motor operates.
operates. Through
based on the forces transmitted to the mainframe and suspension as the motor operates.
Through
this process, this process, this sectionthe presents theequations
dynamic equations of the tilting vehicle.
Through thisthis sectionthis
process, presents dynamic
section presents the dynamic of the tilting
equations of vehicle.
the tilting vehicle.

(a) (b)
(a) (b)
Figure 2. Hardware-in-the-loop simulation of the narrow vehicle: (a) normal vehicle and (b) simple
Figure 2.Hardware-in-the-loop
Hardware-in-the-loopsimulation
tilting2.vehicle.
Figure simulationofofthe
thenarrow
narrowvehicle:
vehicle:(a)
(a)normal
normalvehicle
vehicleand
and(b)
(b)simple
simple
tilting vehicle.
tilting vehicle.
2.1. Lateral Dynamics for Narrow Tilting Vehicle
2.1.
2.1.Lateral
Lateral Dynamics
Dynamicsfor forNarrow
Narrow Tilting Vehicle
In conventional vehicles, theTilting Vehicle is basically based on the difference in load
roll dynamics
In
Inconventional
between conventional vehicles,
the left and vehicles, theroll
the
right suspension rolldynamics
dynamics
due is is
to lateral basically
basically
force inbasedbasedon on
cornering the the difference
difference
situations. Inincon-in
load
load between
between the the
left left
and and
right right suspension
suspension due todue to
lateral lateral
force inforce in
corneringcornering
trast, tilting vehicles can actively make the roll angle through actuators. Based on the lat- situations.
situations. In con-
In contrast,
trast,
eral tiltingtilting
dynamics of avehicles
vehicles general can
can actively actively
vehicle and make
make the rollthe
adding roll
angle
the angle
of through
through
effect actuators.
actuators.
the tilting Based on
structure, Based on
thedy-
the lat-
the
erallateral
namics
dynamicsdynamics
equation of for of a general
the tilting
a general vehicle
vehicle
vehicle and
andcan adding
be derived.
adding the
the effect effect
The of of
lateral the tilting
force during
the tilting structure, the
cornering
structure, the dy-
dynamics equationfor
namics equation forthe
thetilting
tiltingvehicle
vehiclecan
can bebe derived.
derived. TheThe lateral
lateral force
force during
duringcornering
cornering
Actuators 2024, 13, x FOR PEER REVIEW 4 of 15
Actuators 2024, 13, 210 4 of 15

situations, as depicted in Figure 3a, can be derived similarly to conventional vehicles, and
situations,
its equationsasare
depicted
derivedinasFigure 3a, can
Equations be derived similarly to conventional vehicles, and
(1)–(4).
its equations are derived as Equations (1)–(4).
𝑚𝑎 = 𝐹 + 𝐹 𝑐𝑜𝑠(𝛿) ≈ 𝐹 + 𝐹 (1)
may = Fr + Ff cos(δ) ≈ Fr + Ff (1)
𝐼 ..𝜓 = −𝑙 𝐹 + 𝑙 𝐹 𝑐𝑜𝑠 (𝛿) ≈ −𝑙 𝐹 + 𝑙 𝐹 (2)
Iz ψ = −lr Fr + l f Ff cos(δ) ≈ −lr Fr + l f Ff (2)
𝑉, −𝑉 𝛿 + 𝑉 + 𝜓. 𝑙  −𝑉 𝛿 + 𝑉 + 𝜓𝑙 . 
𝐹 = −𝐶 , 𝛼 ≈ −𝐶 V , = −𝐶 , −Vlon δ + Vlat + ψl f ≈ −𝐶 , −Vlon δ + Vlat + ψl f (3)
𝑉, 𝑉
= −Cα, f 𝑉 + 𝑉 + 𝜓. 𝑙 𝛿 ≈ −Cα, f
A,η
Ff = −Cα, f α f ≈ −Cα, f (3)
VA,ξ Vlon + Vlat + ψl f δ Vlon
𝐼 𝜓 = −𝑙 𝐹 + 𝑙 𝐹 𝑐𝑜𝑠 (𝛿) ≈ −𝑙 𝐹 + 𝑙 𝐹 (4)
..
Iz ψ = −lr Fr + l f Ff cos(δ) ≈ −lr Fr + l f Ff (4)

(a) (b)
Figure
Figure3.3.The vehicle
The dynamics
vehicle model:
dynamics (a) lateral
model: dynamics
(a) lateral model
dynamics when
model cornering
when and (b)
cornering androll
(b)dy-
roll
namics model.
dynamics model.

The
Thelateral
lateraldynamics
dynamicsofofa atilting
tiltingvehicle,
vehicle,asasdepicted
depictedininFigure
Figure3b,3b,can
canbebeexpressed
expressed
ininterms
termsofoflateral
lateralforces
forcesacting
actingon
onthe
theground,
ground,lateral
lateralacceleration,
acceleration,and
andthethevehicle’s
vehicle’sroll
roll
angle.
angle.Equation
Equationofofthethedynamics
dynamics for force R𝑅y and
for the lateral force andvertical forceRz𝑅acting
verticalforce acting
onon
the
the
tiretire from
from thethe ground
ground is derived
is derived as Equations
as Equations (5) (5)
andand
(6).(6).
𝑅 = 𝑚𝑦.. + 𝑚𝜑𝑉
. + 𝑚ℎ𝜃..𝑐𝑜𝑠𝜃 − 𝑚ℎ𝜃
. 2 𝑠𝑖𝑛𝜃 = 𝑚𝑎 (5)
Ry = my + m φV + mhθcosθ − mhθ sinθ = may (5)
𝑅 = 𝑚𝑔 − 𝑚ℎ𝜃..𝑠𝑖𝑛𝜃 − 𝑚ℎ𝜃 . 2 𝑐𝑜𝑠𝜃 (6)
Rz = mg − mhθsinθ − mhθ cosθ (6)
When considering the rotational moment generated by the actuator at the center of
gravity 𝐺 asconsidering
When 𝑀 , the equation
the rotational moment
of the roll generated
dynamics by the
is derived as actuator
follows: at the center of
gravity G as Mtilt , the equation of the roll dynamics is derived as follows:
𝐼 𝜃 = 𝑅 ℎ𝑠𝑖𝑛𝜃 − 𝑅 ℎ𝑐𝑜𝑠𝜃 + 𝑀
.. (7)
= 𝑚𝑔ℎ𝑠𝑖𝑛𝜃 − 𝑚ℎ Ix θ𝜃= 𝑠𝑖𝑛Rz𝜃hsinθ
− 𝑚ℎ −𝜃 hcosθ + M−tilt
Ry𝑐𝑜𝑠𝜃𝑠𝑖𝑛𝜃 𝑚𝑎 ℎ𝑐𝑜𝑠𝜃 + 𝑀
.. .2 (7)
= mghsinθ − mh2 θsin2 θ − mh2 θ cosθsinθ − may hcosθ + Mtilt
2.2. Tilting Mechanism Dynamics
2.2. Tilting Mechanism Dynamics
The tilting structure of the vehicle proposed in this paper is mechanically connected
The tilting structure of the vehicle proposed in this paper is mechanically connected
to a MacPherson strut-type suspension through a rotating plate. The torque generated by
to a MacPherson strut-type suspension through a rotating plate. The torque generated by
the tilting actuator is not directly transmitted to the vehicle’s body moment. Instead, it is
the tilting actuator is not directly transmitted to the vehicle’s body moment. Instead, it is
configured in a complex structure of 2DOF with suspension parts and body parts. Unlike
configured in a complex structure of 2DOF with suspension parts and body parts. Unlike
a astructure
structurewhere
wherethethemotor
motorisisfixed
fixedatataaspecific
specificposition
positiontotogenerate
generatetorque,
torque,this
thissystem
system
isisdesigned to transmit torque to both the rotor and stator sides. The housing
designed to transmit torque to both the rotor and stator sides. The housing containingcontaining
the
themotor’s
motor’sstator
statorisisattached
attachedtotothethebody,
body,which
whichcorresponds
correspondstotothe
thesprung
sprungmass.
mass.The
The
Actuators 2024,13,
Actuators2024, 13,210
x FOR PEER REVIEW 5 of
of1515

rotor
rotorand
andshaft
shaftare
areconnected
connectedtotoaaplate
plateattached
attachedtotothe
theupper
upperpart
partofofthe
thesuspension.
suspension.The
The
structure of the tilting part can be expressed as shown in Figure 4.
structure of the tilting part can be expressed as shown in Figure 4.

Figure4.4.Simplified
Figure Simplifieddynamics
dynamicsmodel
modelofoftilting
tiltingstructure.
structure.

The
Thelinear
linearmotion
motionofofthe thesuspension
suspensionresultingresultingfromfromthe therotation
rotationofofthe
theplate
plateconnected
connected
totothe
the motor shaft in Figure 4 is represented as a single rotational component.The
motor shaft in Figure 4 is represented as a single rotational component. Thetorque
torque
generated
generatedbetween
betweenthe rotor
the rotor andand stator is transmitted
stator is transmitted through the housing
through and motor
the housing shaft.
and motor
The torque applied to the motor shaft is transmitted through the rotating
shaft. The torque applied to the motor shaft is transmitted through the rotating plate, gen- plate, generating
the motion
erating theofmotion
the suspension. As the motor
of the suspension. As the housing
motorishousing
connected to the vehicle’s
is connected to thebody, its
vehicle’s
force acts as the body moment. This force relationship can be
body, its force acts as the body moment. This force relationship can be expressed mathe- expressed mathematically
as follows: as follows:
matically
Tm = Tm,sha f t − Tm,housing = Tm,sha f t − Mtilt (8)
𝑇 =𝑇 , −𝑇 , =𝑇 , −𝑀 (8)
The tilting structure can be defined by the equation of the angle of the motor shaft
(θsha f tThe
) andtilting structure
the roll angle ofcan thebe bodydefined
(θbodyby the equation
), where of the between
the difference angle of the two
motor shaft
angles
is(𝜃θm , defined
) and the
as roll
θm = angle
θsha foft − the
θ body
body . (𝜃
Based on), where
this the difference
relationship, the between
dynamics the
of two
the angles
tilting
is 𝜃 , defined
actuator as 𝜃 =as
can be derived 𝜃 Equation − 𝜃 (9). . Based on this relationship, the dynamics of the tilt-
ing actuator can be ..derived as Equation . (9)..
Jm θ sha f t = − Bm θ m − Bsusp θ sha f t − k susp θsha f t + Tm,sha f t
𝐽 𝜃 .= −𝐵 𝜃 − . 𝐵 𝜃 −𝑘 𝜃 +𝑇 , (9)
= − Bm θ m − Bsusp θ sha f t − k susp θsha f t + Tm + Mtilt (9)
= −𝐵 𝜃 − 𝐵 𝜃 −𝑘 𝜃 +𝑇 +𝑀
3. Design of the Roll Stability Criteria
The objective
3. Design of this
of the Roll paper isCriteria
Stability to improve the stability of roll through the control of tilting
vehicles. Therefore, it is necessary to define stability criteria for evaluating roll stability.
The objective of this paper is to improve the stability of roll through the control of
Tilting vehicles can be conceptually likened to motorcycles in their ability to generate roll
tilting vehicles. Therefore, it is necessary to define stability criteria for evaluating roll sta-
angles. Generally, in the context of a two-wheeled system like a motorcycle, the driver
bility. Tilting vehicles can be conceptually likened to motorcycles in their ability to gener-
feels safe about lateral motion when the lateral acceleration of the driver converges to zero.
ate roll angles. Generally, in the context of a two-wheeled system like a motorcycle, the
Hence, the driver’s lateral acceleration can be one of the criteria for evaluating roll stability.
driver feels safe about lateral motion when the lateral acceleration of the driver converges
In addition to lateral acceleration, two more specific stability criteria are proposed, which
to zero. Hence, the driver’s lateral acceleration can be one of the criteria for evaluating roll
are directly associated with the rollover and can be evaluated numerically.
stability. In addition to lateral acceleration, two more specific stability criteria are pro-
posed,
3.1. which
Lateral Loadare directly
Transfer associated with the rollover and can be evaluated numerically.
Ratio
One of the criteria for evaluating the roll stability of a vehicle is the lateral load transfer
3.1. Lateral Load Transfer Ratio
ratio (LTR), which is defined as follows:
One of the criteria for evaluating the roll stability of a vehicle is the lateral load trans-
fer ratio (LTR), which is defined as follows: F − Fzr
LTR ≜ zl (10)
Fzl + Fzr
𝐹 −𝐹
𝐿𝑇𝑅 ≜ (10)
𝐹 + 𝐹in vertical loads between the left and
This index represents the ratio of the difference
right sides. In a situation where there is no difference in vertical load between the left
This index represents the ratio of the difference in vertical loads between the left and
and right sides, this index is measured as 0. As the difference in vertical loads increases,
right sides. In a situation where there is no difference in vertical load between the left and
the absolute value of this index increases. In a scenario where one wheel of the vehicle
right sides, this index is measured as 0. As the difference in vertical loads increases, the
loses contact with the ground, leading to a rollover situation, one vertical load becomes 0,
absolute value of this index increases. In a scenario where one wheel of the vehicle loses
resulting in the LTR becoming either 1 or −1. Therefore, maintaining this value within a
contact with the ground, leading to a rollover situation, one vertical load becomes 0,
certain range is crucial for preventing vehicle rollover.
Actuators 2024, 13, x FOR PEER REVIEW 6 of 15

Actuators 2024, 13, 210 6 of 15

resulting in the LTR becoming either 1 or −1. Therefore, maintaining this value within a
certain range is crucial for preventing vehicle rollover.
The
Thesumsumofofthethevertical
verticalloads
loadsonon
the left
the and
left right
and sides
right of the
sides vehicle
of the is equivalent
vehicle to
is equivalent
the totaltotal
to the vertical loadload
vertical and can
andbe simply
can expressed
be simply as 𝑚𝑔.
expressed mg.difference
asThe in vertical
The difference loads
in vertical
loads between
between the left the
andleft and
right rightassides,
sides, shown asinshown
Figurein5,Figure 5, is influenced
is influenced by the
by the lateral lateral
acceler-
acceleration
ation due to thedue to the cornering
cornering and the
and the force force of Furthermore,
of gravity. gravity. Furthermore, considering
considering the effect the
of
effect ofacceleration,
angular angular acceleration, this difference
this difference can be expressed
can be expressed by Equationby Equation
(11). (11).

22  .. + 𝑚ℎ𝑎 − 𝑚𝑔ℎ𝜃
F𝐹zl −−F𝐹zr == 𝑇 𝐼Ix𝜃θ + mhay − mghθ (11)
(11)
Tw

(a) (b)
Figure
Figure5.5.The
Thenarrow
narrowtilting
tiltingvehicle
vehiclestructure:
structure:(a)
(a)structure
structureofofthe
thetilting
tiltingvehicle
vehiclewith
withrotational
rotationalplate
plate
and (b) simplified dynamics model of the tilting vehicle.
and (b) simplified dynamics model of the tilting vehicle.

Consequently,
Consequently, the LTR value
value can
can be
bederived
derivedas
asEquation
Equation(12)
(12)bybyconsidering
considering the
the lat-
lateral
eral acceleration
acceleration andand
roll roll angle.
angle.
2  ..
= 2
𝐿𝑇𝑅 = 𝐼I 𝜃θ +
+ 𝑚ℎ𝑎 − mghθ
𝑚𝑔ℎ𝜃

LTR mhay − (12)
(12)
𝑇 𝑚𝑔
Tw mg
x

3.2.Zero
3.2. ZeroMoment
MomentPoint
Point
Theconcept
The conceptofofthe
thezero
zeromoment
momentpoint point(ZMP)
(ZMP)isisaacritical
criticalparameter
parameterininthe thefield
fieldofof
humanoid robotics and dynamic stability analysis. ZMP refers to the point
humanoid robotics and dynamic stability analysis. ZMP refers to the point on the ground on the ground
atatwhich
whichthe themoment
momentacting
actingon
onthe
thecenter
centerof ofmass
massofofthe
thewalking
walkingor orbalancing
balancingsystem
systemisis
equaltotozero.
equal zero.Mathematically,
Mathematically,the theZMP
ZMPcan canbebedefined
definedasasthe
thepoint
pointononthe
theground
groundplaneplane
where the sum of the moments
where the sum of the moments is zero. is zero.
Applyingthis
Applying thisconcept
concepttotovehicles
vehiclesenables
enablesthethedetection
detectionof ofvehicle
vehiclestability
stabilitythrough
through
the position of the ZMP. When the ZMP is located between the left and
the position of the ZMP. When the ZMP is located between the left and right wheels right wheels of the
of
vehicle, rollover does not occur because the point where the moment
the vehicle, rollover does not occur because the point where the moment becomes 0 isbecomes 0 is inside
the contact point. A rollover occurs when the ZMP is located outside the vehicle’s wheels,
inside the contact point. A rollover occurs when the ZMP is located outside the vehicle’s
causing a rotating moment outside the vehicle. Therefore, it can be considered safer when
wheels, causing a rotating moment outside the vehicle. Therefore, it can be considered
the ZMP is closer to the midpoint between the two wheels, and this point can be used as
safer when the ZMP is closer to the midpoint between the two wheels, and this point can
stability criteria.
be used as stability criteria.
In tilting vehicles, the ZMP can be defined as the point where the resultant force of
In tilting vehicles, the ZMP can be defined as the point where the resultant force of
lateral and gravitational contacts the ground. Assuming that the lateral acceleration of the
lateral and gravitational contacts the ground. Assuming that the lateral acceleration of the
vehicle can be measured, the direction of the force vector originating from the center of
vehicle can be measured, the direction of the force vector originating from the center of
mass can be expressed as Equation (13).
mass can be expressed as Equation (13).
→ → →
𝐹⃗ =
F CoG F⃗ y +
=𝐹 +𝐹⃗F z==𝑚𝑎
ma𝑦y ŷ−−𝑚𝑔𝑧̂
mgẑ (13)
(13)
 as illustrated in Figure 6b, we obtain the contact point. When considering the midpoint
between the two wheels as the reference point, the displacement of the vehicle’s center of
mass, influenced by the vehicle’s rotation, is given by a 𝑦-axis position of ℎ 𝑠𝑖𝑛 𝜃 and a
𝑧-axis position of ℎ 𝑐𝑜𝑠 𝜃. Through this, distance from the center point of the two wheels
Actuators 2024, 13,
Actuators x FOR
2024, PEER REVIEW
13, 210 to the ZMP can be calculated, and the formula of the ZMP is derived as follows: 7 of 15 7 of 15

𝑎 ℎ
𝑍𝑀𝑃 = ℎ 𝑠𝑖𝑛 𝜃 + 𝑐𝑜𝑠 𝜃 (14)
𝑔
Extending
Extending this vector
this vector and performing
and performing the mathematical
the mathematical intersection
intersection with
with the the ground,
ground,
as illustrated
as illustrated in Figure
in Figure 6b, we6b,obtain
we obtain the contact
the contact point.point.
WhenWhen considering
considering the midpoint
the midpoint
between
between the twothewheels
two wheels
as theas the reference
reference point,point, the displacement
the displacement of theofvehicle’s
the vehicle’s
centercenter
of of
mass, influenced by the vehicle’s rotation, is given by a 𝑦-axis position of ℎ 𝑠𝑖𝑛 𝜃 andaaz-axis
mass, influenced by the vehicle’s rotation, is given by a y-axis position of hsinθ and
positionofofhcosθ.
𝑧-axisposition Through
ℎ 𝑐𝑜𝑠 𝜃. Through this, distance
this, distance from
fromthe center
the centerpoint
pointofofthe
thetwo
twowheels
wheelsto the
ZMP can be calculated, and the formula of the ZMP is derived
to the ZMP can be calculated, and the formula of the ZMP is derived as follows: as follows:

𝑎 ℎ ay h
𝑍𝑀𝑃 ZMP
= ℎ 𝑠𝑖𝑛
= 𝜃hsinθ
+ + 𝑐𝑜𝑠 𝜃cosθ (14) (14)
𝑔 g

(a) (b)
Figure 6. The zero moment point (ZMP): (a) ZMP of the walking system and (b) ZMP of the tilting
vehicle.

4. Design of a Robust Tilting Control System

The control block diagram is designed as a cascade structure to enhance roll stability,
as illustrated in Figure 7. The cascade structure offers several advantages, including im-
proved performance, robustness, and easier tuning. By separating the control tasks into
inner and outer(a) loops, the cascade structure allows each loop (b)
to be tuned independently,
simplifying
FigureFigure
6. The 6.
zero the overall
moment control
point (ZMP): design. The inner loop can be optimized for fasttilting
and accu-
The zero moment point(a) ZMP of
(ZMP): (a)the
ZMPwalking
of thesystem
walkingandsystem
(b) ZMP andof (b)
the ZMP of the
rate
vehicle. roll angle control, while the outer loop can focus on adapting to different driving
tilting vehicle.
conditions. This hierarchical approach improves the system’s robustness against disturb-
ances
4. Design ofand
4. Design enhances
of a Robust
a Robust theTilting
Tilting vehicle’s
Control overall
Control
System handling performance.
System
The inner loop
The control controller isisdesigned as
asaa roll anglestructure
control loop. Its primary function
The control blockblock
diagramdiagram designed
is designed cascade
as a cascade to enhance
structure to enhance rollroll stability,
stability, as
is to track
illustrated the desired roll angle with high precision and quick response. By focusing on
as illustrated in in Figure7.7.The
Figure Thecascade
cascadestructure
structureoffers
offers several advantages,
advantages, including
includingimproved
im-
the roll angle robustness,
performance, directly, theand inner looptuning.
easier ensuresByrapid correction,
separating thereby
the control enhancing
tasks into inner the ve-
proved performance, robustness, and easier tuning. By separating the control tasks into and
hicle’s stability in real time.
inner outer loops,loops,
and outer the cascade structure
the cascade allowsallows
structure each loop
eachto be tuned
loop independently,
to be tuned independently,
∗
simplifying
the The outer
overall loop,
control on the
design. Theother
inner hand,
loop generates
can be the target
optimized for roll
fast angle(𝜃
and ) based
accurate rollon the
angle
simplifying the overall control design. The inner loop can be optimized for fast and accu-
vehicle’s
control, overallthestate,
outerincludingcan speed, lateral acceleration, anddriving
other dynamic parame-
rate roll anglewhile
control, while loop
the outer focus
loop oncanadapting
focus ontoadapting
different to different conditions.
driving This
ters. This loop
hierarchical takes aimproves
approach broader view the of the vehicle’s
system’s robustness dynamics
against and adjusts the
disturbances and target roll
enhances
conditions. This hierarchical approach improves the system’s robustness against disturb-
angle
the to optimize
vehicle’s overall stability
handlingunder varying conditions.
performance.
ances and enhances the vehicle’s overall handling performance.
The inner loop controller is designed as a roll angle control loop. Its primary function
is to track the desired roll angle with high precision and quick response. By focusing on
the roll angle directly, the inner loop ensures rapid correction, thereby enhancing the ve-
hicle’s stability in real time.
The outer loop, on the other hand, generates the target roll angle(𝜃 ∗ ) based on the
vehicle’s overall state, including speed, lateral acceleration, and other dynamic parame-
ters. This loop takes a broader view of the vehicle’s dynamics and adjusts the target roll
angle to optimize stability under varying conditions.

Figure 7. Block diagram of the vehicle controller.

The inner loop controller is designed as a roll angle control loop. Its primary function
is to track the desired roll angle with high precision and quick response. By focusing on the
roll angle directly, the inner loop ensures rapid correction, thereby enhancing the vehicle’s
stability in real time.
The outer loop, on the other hand, generates the target roll angle(θ ∗ ) based on the
vehicle’s overall state, including speed, lateral acceleration, and other dynamic parameters.
Actuators 2024, 13, 210 8 of 15
Actuators 2024, 13, x FOR PEER REVIEW 8 of 15

This loop takes a broader view of the vehicle’s dynamics and adjusts the target roll angle to
optimize stability
Figure 7. Block under
diagram ofvarying conditions.
the vehicle controller.

4.1. Roll Angle Controller

4.1. Roll Angle Controller
The roll angle controller of the vehicle is designed based on the dynamic model
The roll angle controller of the vehicle is designed based on the dynamic model be-
between the vehicle’s roll angle and motor torque. The relationship between the motor
tween the vehicle’s roll angle and motor torque. The relationship between the motor angle
angle and the roll angle, previously expressed in a 2DOF, is simplified into a 1DOF rotational
and the roll angle, previously expressed in a 2DOF, is simplified into a 1DOF rotational
model. The nominal model in the control logic is expressed as simply as possible, and an
model. The nominal model in the control logic is expressed as simply as possible, and an
additional robust controller is applied. Inertia coefficient Jn is the nominal roll inertia for
additional robust controller is applied. Inertia coefficient Jn is the nominal roll inertia for
the entire vehicle, while Bn is the nominal damping, such as friction from bushings and
the entire vehicle, while Bn is the nominal damping, such as friction from bushings and
reduction gear in the vehicle. The dynamic equation of the model is expressed as follows:
reduction gear in the vehicle. The dynamic equation of the model is expressed as follows:
θ (s𝜃(𝑠)
) 11
= (15)
s) =Jn𝐽s2𝑠++
T (𝑇(𝑠) Bn𝐵s 𝑠 (15)

The
Theroll
rollangle
anglecontroller
controllerfor forthe
thevehicle
vehicleconsists of of
consists feedback, feedforward,
feedback, andand
feedforward, distur-
dis-
bance observer
turbance (DOB)
observer based
(DOB) on the
based on nominal
the nominalmodel. A complete
model. block
A complete diagram
block of the
diagram ofroll
the
angle controller
roll angle is shown
controller in Figure
is shown 8. The
in Figure 8. feedback controller
The feedback is designed
controller usingusing
is designed pole-zero
pole-
cancellation basedbased
zero cancellation on theonnominal model
the nominal in Equation
model (15).(15).
in Equation The The
feedback control
feedback equation
control equa-
with
tion tuning parameter
with tuning w f b is𝑤
parameter expressed as follows:
is expressed as follows:

( Jn(𝐽s +
𝑠+ Bn𝐵)w)𝑤
CFB𝐶(s)(𝑠)
== 𝜏 𝑠+1
fb (16)
(16)
τf b s + 1

Figure8.8.Block
Figure Blockdiagram
diagramof
ofthe
theroll
rollangle
anglecontroller.
controller.

To
Toimprove
improvethe theposition
positioncontrol
controlresponse
responsetime,
time,the
thefeedforward
feedforwardisisdesigned
designedby bytaking
taking
the inverse of the nominal model and multiplying it with a second-order order
the inverse of the nominal model and multiplying it with a second-order order low-pass low-pass
filter
filterfor
forstability.
stability.The
Theequation
equationof
ofthe
thefeedforward
feedforwardcontroller
controllerisisexpressed
expressedasasfollows:
follows:
Jn(𝐽s2𝑠+ +
Bn𝐵s 𝑠)𝑤

2
wf f
CFF𝐶(s)(𝑠)
= =2 (17)
(17)
s 𝑠+ + 2𝜁𝑤
2ζw 𝑠 + 2𝑤
f f s + wf f

The tracking performance of the position control may be unstable due to the model
The tracking performance of the position control may be unstable due to the model un-
uncertainty if only feedback and feedforward controllers are used. Additionally, in cor-
certainty if only feedback and feedforward controllers are used. Additionally, in cornering
nering situations, the lateral force is transmitted to the motors, resulting in high position
situations, the lateral force is transmitted to the motors, resulting in high position error. To
error. To compensate for these model uncertainties and disturbances, a disturbance ob-
compensate for these model uncertainties and disturbances, a disturbance observer (DOB)
server (DOB) is used as an additional control structure. DOB predicts the current input of
is used as an additional control structure. DOB predicts the current input of the nominal
the nominal model and compares it to the actual input, predicting the difference as a dis-
model and compares it to the actual input, predicting the difference as a disturbance. This
turbance. disturbance
predicted This predicted disturbance
passes throughpasses through a low-pass
a second-order second-order
filter,low-pass
which isfilter, which
expressed
is expressed as 𝑄(𝑠), to observe disturbances below a certain frequency,
as Q(s), to observe disturbances below a certain frequency, while signals beyond thiswhile signals
beyond this
frequency frequency
range range This
are blocked. are blocked.
structureThis
can structure
predict andcancompensate
predict andforcompensate
disturbances,for
disturbances, effectively removing noise and enhancing the tracking performance of the
position control. The mathematical expression for the second-order low-pass filter is ex-
pressed as follows:
Actuators 2024, 13, 210 9 of 15

effectively removing noise and enhancing the tracking performance of the position control.
The mathematical expression for the second-order low-pass filter is expressed as follows:

w2Q
Q(s) = (18)
s2 + 2ζwQ s + w2Q

4.2. Roll Stability Controller

The outer control loop, which generates the target roll angle of the tilting vehicle, is
based on lateral acceleration during cornering. The lateral acceleration of the vehicle’s body,
originating from the center of gravity, is typically expressed as shown in Equation (19).
..
.. ..
aym = y + γv x cosθ + hθ − gsinθ = ay cos θ + hθ − gsinθ (19)

The target roll angle is calculated to converge lateral acceleration to zero based on this
..
equation. For simplicity, the lateral acceleration caused by the side slip y and the angular
..
acceleration θ are assumed to be zero in the equation. The vehicle’s yaw rate γ is calculated
as γ = v x /R, and the cornering radius R can be expressed in relation to the steering angle
δ f and the wheelbase L. The formula for calculating the target roll angle is represented as a
function of the vehicle’s longitudinal velocity and steering angle, as shown in Equation (20).

ay v2x v2x δ f
   
θ d ≈ tan−1 =tan−1 ≈ (20)
g gR gL

Directly using the values generated by vehicle speed and steering can result in a
too-rapid response in comparison to the steering angle, potentially causing discomfort
for the driver. To control the response speed of the target roll angle, a steering response
coefficient τ has been added to the controller. Additionally, to prevent the excessive roll
angle, k has been included as a tuning parameter for the target roll angle. The transfer
function H (s) for tuning the roll angle is expressed in Equation (21), and the final formula
for generating the target roll angle is represented as shown in Equation (22).

k
H (s) = (21)
τs + 1

v2x δ f
 
k
θd = θ d H (s) = (22)
gL τs + 1

5. Simulation Result
The tilting vehicle introduced in this paper is based on an actual vehicle with tilting at-
tached to a MacPherson strut-type suspension. For the simulation, the Simscape Multibody
model is constructed based on 3D drawings of the actual vehicle. The connection structure
between each component is constructed based on rotational and linear movements, giving
each part the necessary degree of freedom to allow movement of each part. The component
connection of the physical model is designed to be as similar as possible to the actual system,
but since the model is too complex to measure and consider all friction and deformation
of each part. Therefore, the multibody model is constructed by prioritizing simplicity
while preserving similarity, such as applying friction only to areas with significant friction
components. Components related to the tilting structure, including the actuator motor and
suspension, are designed by combining analytically derived models proposed in equation
11 and multibody physical models to ensure the accuracy of the simulation model.
A parameter matching process is performed to ensure similarity between the sim-
ulation model and the actual model. Various sine signals of different magnitudes and
frequencies are applied to perform parameter matching based on the roll angle and motor
position. The parameters such as friction, inertia, and spring constant are adjusted to
 Actuators2024,
Actuators 2024,13,
13,xxFOR
FORPEER
PEERREVIEW
REVIEW 10 ofof 15
10 15

Actuators 2024, 13, 210 10 of 15

frequenciesare
frequencies areapplied
appliedto toperform
performparameter
parametermatching
matchingbasedbasedononthe
theroll
rollangle
angleand
andmotor
motor
position. The parameters such as friction, inertia, and spring constant
position. The parameters such as friction, inertia, and spring constant are adjusted to are adjusted to
match
match the results with the actual vehicle’s behavior. The results for sine input response
matchthe theresults
resultswith
withthe theactual
actualvehicle’s
vehicle’sbehavior.
behavior.The Theresults
resultsfor
forsine
sineinput
inputresponse
response
comparison
comparison are
are shown
shown in
in Figure
Figure 99 and
and Figure
Figure 10.
10. It
It is
is difficult
difficult
comparison are shown in Figures 9 and 10. It is difficult to obtain the same to
to obtain
obtain the
the same
same results
results
results due
due
due
to theto the
to the differences
differences
differences between
between
between the
the the actual
actual
actual vehicle
vehicle
vehicle and
andand
the thethe simulation
simulation
simulation model,
model,
model, such such
such asleft
as
as left left
and
and
and right
rightright
frictionfriction
friction conditions,
conditions,
conditions, initial
initialinitial position,
position,
position, and
and
and gear gear backlash.
gear backlash.
backlash. Therefore,
Therefore,
Therefore, parameter parameter
parameter
matching
matching
matching
is performedisisperformed
performed
to produce toproduce
to produce
as similar assimilar
as similar
behavior behavior
behavior aspossible
as
as possible possible byfocusing
by
by focusingfocusing
on the onphase
on thephase
the phase
and
and
and magnitude
magnitude
magnitude when
whenwhenvarious various
various signals
signalssignals are given.
are The
are given. given. The parameters
The parameters
parameters applied in applied
applied
Simulink in Simulink
in multibody
Simulink
multibody
multibody
are listed in are
are listed
listed
Table inTable
1. in Table1.1.

Figure9.9.Position
Figure Position responseof
of thereal
real systemand
and simulation.
Figure 9. Positionresponse
response ofthe
the realsystem
system andsimulation.
simulation.

Figure 10.
Figure10.
Figure Velocity response
Velocityresponse
10.Velocity of
responseof the
ofthe real
thereal system
realsystem and
systemand simulation.
andsimulation.
simulation.

Table
Table1.1.
Table The
1.The parameters
Theparameters of
parametersof the
ofthe simulation
thesimulation model.
simulationmodel.
model.
𝑴𝒔𝒑𝒓𝒖𝒏𝒈
𝑴 𝑴𝑴𝒖𝒏𝒔𝒑𝒓𝒖𝒏𝒈 𝒌𝒌𝒔𝒖𝒔𝒑𝒆𝒏𝒔𝒊𝒐𝒏
𝒔𝒖𝒔𝒑𝒆𝒏𝒔𝒊𝒐𝒏 𝑪𝑪𝒔𝒖𝒔𝒑𝒆𝒏𝒔𝒊𝒐𝒏
𝒔𝒖𝒔𝒑𝒆𝒏𝒔𝒊𝒐𝒏 𝒉𝒉𝒎𝒐𝒕𝒐𝒓
𝒎𝒐𝒕𝒐𝒓 𝒉𝒉𝑪𝒐𝑮
𝑪𝒐𝑮 𝑳𝑳𝒕𝒓𝒂𝒄𝒌𝒘𝒊𝒅𝒕𝒉
𝒕𝒓𝒂𝒄𝒌𝒘𝒊𝒅𝒕𝒉 𝑩𝑩𝒇𝒓𝒊𝒄𝒕𝒊𝒐𝒏
𝒇𝒓𝒊𝒄𝒕𝒊𝒐𝒏
𝒔𝒑𝒓𝒖𝒏𝒈 𝒖𝒏𝒔𝒑𝒓𝒖𝒏𝒈
M550
sprung
kg Munsprung
100kg
kg ksuspension
9810 N/m Csuspension
2400 N/(m/s) hmotor
0.31 m hCoG
0.43 m Ltrackwidth
0.825 m Bfriction
120 Nm/(deg/s)
550 kg 100 9810 N/m 2400 N/(m/s) 0.31 m 0.43 m 0.825 m 120 Nm/(deg/s)
550 kg 100 kg 9810 N/m 2400 N/(m/s) 0.31 m 0.43 m 0.825 m 120 Nm/(deg/s)
Thecontroller
The controllerdesigned
designedin inthis
thispaper
paperadopts
adoptsaacascade
cascadestructure
structurewith withan aninternal
internalrollroll
angle
angle The control
control loop.
loop. To To ensure
ensure in the
thethisperformance
performance and safety
anda safety of the external
ofstructure
the external control logic, itit
controller designed paper adopts cascade withcontrol logic,roll
an internal
is essential
isangle
essential to first
to first verify the performance of the internal control loop. To evaluate the
control loop.verify the performance
To ensure the performance of the andinternal
safety control loop. Tocontrol
of the external evaluate theit
logic,
performanceof
performance ofthe
theinternal
internalroll rollangle
anglecontrol,
control,aasimulation
simulationisisconducted
conductedin inthe
themultibody
multibody
is essential to first verify the performance of the internal control loop. To evaluate the
environmentusing
environment using rollposition
position inputsfor fortrapezoidal,
trapezoidal,random,
random,and anddouble
doublelane lanechange
change
performance of theroll internal roll inputsangle control, a simulation is conducted in the multibody
scenarios.The
scenarios. Thetrapezoidal
trapezoidalsignal signalhas hasananangular
angularvelocity
velocityof of10 10deg/s
deg/swith withaaconvergence
convergence
environment using roll position inputs for trapezoidal, random, and double lane change
angle
angle of 6 deg,
of 6 deg,
scenarios. Theand and the random
the random
trapezoidal signal signal
signal
has isis a 1
anaangularHz Gaussian
1 Hz Gaussian noise
velocitynoise with
of 10with
deg/s a maximum
a maximum position
position
with a convergence
of 5
ofangle deg.
5 deg.of 6 deg, and the random signal is a 1 Hz Gaussian noise with a maximum position
Thesimulation
of 5 The
deg. simulationresults resultsare arepresented
presentedin inFigure
Figure11, 11,confirming
confirmingthat thatthetheroll
rollcommand
command
tracking
tracking error
Theerror is below
is below
simulation 0.66 deg.
0.66 deg.
results Furthermore,
Furthermore,
are presented for the
for the11,
in Figure model
model with a disturbance
with a that
confirming disturbance observer
observer
the roll command
(DOB),
(DOB),
tracking the
the maximum
maximum
error is below error
0.66isis
error 0.3040
0.3040
deg. deg,and
deg,
Furthermore,andthe the root
forroot meansquare
the mean
model square (RMS)
with a(RMS) errorisobserver
error
disturbance is0.0847
0.0847
deg
deg
(DOB),in the
in the case
thecase
maximum of the random
of the error
random signal.
signal.
is 0.3040 deg, These
These
and the values
values are
root are
mean smaller
smaller compared
squarecompared
(RMS) error to the
toisthe model
model
0.0847 deg
without
without
in the caseDOB,
DOB, which
of which
the hasaamaximum
has
random maximum
signal. error
error
These of
valuesof0.3078
0.3078 andan
and
are smaller ancompared
RMSerror
RMS errorto ofthe
of 0.0922.
0.0922.
model Although
Although
without
these
these results
DOB,results
which has indicate
indicate that the
that
a maximum the performance
performance
error of 0.3078 and with
with DOB
anDOBRMSisis better,
better,
error the difference
the
of 0.0922. difference
Althoughisisthesenot
not
significant
results indicate
significant enough
enough to
thattothe fully demonstrate
performance
fully demonstrate withthe the
DOB robustness
is better, the
robustness of DOB. This
difference
of DOB. is
This isisnot because the
significant
because the
simulation
enough to model
simulation model
fully presentedin
demonstrate
presented inFigure
the Figure 11was
robustness
11 was conducted
of conducted
DOB. iswithout
This without
because external disturbances,
the simulation
external model
disturbances,
presented in Figure 11 was conducted without external disturbances, representing condi-
tions when the vehicle is stationary state. However, in real driving situations, numerous
disturbances occur.
 force was applied to the center of mass. The simulation results under these conditions are
shown in Figure 12. Unlike the disturbance-free scenario, the graphs reveal a significant
difference in control performance. The RMS error decreased from 0.3335 to 0.2053, repre-
senting a 38% reduction, while the peak error decreased from 0.7002 to 0.5948, represent-
Actuators 2024, 13, 210 ing a 15% reduction. These results confirm the advantages of robust control provided by
11 of 15
the DOB.

Roll control result (W.O. DOB) Roll control result (W.O. DOB) Roll control result (W.O. DOB)

Angle [deg]
Angle [deg]

Angle [deg]
Roll control result (W. DOB) Roll control result (W. DOB) Roll control result (W. DOB)

Angle [deg]
Angle [deg]

Angle [deg]
Roll control error Roll control error Roll control error
Current [A]
Current [A]

Current [A]
W.O. DOB W.O. DOB W.O. DOB
W. DOB W. DOB W. DOB

Time [s] Time [s] Time [s]

(a) (b) (c)

Figure 11. Roll angle control results without external forces: (a) trapezoidal input; (b) random input;
Figure 11. Roll angle control results without external forces: (a) trapezoidal input; (b) random input;
and (c) double lane change input.
and (c) double lane change input.

To simulate various disturbances encountered during driving, a random external force

was applied to the center of mass. The simulation results under these conditions are shown
in Figure 12. Unlike the disturbance-free scenario, the graphs reveal a significant difference
Actuators 2024, 13, x FOR PEER in control performance. The RMS error decreased from 0.3335 to 0.2053, representing
REVIEW 12 of 15 a
38% reduction, while the peak error decreased from 0.7002 to 0.5948, representing a 15%
reduction. These results confirm the advantages of robust control provided by the DOB.

Roll control result (W.O. DOB) Roll control result (W.O. DOB) Roll control result (W.O. DOB)
Angle [deg]

Angle [deg]

Angle [deg]

Roll control result (W. DOB) Roll control result (W. DOB) Roll control result (W. DOB)
Angle [deg]

Angle [deg]

Angle [deg]

Roll control error Roll control error Roll control error

Current [A]

Current [A]

Current [A]

W.O. DOB W.O. DOB W.O. DOB

W. DOB W. DOB W. DOB

Time [s] Time [s] Time [s]

(a) (b) (c)

Figure 12. Roll angle control results with external forces: (a) trapezoidal input; (b) random input;
Figure 12. Roll angle control results with external forces: (a) trapezoidal input; (b) random input; and
and (c) double lane change input.
(c) double lane change input.
While multibody simulations are suitable for the lateral behavior in a static state, cre-
ating a dynamic driving environment is challenging. Therefore, a driving simulation tool
is additionally needed, and CarSim2019 is a useful co-simulation software that can be used
with Simulink. Factors such as lateral acceleration, lateral force, and driver’s steering dur-
ing driving situations are imported from CarSim to multibody to generate lateral behav-
ior. Conversely, the vehicle’s roll angle resulting from lateral behavior is imported into
 Time [s] Time [s]

(a) (b) (c)

Actuators 2024, 13, 210
Figure 12. Roll angle control results with external forces: (a) trapezoidal input; (b) random12input;
of 15
and (c) double lane change input.

While multibody
multibody simulations
simulationsarearesuitable
suitableforfor
thethe
lateral behavior
lateral behaviorin ain
static state,state,
a static cre-
ating a dynamic
creating a dynamic driving environment
driving environment is challenging. Therefore,
is challenging. a driving
Therefore, simulation
a driving simulationtool
tool is additionally
is additionally needed,
needed, and CarSim2019
and CarSim2019 is a useful
is a useful co-simulation
co-simulation software
software thatbe
that can can be
used
used
with with Simulink.
Simulink. Factors
Factors such such as lateral
as lateral acceleration,
acceleration, laterallateral
force,force, and driver’s
and driver’s steering
steering dur-
during driving
ing driving situations
situations are imported
are imported from from
CarSim CarSim to multibody
to multibody to generate
to generate lateral
lateral behav-
behavior. Conversely, the vehicle’s roll angle resulting from lateral behavior
ior. Conversely, the vehicle’s roll angle resulting from lateral behavior is imported into is imported
into CarSim
CarSim based based
on theonvalues
the values obtained
obtained from from the multibody.
the multibody. The entire
The entire co-simulation
co-simulation envi-
environment is shown
ronment is shown in Figure
in Figure 13. 13.

Figure 13. Co-simulation diagram of

Figure 13. of the
the system.
system.

The driving scenarios for the tilting vehicle include double lane changes and step
steer conditions. The maximum driving speed of the tilting vehicle is set as 60 km/h, and
the simulation for a single lane change is conducted under constant speed conditions of
50 km/h. In the step steer simulation, 120 degrees of steering angle is applied at a constant
driving speed. To compare the effect of the tilting control, two vehicle control methods
were applied. For regular vehicles without tilting, the plate used to rotate the vehicle is
fixed to the vehicle’s body. The motor shaft for tilting motion is fixed at zero, and this
model operates like a conventional vehicle with a standard suspension. The model with the
tilting control is designed to actively create a roll angle based on the controller suggested in
Figure 7.
Roll stability of the vehicle is evaluated by the driver’s lateral acceleration, LTR,
and ZMP. From the perspective of the roll angle of the vehicle, drivers feel safe when
tilting control is applied as the vehicle leans inside of the cornering radius. In contrast,
without control, the vehicle leans outward of the cornering radius, which can cause rollover.
Similarly, the driver’s lateral acceleration significantly decreases when controlled, making
the driver feel more stable. In terms of LTR, the roll stability is significantly reduced by
more than 40% in RMS compared to the uncontrolled scenario. The increase of stability is
seen in both the single lane change driving, shown in Figure 14, and the constant radius
cornering driving shown in Figure 15. These results show that the lateral loads on the
vehicle are well distributed, reducing the risk of one side lifting off the ground. When
evaluating stability based on the ZMP, the results show that the ZMP of the vehicle with
tilting control is stably positioned near the center. This indicates that the tilting control
effectively contributes to stability in terms of ZMP.

Actuators 2024, 13, 210
in both the single lane change driving, shown in Figure 14, and the constant radius cor-
nering driving shown in Figure 15. These results show that the lateral loads on the vehicle
are well distributed, reducing the risk of one side lifting off the ground. When evaluating
stability based on the ZMP, the results show that the ZMP of the vehicle with tilting con-
trol is stably positioned near the center. This indicates that the tilting control effectively
13 of 15
contributes to stability in terms of ZMP.

Lateral Acceleration Lateral load transfer ratio

]
2
Accel [m/s

LTR
W.O. Control
W. Control

Roll angle Zero moment point

Angle [degree]

Position [m]
Actuators 2024, 13, x FOR PEER REVIEW 14 of 15
Time [s] Time [s]

Figure 14. The simulation results of the double lane change.

Figure 14. The simulation results of the double lane change.
Lateral Acceleration Lateral load transfer ratio
]
2
Accel [m/s

LTR

W.O. Control
W. Control

Roll angle Zero moment point

Angle [degree]

Position [m]

Time [s] Time [s]

Figure 15. The simulation results of step steer.

Figure 15. The simulation results of step steer.
6. Conclusions
6. Conclusions
This paper investigates the roll stability control of a narrow tilting vehicle (NTV) with
This
a novel paperstructure,
tilting investigates the roll astability
leveraging control
real vehicle of a narrow
model tilting
to enhance its vehicle (NTV)
practical with
relevance.
a novel tilting
Accurate structure,dynamic
and detailed leveraging a real vehicle
equations model to
were derived enhance
using the 3Dits model
practical
of relevance.
the actual
Accurate
vehicle, andandthrough
detailed parameter
dynamic equations
matching, were
thederived using
simulation the 3D
model model
was of the actual
implemented to
vehicle, and through parameter matching, the simulation model was implemented to rep-
replicate real vehicle conditions. As a control structure, the cascade structure consisting
licate real vehicle
of an inner conditions.
robustness controlAs a control
loop and anstructure, the cascade
outer stability controlstructure
loop wasconsisting
adopted.of an
This
inner robustness
approach securescontrol
robust loop
angleand an outer
control in thestability control
inner loop loop
while was adopted.
effectively managingThis roll
ap-
proach secures robust angle control in the inner loop while effectively managing roll sta-
bility in the outer loop. The proposed roll angle controller simplifies the model to reduce
instability, and the incorporation of a disturbance observer (DOB) mitigates the effects of
external disturbances and model errors, demonstrating an improvement in control per-
formance. Moreover, this study enhances the evaluation of vehicle roll stability by intro-
Actuators 2024, 13, 210 14 of 15

stability in the outer loop. The proposed roll angle controller simplifies the model to reduce
instability, and the incorporation of a disturbance observer (DOB) mitigates the effects
of external disturbances and model errors, demonstrating an improvement in control
performance. Moreover, this study enhances the evaluation of vehicle roll stability by
introducing the zero moment point (ZMP) criterion, traditionally used in pedestrian robot
research, alongside the commonly utilized lateral tire force ratio (LTR). By applying both
LTR and ZMP as stability indicators, the research provides a comprehensive assessment of
the tilting vehicle’s roll stability from multiple perspectives.

Author Contributions: Conceptualization, H.C.; methodology, S.L. and K.N.; software, S.L.; vali-
dation, S.L., H.C. and K.N.; formal analysis, S.L. and K.N.; investigation, S.L. and H.C.; resources,
H.C.; data curation, S.L.; writing—original draft preparation, S.L.; writing—review and editing, K.N.;
visualization, S.L.; supervision, K.N.; project administration, H.C.; funding acquisition, K.N. All
authors have read and agreed to the published version of the manuscript.
Funding: This work was supported by the National Research Foundation of Korea (NRF) grant
funded by the Korean government (MSIT) (No. 2022R1C1C1011785) and by the Korea Evalua-
tion Institute of Industrial Technology (KEIT) grant funded by the Korean government (MOTIE)
(No. 20023833).
Data Availability Statement: The original contributions presented in the study are included in the
article, further inquiries can be directed to the corresponding author/s.
Conflicts of Interest: Author Hyeonseok Cho was employed by the company Hyundai (Kia Namyang)
Research and Development Center. The remaining authors declare that the research was conducted
in the absence of any commercial or financial relationships that could be construed as a potential
conflict of interest.

References
1. Kockelman, K.M.; Zhao, Y. Behavioral Distinctions: The Use of Light-Duty Trucks and Passenger Cars. J. Transp. Stat. 2000, 3,
47–60. [CrossRef]
2. Grzegożek, W.A. Analysis of a Construction Solution for Narrow Track Vehicle (NTV). IOP Conf. Ser. Mater. Sci. Eng. 2018, 421,
022012. [CrossRef]
3. Poelgeest, A. The Dynamics and Control of a Three-Wheeled Tilting Vehicle. Ph.D. Thesis, University of Bath, Bath, UK, 2011.
4. Chatterjee, M.; Kale, M.; Chaudhari, B.N. Mathematical Modelling of Chassis Dynamics of Electric Narrow Tilting Three Wheeled
Vehicle. In Proceedings of the 2015 Annual IEEE India Conference (INDICON), New Delhi, India, 17–20 December 2015; pp. 1–6.
5. Chiou, J.-C.; Chen, C.-L. Modeling and Verification of a Diamond-Shape Narrow-Tilting Vehicle. IEEE/ASME Trans. Mechatron.
2008, 13, 678–691. [CrossRef]
6. Larish, C.; Piyabongkarn, D.; Tsourapas, V.; Rajamani, R. A New Predictive Lateral Load Transfer Ratio for Rollover Prevention
Systems. IEEE Trans. Veh. Technol. 2013, 62, 2928–2936. [CrossRef]
7. Ding, X.; Wang, Z.; Zhang, L. A Vehicle Rollover Prediction System Based on Lateral Load Transfer Ratio. In Proceedings of the
2020 Chinese Automation Congress (CAC), Shanghai, China, 6–8 November 2020; pp. 7256–7261.
8. Tsourapas, V.; Piyabongkarn, D.; Williams, A.C.; Rajamani, R. New Method of Identifying Real-Time Predictive Lateral Load
Transfer Ratio for Rollover Prevention Systems. In Proceedings of the 2009 American Control Conference, Saint Louis, MO, USA,
10–12 June 2009; pp. 439–444.
9. Shin, D.; Woo, S.; Park, M. Rollover Index for Rollover Mitigation Function of Intelligent Commercial Vehicle’s Electronic Stability
Control. Electronics 2021, 10, 2605. [CrossRef]
10. Kajita, S.; Kanehiro, F.; Kaneko, K.; Fujiwara, K.; Harada, K.; Yokoi, K.; Hirukawa, H. Biped Walking Pattern Generation by Using
Preview Control of Zero-Moment Point. In Proceedings of the 2003 IEEE International Conference on Robotics and Automation
(Cat. No.03CH37422), Taipei, Taiwan, 14–19 September 2003; Volume 2, pp. 1620–1626.
11. Erbatur, K.; Okazaki, A.; Obiya, K.; Takahashi, T.; Kawamura, A. A Study on the Zero Moment Point Measurement for Biped
Walking Robots. In Proceedings of the 7th International Workshop on Advanced Motion Control. Proceedings (Cat. No.02TH8623),
Maribor, Slovenia, 3–5 July 2002; pp. 431–436.
12. Rajamani, R.; Piyabongkarn, D.; Tsourapas, V.; Lew, J.Y. Parameter and State Estimation in Vehicle Roll Dynamics. IEEE Trans.
Intell. Transp. Syst. 2011, 12, 1558–1567. [CrossRef]
13. Rajamani, R.; Gohl, J.; Alexander, L.; Starr, P. Dynamics of Narrow Tilting Vehicles. Math. Comput. Model. Dyn. Syst. 2003, 9,
209–231. [CrossRef]
14. Piyabongkarn, D.; Keviczky, T.; Rajamani, R. Active direct tilt control for stability enhancement of a narrow commuter vehicle.
Int. J. Automot. Technol. 2004, 5, 77–88.
Actuators 2024, 13, 210 15 of 15

15. Kidane, S.; Rajamani, R.; Alexander, L.; Starr, P.J.; Donath, M. Development and Experimental Evaluation of a Tilt Stability Control
System for Narrow Commuter Vehicles. IEEE Trans. Control. Syst. Technol. 2010, 18, 1266–1279. [CrossRef]
16. Tang, C.; Ataei, M.; Khajepour, A. A Reconfigurable Integrated Control for Narrow Tilting Vehicles. IEEE Trans. Veh. Technol. 2019,
68, 234–244. [CrossRef]
17. Tang, C.; Khajepour, A. Integrated Stability Control for Narrow Tilting Vehicles: An Envelope Approach. IEEE Trans. Intell. Transp.
Syst. 2021, 22, 3158–3166. [CrossRef]
18. Tang, C.; Khajepour, A. Narrow Tilting Vehicles: Mechanism, Dynamics, and Control; Morgan & Claypool Publishers: San Rafael, CA,
USA, 2019; ISBN 978-3-031-01501-4.
19. Ren, Y.; Dinh, T.Q.; Marco, J.; Greenwood, D.; Hessar, C. Nonlinearity Compensation Based Tilting Controller for Electric Narrow
Tilting Vehicles. In Proceedings of the 2018 5th International Conference on Control, Decision and Information Technologies
(CoDIT), Thessaloniki, Greece, 10–13 April 2018; pp. 1085–1090.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.