Recei ed 28 Oc obe 2024, accep ed 25 No embe 2024, da e o publica ion 29 No embe 2024,
da e o cu en e sion 9 Decembe 2024.
Digi al Objec Iden i ie 10.1109/ACCESS.2024.3509226
Towa d Digi al Twin o O -Road Vehicles Using
Robo Simula ion F amewo ks
ARIANNA RANA 1,2, ANTONIO PETITTI 1, ANGELO UGENTI 2, ROCCO GALATI 2,
GIULIO REINA 2, AND ANNALISA MILELLA 1
1Ins i u e o In elligen Indus ial Technologies and Sys ems o Ad anced Manu ac u ing (STIIMA), Na ional Resea ch Council o I aly (CNR), 70126 Ba i, I aly
2Depa men o Mechanics, Ma hema ics, and Managemen , Poly echnic Uni e si y o Ba i, 70126 Ba i, I aly
Co esponding au ho : An onio Pe i i ([email p o ec ed].i )
This wo k was pa ially unded by he ollowing p ojec s: AgRibo -Ha nessing Robo ics, XR/AR, and 5G o a New E a o Sa e,
Sus ainable, and Sma Ag icul u e, Eu opean Union’s Ho izon Eu ope esea ch and inno a ion p og amme unde g an ag eemen
(No.101183158); gi ing Smell sense To Ag icul u al Robo ics (STAR), ERA-NET COFUND ICT AGRI-FOOD (G an No. 45207); CNR
Dipa imen o di Ingegne ia, ICT e Tecnologie pe l’Ene gia e i T aspo i (DIITET) p ojec DIT.AD022.207, STRIVE-le Scienze pe le
TRansizioni Indus iale, Ve de ed Ene ge ica (FOE 2022), sub ask ac i i y Ag o-Sensing2. This publica ion is also pa o he p ojec No d
O es Digi ale e Sos enibile (NODES), which has ecei ed unding om he Minis e o dell’Uni e si à e della Rice ca (MUR)–M4C2 1.5 o
Piano Nazionale di Rip esa e Resilienza (PNRR) unded by he Eu opean Union–Nex Gene a ionEU (G an ag eemen no. ECS00000036).
ABSTRACT Digi al wins p o ide a powe ul ool o es ing and main aining p oduc s and p ocesses
in se e al applica ion ields, including manu ac u ing, sma ci ies, heal hca e, and ag icul u e, aiming o
op imize ope a ional e iciency, esou ce usage, and planning accu acy. Resea ch p esen ed in his pape
deals wi h he de elopmen o he digi al e sion o o - oad ehicles. Two di e en obo simula ion
amewo ks a e in es iga ed. The i s one is based on Gazebo, an open-sou ce 3D obo ics simula o , o es
and alida e he algo i hms de eloped in he ROS amewo k; he second one adop s he ehicle mechanical
assembly in MSC Adams, a mul ibody modeling so wa e used o s udy he dynamics o complex mechanical
sys ems. Bo h models a e de eloped o a acked obo ha uses an inno a i e a icula ed passi e suspension
sys em on ei he side ha allows each g ound wheel o mo e independen ly wi h espec o he ehicle
body, p o iding ema kable adap abili y o i egula e ain. In addi ion, a Gazebo model is de eloped o
a ou -wheel d i e/s ee ing obo , including obo senso s such as GNSS, IMU, and isual senso s and a
model o a ypical ag icul u al en i onmen (i.e., a ineya d). The pape p esen s he de ails o model design
and implemen a ion while in es iga ing he bes choice in de eloping he digi al win o o - oad ehicles
ope a ing in he ield. Addi ionally, an ag icul u al scena io has been selec ed as a use case o acili a e he
e alua ion o he analyzed amewo ks. Ou indings demons a e ha he Gazebo amewo k could se e
as a sui able obo simula ion amewo k o c ea ing digi al wins o ehicles, p o ided i inco po a es
eal- ime senso measu emen s designed o iden i ying soil-wheel in e ac ion dynamics. In con as , he
mul ibody model p o ides a highe - ideli y dynamic model, including ack- e ain in e ac ion. Despi e i s
ad an ages, his model has subs an ially highe compu a ional cos s, which limi s i s applicabili y o eal-
ime simula ions, making i less easible o p ac ical use in he ield.
INDEX TERMS Ag icul u al obo ics, digi al win, ield obo ics, mobili y o e de o mable e ain,
mul ibody modeling and simula ion, obo simula ion amewo ks.
I. INTRODUCTION
A Digi al Twin (DT) [1],[2] is a digi al, ypically com-
pu a ional, ep esen a ion o a p ocess o a angible en i y
like a obo ic ehicle. C ea ing a DT in ol es a i icial
The associa e edi o coo dina ing he e iew o his manusc ip and
app o ing i o publica ion was Lukasz Wisniewski .
in elligence, machine lea ning, and/o so wa e analy ics
alongside physics-based modeling o cons uc a digi al
simula ion model capable o eplica ing he cha ac e is ics
and ac ions o he physical coun e pa [3]. To achie e his,
he model o en equi es ongoing aining and e inemen
using senso da a and a p obabilis ic compu a ional ame-
wo k ha accommoda es unce ain ies [4]. Digi al wins
VOLUME 12, 2024
2024 The Au ho s. This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial-NoDe i a i es 4.0 License.
Fo mo e in o ma ion, see h ps://c ea i ecommons.o g/licenses/by-nc-nd/4.0/ 178047
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
a e inc easingly p e alen in a ious applica ion ields such
as indus y, manu ac u ing [5], ae ospace [6], sma ci ies,
heal hca e [7], ag icul u e [8], e c [9]. In [10], a ocus on
he p ecise s a e synch oniza ion be ween he DT o a mobile
obo and he physical asse is gi en. The p oposed app oach
deals wi h he co ec posi ion e alua ion in he sma ac o y
con ex . In [11], on he o he hand, he DT has been used
o es localiza ion and na iga ion algo i hms unde di e en
ope a ing condi ions o an au onomous ehicle in a p oduc-
ion hall. Mo eo e , ano he app oach o he DT has been
p oposed in [12], whe e he sys em is composed o a human
ope a o equipped wi h an exoskele on obo and a i ual
eali y de ice. In o ma ion be ween he a ious pa s o he
sys em is exchanged in eal- ime o accele a e he ins uc ion
o obo ic asks in p oduc ion en i onmen s. In [13], a digi al
win o a comme cial VTOL con e iplane ae ial ehicle has
been de eloped in Gazebo, enabling compa ison o ake-o ,
ho e ing, and landing maneu e s wi h and wi hou he wind
physics model. Howe e , signi ican ad ancemen s a e s ill
needed, especially in add essing challenges ela ed o ou doo
en i onmen s [14],[15],[16]. In [17], digi al wins a e
p oposed as a sui able ool o i ual p o o yping and damage
de ec ion in space explo a ion o e applica ions. Howe e ,
as his is a p elimina y easibili y s udy, no alida ion and
e i ica ion ha e been p esen ed. In [18], he digi al win
o a pas u e was de eloped, which is mainly o echnical
e i ica ion and da a acquisi ion based on machine ision.
Fu he mo e, [19] p oposes a DT p o o ype in a sma a m.
In pa icula , he DT se es as a subs i u e o he physical
sys em du ing he de elopmen phase o a sma a ming
applica ion.
A DT comp ises h ee componen s: he physical asse ,
he co esponding i ual coun e pa , and he linking da a
exchange. This pape ocuses on de eloping a i ual
ep esen a ion o o - oad obo ic ehicles. Al hough many
3D simula o s o simula e obo s in ou doo en i onmen s
unde di e en condi ions a e a ailable [20],[21],[22],
we explo e wo dis inc simula ion en i onmen s. The i s
employs ROS and Gazebo 11 o es and alida e algo i hms
de eloped wi hin he ROS amewo k, while he second
le e ages he ehicle mechanical assembly wi hin MSC
Adams. Gazebo [23] is an open-sou ce simula ion so wa e
bo n in 2002 a he Uni e si y o Sou he n Cali o nia as
pa o a Ph.D. esea ch p ojec . I is one o he mos
amous simula o amewo ks used by he obo ic commu-
ni y [21],[24],[25]. I is a powe ul so wa e ool ha o e s
a comp ehensi e 3D simula ion en i onmen o c ea ing
and simula ing a ious obo ic sys ems, en i onmen s, and
scena ios. One o i s key ad an ages is he abili y o es
and alida e algo i hms, obse e obo ic beha io s, and
mo e in a sa e y en i onmen [26],[27]. Fu he mo e,
Gazebo allows us o model obo s, speci ying hei physical
p ope ies, i.e., collision, isual appea ance, and ine ia.
I also p o ides a wide ange o senso s ha can be added o
he obo model [28], p o iding pe cep ion capabili ies and
allowing he e alua ion and expe imen a ion o pe cep ion,
planning, and con ol algo i hms. Gazebo is o en used
alongside obo ic amewo ks [29]. Speci ically, i has a
s ong in eg a ion wi h ROS, whe eas Gazebo is employed o
es and alida e obo ics applica ions, which a e de eloped
in ROS, o example, in a simula ed en i onmen be o e
deploymen in he eal wo ld. The Gazebo ROS package
acili a es he communica ion be ween ROS and Gazebo.
This package ac s as a b idge, enabling he simula ion o
ROS-based obo ic sys ems wi hin he Gazebo en i onmen .
The e o e, de elope s can combine hei ROS code wi h
Gazebo, allowing hem o es and assess he pe o mance
o hei de eloped algo i hms. MSC Adams, on he o he
hand, is a so wa e package used o mul i-body dynamics
simula ion. I is a widely used ool in enginee ing, pa icula ly
in he ields o mechanical, au omo i e, ae ospace, and o he
indus ies whe e he beha io o in e connec ed mo ing
pa s needs o be analyzed. Mo eo e , he buil -in oolki
Adams T acked Vehicle (ATV) is designed speci ically o
simula ing and analyzing he dynamics o acked ehicles.
This simula ion en i onmen also includes ools o simula e
he in e ac ion be ween he acks and di e en ypes
o e ain, including so soils. The so wa e also allows
o he simula ion o suspension sys ems, including ack
ensioning de ices. Because o hese ea u es, MSC Adams
is able o simula e he dynamics o an o - oad obo wi h
much highe accu acy han Gazebo. On he o he hand,
because o he highe compu a ional bu den and lack o
simula ed ex e ocep i e pe cep ion, he mul ibody model
migh be challenging o implemen o a eal- ime digi al win
applica ion.
The aim is o de ine he DT o an o - oad obo ic ehicle o
exploi he ad an age o he wo dis inc obo ics simula o s
while in es iga ing he possibili y o be con inuously upda ed
using senso y da a. Speci ically, he con ibu ion o his pape
is wo old:
(i) he Gazebo isual models o wo cus om-buil o - oad
ehicles ha e been de eloped;
(ii) an ex ensi e compa ison be ween Gazebo and MSC
Adams has been unde aken o de e mine he mos
sui able amewo k o de eloping he digi al win o
o - oad ehicles ope a ing in ield condi ions.
The pape is o ganized as ollows. Sec ion II p o ides
an o e iew o he obo ic pla o ms. Sec ion III de ails he
de elopmen o isual models o he wo obo s c ea ed
in Gazebo, while Sec. IV b ie ly in oduces he mul ibody
model de eloped in MSC Adams. Sec ion Vdesc ibes he
es s conduc ed using he wo simula o s, ocusing on a
speci ic ag icul u al obo ics use case, and, inally, Sec. VI
ou conclusions om he in es iga ion a e d awn.
II. ROBOTIC PLATFORMS
In his pape , wo cus om-buil o - oad ehicles, namely
a acked obo and a wheeled obo , a e conside ed,
espec i ely, a ailable a Poli ecnico o Ba i, I aly, and a
CNR-STIIMA, I aly.
178048 VOLUME 12, 2024
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
The ully unc ioning p o o ype o he all- e ain acked
o e named Polibo [30] is shown in Fig. 1, while Fig. 2
p esen s a CAD ende ing wi h he indica ion o he o e all
dimensions highligh ing also how he ubbe ack is w apped
a ound he d i e sp ocke a he op, he idle wheel a he
on , and he ou g ound wheels. The dis inc i e ea u e
o he acked locomo ion sys em includes an a icula ed
passi e suspension sys em on each side, enabling indi idual
mo emen o each g ound wheel and imp o ing adap abili y
o une en e ain. Mo eo e , he ou oad wheels dis ibu e
he ehicle’s weigh ac oss he con ac pa ch.
FIGURE 1. The Polibo equipped wi h a mul i-senso sys em.
FIGURE 2. Isome ic CAD iew o he Polibo wi h he indica ion o he
main dimensions.
As epo ed in Table 1, Polibo is based on he Robo
Ope a ing Sys em (ROS) and is equipped wi h wo 350 W
24 VDC b ush mo o s, each coupled wi h an angula gea box
wi h a 30:1 educing a io. The d i ing mo o and he angula
gea box ha e been selec ed o deli e a peak o que o 35Nm
TABLE 1. Technical speci ica ions o Polibo .
a 80 RPM wi h a maximum payload o 50 Kg in addi ion
o he possibili y o o e come slopes up o 40 deg ees.
The unde ca iage comp ises wo sub ames a ached o
he main body ame ia wo b acke s. A ached o he
sub ames ia e olu e join s, ou swing a ms can o a e
independen ly ca ying a oad wheel each. The powe ain
componen s a e housed wi hin he cen al chassis, while he
ou pu sha is di ec ly connec ed o he d i e sp ocke s. The
ma e ial used o he bel is a composi e o mel ed na u al
ubbe and ibe glass, wi h an inne s eel co d. All wheels
a e cons uc ed om UHMW, a high-densi y polye hylene
known o i s excep ional wea esis ance. Polibo ea u es
an uppe la su ace designed o accommoda e addi ional
de ices and senso s such as came as, LiDARs, IMUs,
o lap ops. Embedded in he main body, an In el i7 compu e
has 16 GB RAM DDR, 256 GB SSD, Wi-Fi, and Blue oo h
in e aces. The p ima y ope a ing sys em on he compu e is
Ubun u, which is u ilized o unning ROS and commanding
he mo o con olle s ia an RS232 se ial po . The p ima y
powe sou ce comp ises a 24 VDC 30 Ah LiPo ba e y pack
p o iding a s anda d au onomy o app oxima ely 3 hou s.
The second obo a chi ec u e conside ed in his wo k,
he e e e ed o as CNRbo , is a ou -wheel d i ing and
ou -wheel s ee ing ehicle wi h a ocke suspension sys em
based on wo swing a ms able o o a e a ound a ho izon al
axis while ancho ed o a se o dampe s; his design is
essen ial o keep he main ame as s able as possible by
educing ib a ions e en on ough e ains by enhancing he
e iciency o he senso s (i.e., came as, lase s, and IMUs)
ins alled on he ehicle. The ehicle p o o ype is shown
in Fig. 3, whe eas he CAD model is epo ed in Fig. 4.
CNRbo , is a ROS wheeled pla o m p o iding an Acke mann
s ee ing geome y, which can be equipped wi h a ious
p op iocep i e and ex e ocep i e senso s ha p o ide highly
accu a e in o ma ion abou he en i onmen and he ehicle’s
ope a ional condi ions. CNRbo has been designed wi h a
o al o ou b ushless d i ing mo o s and ou plane a y
gea boxes wi h a gea a io o 4:1 o p ope ly adjus bo h
he ou pu o que and eloci y, p o iding up o 40 Nm
wi h an o e all peak o que o abou 50 Nm when spinning
a 80 RPM. In his case, he maximum payload is 100 Kg
wi h a capabili y o o e come slopes up o 40 deg ees.
To enhance he s abili y and he pe o mance o he mo o s
making hem each he desi ed speed wi h minimal delay and
VOLUME 12, 2024 178049
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
TABLE 2. Technical speci ica ions o CNRbo .
o e shoo , a PID con olle has been implemen ed in bo h
obo s o con ol he ehicle in Closed Speed Loop by using
he da a coming om he mo o s encode s. Thanks o i s
10-inch o - oad wheels, i can be used bo h o indoo and
ou doo applica ions, and i s high payload allows use s o add
addi ional de ices and lap ops. The la ge g ound clea ance
enables he ehicle o ope a e in e ains wi h high densi y
o obs acles such as ocks and deb is. The ea panel o e s
se e al powe ou pu s in o de o connec ex e nal senso s and
de ices. This wheeled mobile obo (WMR) is elec ically
powe ed by wo ully isola ed AGM ba e ies wi h a powe
ou pu a ing o 2400 W. Table 2summa izes he main elec ic
and mechanical ea u es o he CNRbo .
Finally, bo h obo s can be emo ely con olled using a
s anda d joys ick con olle ha ope a es ia wi eless o adio
echnologies. The ope a ing ange o he de ice depends
on he communica ion echnology used. Wi h wi eless
echnology, he ange is app oxima ely 10 me e s, while adio
echnology allows o a signi ican ly ex ended ange.
III. ROS-GAZEBO MODEL OF THE ROBOTIC PLATFORMS
This sec ion desc ibes he implemen a ion o he isual model
o he wo ehicles p esen ed in Sec. II and he de elopmen
o a simula ed en i onmen .
A. UNIFIED ROBOTIC DESCRIPTION FORMAT
In Gazebo, a simula ed model o a obo is gene a ed using he
Uni ied Robo ic Desc ip ion Fo ma (URDF) [31]. URDF is
an XML ile o ma widely used in he ROS (Robo Ope a ing
Sys em [32],[33], a amewo k o de eloping obo ics
sys ems) en i onmen as i p o ides a means o desc ibe he
kinema ics, dynamics, and isual aspec s o he obo model.
In gene al, a URDF ile is used o desc ibe a obo as a ee
o links connec ed by join s. The links e e o each physical
componen o he obo , while he join s ep esen how a link
mo es ela i e o ano he link by de ining he loca ion o he
links in space.
The isual model can be c ea ed using basic geome ic
shapes, such as cubes, cylinde s, sphe es, and mo e, o spe-
ci ic meshes o a mo e de ailed and ealis ic ep esen a ion
o he model i sel .
Fo his s udy, he ollowing s eps a e pe o med. Fi s ,
a CAD assembly o he obo is buil in SolidWo ks (SW)
FIGURE 3. The obo CNRbo wi h i s senso sui e.
FIGURE 4. Isome ic CAD iew o he all- e ain o e CNRbo wi h he
indica ion o he main dimensions.
(a 3D design so wa e sui e). Then, he model equi ed by
he ROS-Gazebo amewo k is gene a ed by employing a
so wa e ool called SolidWo ks o URDF expo e [34]. The
178050 VOLUME 12, 2024
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
so wa e ool enables he con e sion o he CAD model and
gene a es a ROS-like package ha inco po a es in o ma ion
abou meshes, ma e ials, ex u es, and he URDF ile. The
URDF ile includes he di e en elemen s ( igid bodies)
composing he ehicle. I is c i ical o de ine he elemen s
in he igh o de o co ec ly build he ee ha desc ibes
he kinema ic ela ionships be ween he componen s o he
ehicle.1Fu he mo e, he URDF ile gene a ed by he ool
con ains in o ma ion ega ding he isual ep esen a ion o
he CAD model con e ed in o he speci ic o ma STL
(STe eo Li hog aphy in e ace) o isualiza ion in ROS.
The ool also gene a es collision geome y in o ma ion
and physical p ope ies such as mass, ine ia, and o he
pa ame e s equi ed by physics engines like Gazebo, which
will be discussed in mo e de ail in Sec. III-C. The s eps
ollowed o ob ain he digi al model o he ehicles a e as
ollows:
1) Design he o e wi h CAD so wa e;
2) Open ‘‘Solidwo ks o URDF expo e ’’ en i onmen ;
3) Selec he i s link, called he base link, i.e. he main
componen om which all child links a e de i ed. Each
URDF has solely a base link;
4) Selec he necessa y numbe o child en ha each pa en
link will ha e;
5) Fo each pa en -child connec ion, assign he ype o
join (e.g., e olu e, p isma ic, ixed, e c.) ha connec s
hem. Fo each join , i is necessa y o se se e al
p ope ies, such as posi ion limi s, eloci y, e o , axis
o o a ion/ ansla ion (only o mo able join s), e c.
6) Expo he URDF ile and associa ed iles, such as
meshes, ma e ials, and ex u es equi ed o he isual
and physical ep esen a ion o he obo .
I is impo an o gi e a unique name o each link and each
join . In Fig. 5, he low cha ep esen ing he s eps used o
ob ain he digi al model o he ehicles is shown.
B. VISUAL ROBOT MODELS USING URDF
Since Polibo and CNRbo a e cus omized obo s, isual
models ha e been c ea ed ollowing he p ocedu e desc ibed
in Sec. III-A. The URDF models o he Polibo and CNRbo ,
espec i ely, a e desc ibed in he emainde o his sec ion.
As s a ed in Sec. II, he Polibo is a acked obo [35],[36].
Howe e , o ease o implemen a ion, since a acked obo
employs a skid s ee ing mechanism, each ack is simula ed
as wo wheels, which a e con olled wi h he same eloci y
e e ence [37]. Mo eo e , de ining he ansmission a ibu e
o each non- ixed join is necessa y o simula e i p ope ly.
Thus, ou ansmission a ibu es ha e been de ined o he
Polibo , one o each wheel. Finally, o comple e he model
and enable he obo o mo e, a con olle [38] o skid s ee
d i ing is se o con ol he wheels’ eloci y. In Fig. 6, he
URDF model o he Polibo in he RViz [39] en i onmen is
shown. RViz is a 3D isualiza ion ool in eg a ed in o ROS
1This ansla es in o he TF (T ansFo m) ee o he coo dina e ans o -
ma ions in he ROS en i onmen .
FIGURE 5. Flow cha o he me hodology used o de eloping he URDF
model.
ha enables isualiza ion o he obo model wi h i s links
and join s and p o ides an in e ac i e, eal- ime iew o he
en i onmen a ound he obo . As can be obse ed, he obo
is equipped wi h a simula ed senso sui e moun ed on boa d
ha ep esen s he eal senso s. The senso is conside ed an
ex ension o he obo i sel , so he geome ic shape o he
senso can be de ined wi hin he URDF ile o he obo i sel .
Mo eo e , he simula ed senso emula es he beha io o he
eal senso , ex ac ing in o ma ion abou he en i onmen and
in e ac ing wi h ROS. Speci ically, he Polibo is equipped
wi h an In el Realsense D435 came a, a GPS ecei e , an IMU
senso , and a Velodyne LiDAR, which enable he pe cep ion
o he su ounding i ual en i onmen . Senso da a, such as
poin clouds, came a images, lase scans, and mo e, can be
isualized in RViz.
On he o he hand, he CNRbo has ou d i ing and
s ee ing wheels [40]; hus, ou a ibu es o he s ee ing
angle and ou a ibu es o he wheel o a ion ha e been
de ined. Consequen ly, a con olle mus be de ined o enable
he obo o mo e in he en i onmen and simul aneously
con ol all he wheels. Fo ease o implemen a ion, he skid
s ee ing d i e is used o his obo . Fig. 7depic s he URDF
model o he CNRbo in RViz, equipped wi h a simula ed
senso sui e composed o an In el Realsense D435 came a,
a GPS ecei e , and an IMU senso .
C. SIMULATED ENVIRONMENT
This sec ion in oduces a Gazebo simula ed en i onmen .
F om now on, we will e e o he simula ed en i onmen
as he wo ld. C ea ing a wo ld in Gazebo means c ea ing an
SDF (Simula ion Desc ip ion Fo ma ) model, which allows
o he desc ip ion and de ini ion o di e en en i ies, such
VOLUME 12, 2024 178051
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
FIGURE 6. RViz isualiza ion o he URDF model o he Polibo wi h he
simula ed senso sui e moun ed on boa d. The obo is equipped wi h he
ollowing senso s: an In el Realsense D435 came a, depic ed as he g ey
elemen moun ed a he op; a GPS ecei e , ep esen ed by he blue
cube; an IMU senso , ep esen ed by he o ange cube; a Velodyne LiDAR,
ep esen ed by he black cylinde on he chassis.
FIGURE 7. RViz isualiza ion o he URDF model o he CNRbo wi h he
simula ed senso sui e moun ed on boa d. The obo is equipped wi h he
ollowing senso s: an In el Realsense D435 came a, depic ed as he g ey
elemen moun ed a he op; a GNSS ecei e , ep esen ed by he blue
cube; an IMU senso , ep esen ed by he o ange cube.
as obo s, senso s, objec s wi hin a speci ic en i onmen , and
mo e. URDF and SDF a e XML-based ile o ma s used in he
obo ics en i onmen , bu hey se e di e en pu poses. The
i s desc ibes he obo ’s s uc u e, including join s, links,
isual appea ance, and collision p ope ies; he second, on he
o he hand, is dedica ed o desc ibing he en i e simula ed
en i onmen in which he obo ope a es. I may include
mul iple obo models, senso s, physical p ope ies o he
wo ld, and o he de ails. Fig. 8illus a es he dis inc ion
be ween URDF and SDF.
FIGURE 8. Gazebo scene ea u ing a compa ison o a URDF (yellow
squa e) and SDF ( ed squa e).
Fo con enience, in his wo k, he obo and senso models
a e de ined in he URDF ile because he SolidWo ks ool
gene a ed he obo model, as explained in Sec. III-A. The
SDF model comp ises h ee main componen s: links, join s,
and plugins. The link con ains in o ma ion abou a speci ic
model elemen ’s isual and physical p ope ies. Speci ically,
i desc ibes how he link appea s, i.e., he shape o he
mesh, and he collision p ope ies used o collision-checking
pu poses. Mo eo e , i includes he dynamical p ope ies,
such as mass, cen e o mass, and momen s o ine ia, and
he su ace p ope ies, such as ic ion, bounciness, and mo e.
On he o he hand, a join speci ies he connec ion be ween
wo links, enabling he c ea ion o complex a icula ed
objec s. Senso s a e conside ed as a pa o he obo model.
Consequen ly, i is necessa y o c ea e a link ep esen ing he
geome ical shape o which he senso is a ached and he
join o connec i o he main pa o he obo . Mo eo e ,
he senso elemen is added o he link, speci ying he
senso ype (e.g., came a, LiDAR, IMU, GNSS ecei e , e c)
and speci ic senso pa ame e s. To eplica e he eal-wo ld
beha io o he senso in he simula ed en i onmen , plugins
a e used [28]. These plugins a e sha ed lib a ies, p o iding
addi ional beha io s and capabili ies o he simula ed objec s
in Gazebo. Each senso has se e al con igu able pa ame e s
o ma ch he speci ica ions o i s physical ha dwa e, allowing
o ealis ic simula ion. Fo example, in he case o he
came a, i is possible o se he ho izon al and e ical ield
o iew (FOV), while o a LiDAR, pa ame e s such as he
ange, he numbe o ays o channels, he esolu ion, and he
upda e a e can be con igu ed. Addi ionally, i is possible o
apply he Gaussian noise, which allows he senso o eplica e
i s eal-wo ld beha io closely.
Gazebo p o ides he Model Da abase con aining se e al
p ede ined SDF models [41], including simple shapes like
boxes, sphe es, cylinde s, and mo e complex objec s like
ca s, people, buildings, and mo e. Howe e , use s can c ea e
hei own SDF model, ei he as a combina ion o simple
shapes o as complex meshes using a 3D modeling ool, such
as Blende [42], o a ailable in many online eposi o ies.
Rega ding ou simula ed en i onmen , Fig. 9shows a
ineya d a ailable online [43], con e ed in o a Collada
ile using Blende and impo ed in Gazebo. Al e na i ely,
he scene can be gene a ed by c ea ing a 3D map o he
178052 VOLUME 12, 2024
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
FIGURE 9. Simula ion en i onmen ob ained om a 3D ool.
en i onmen by con e ing cap u ed poin clouds o a eal
en i onmen using a 3D isual senso , such as a came a
o LiDAR. To c ea e a 3D map based on he poin cloud
o he en i onmen , a meshing algo i hm is used [44].
This algo i hm acili a es he gene a ion o a mesh-based
ep esen a ion o map impo in he Gazebo simula ion
en i onmen , ollowing hese ou lined s eps:
1) Downsampling The ob ained dense map is an op ional
ye ad an ageous s ep ha accele a es he mesh econ-
s uc ion p ocess and enhances he o e all ou come.
To ul ill his ask, samples a e gene a ed ollowing a
Poisson-disk dis ibu ion [45];
2) Knowing he no mals is essen ial o econs uc ing he
su ace o he elemen s ha make up he map;
3) The econs uc ion o he su ace begins om he se
o poin s and no mals. A iangle mesh is compu ed
employing The Ball Pi o ing algo i hm [46], which is
based on he p inciple ha h ee poin s cons i u e a
iangle i a ball o a use -de ined adius ouches hem
wi hou encompassing any o he s;
4) Tex u e mapping is c ea ed by iangle-by- iangle
pa ame e iza ion;
5) P ojec ing a 2D image on o he su ace a 3D model o
ex u e mapping is called UV mapping [47]. When he
UV map is eady, he colo can hen be ans e ed o he
econs uc ed su ace;
6) The mesh can be expo ed.
The e o e, he 3D map can be used o c ea e a Gazebo
SDF model. Fig. 10 showcases a ineya d ow ob ained by
ollowing he p ocedu e explained [44].
D. LIMITATIONS
This sec ion ou lines he limi a ion o Gazebo. Gazebo does
no na i ely suppo de o mable e ain. This is he main
issue in using Gazebo as a simula ion amewo k o he
de elopmen o he DT. Thus, simula ing he dynamics o
wheel/soil in e ac ion is no possible.
IV. MULTI-BODY MODEL
In o de o compa e he wo simula ion amewo ks,
we de eloped he Adams model o he Polibo , he obo o
which simula ion in Gazebo would be mo e challenging. This
FIGURE 10. Simula ion en i onmen econs uc ed s a ing om eal da a
acqui ed in he ield by means a RGB-D came a.
app oach aimed o enhance he e ec i eness o analyzing he
wo simula o s.
In MSC Adams ATV, he mul ibody model o Polibo is
cons uc ed as an assembly o se en independen ly modelled
subsys ems ha a e la e in oked and in eg a ed. As shown
in Fig. 12, hese se en subsys ems a e: he hull, he d i e
sp ocke , he on , middle, and ea suspension uni s wi h
hei oad wheels, he ack bel , and he senso ame. Each
subsys em is made up o di e en pa s joined oge he .
Fo example, he on swing a m is composed o 4 pa s,
namely he swing a m, he ensione , he idle wheel, and
he oad wheel. The bigges subsys em in e ms o numbe o
pa s is he ack bel . A single ack comp ises 56 elemen s,
and each elemen is made o wo pa s (a segmen and a
connec o ). This means ha bo h acks yield a pa -coun
o 224 bodies. The whole Polibo assembly is made up o
267 pa s. To maximize he accu acy o he model, all he
pa s o he obo we e expo ed om he 3D CAD models
as Pa asolid iles and hen impo ed in o MSC Adams. The
ine ial p ope ies o all he componen s a e cohe en wi h
hose o he eal p o o ype. The o e all mass o he assembly
is 125 kg, and i s cen e o g a i y is loca ed 40 mm ahead o
he d i ing axle and 242 mm abo e he g ound. The ehicle
powe ain is made up o a educe ha connec s he mo o
sha o he sp ocke , which hen engages wi h he ack, one
pe each side. Ei he a p esc ibed angula eloci y o o que
can be assigned o he mo o sha s.
The on suspension subsys em comp ises ou igid
bodies: wo pa s cons i u ing he swing a m, a oad wheel,
and he idle wheel. One o he pa s o he a m is hinged o
he hull ia a e olu e join so ha i is ee o pi o abou
he a achmen poin . Such o a ion is opposed by a linea
sp ing-dampe o ce de ined o model he shock abso be
placed be ween he a m and he chassis. The o he igid
body o ming pa o he on suspension a m is connec ed
o he pi o ing one by means o he ack ensioning uni ,
which changes he ela i e posi ion o he wo pa s ia a
ansla ional join .
The middle oad wheels a e ca ied by wo swing a ms
cons ained o he hull ia e olu e join s and linked o one
ano he by means o a linea sp ing-dampe o ce. Al hough
VOLUME 12, 2024 178053
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
he o a ions o he wo a ms o he middle suspension uni
a e kinema ically independen , he p esence o he elas ic
elemen makes su e ha when one wheel is li ed, he o he
is pushed agains he g ound, hus allowing he ack o be e
adap o he obs acles and he une en p o ile o he e ain
being a e sed.
The Polibo p o o ype p esen s ubbe -made acks ein-
o ced wi h s eel cable souls. Wi hin he ATV oolki he ack
is composed o a disc e e numbe o segmen s, which a e
au oma ically w apped a ound he olling elemen s be o e he
simula ion. The me hodology ollowed o build he mul ibody
model o Polibo in Adams is shown in Fig. 11 and can be
summa ized as ollows:
1) Design he o e wi h a CAD so wa e;
2) The geome y o all he pa s o he obo a e expo ed
as Pa asolid iles ex ac ed om 3D CAD elemen s;
3) Pa asolid iles a e impo ed in MSC Adams and
con e ed in o igid bodies. The igid bodies a e
connec ed ia e olu e, ansla ional o ixed join s, and
he ob ained sub-assemblies a e sa ed as Templa es;
4) Templa es a e used o c ea e independen symme ic
subsys ems on each side o he main body;
5) Subsys ems a e in eg a ed o o m he assembly o
simula e;
6) The ack elemen subsys em is w apped a ound he oad
wheels and sp ocke o c ea e he ack bel s;
7) Finally, he soil geome y and he pa ame e s a e
de ined.
FIGURE 11. Flow cha o he me hodology used o de eloping he
mul i-body model.
Fo mo e de ails on he mul ibody model o Polibo
(e.g. ine ial pa ame e s, con ac modeling, alida ion agains
expe imen al da a) he eade is in i ed o e e o [48].
FIGURE 12. Polibo mul ibody assembly and subsys ems.
A. LIMITATIONS
This sec ion add esses he limi a ions o he mul i-body
model de eloped wi h MSC Adams. These mainly ela e o
wo aspec s: compu a ional cos and ack-g ound in e ac ion
modeling.
As s a ed abo e, he mul ibody model o Polibo is made up
o 267 di e en bodies which, oge he wi h all he cons ain s
linking hem, b ing he numbe o equa ions sol ed by
MSC Adams a each ime s ep o 1824. This causes a high
compu a ional cos , especially on so soil (such as sand
o loam). Fo example, a compu e wi h an In el Co e i7-
4870HQ CPU clocking a 2.50 GHz and wi h 16 GB o RAM,
using 3 h eads in pa allel, simula es 1 second o Polibo ’s
dynamics on loam in abou 2940 seconds, which makes he
mul ibody model unsui able o eal- ime simula ions, unless
a much mo e powe ul sys em is in eg a ed.
The second main limi a ion ega ds he ack-g ound
in e ac ion modeling. On so soils, MSC Adams uses
Bekke ’s equa ion [49] as p essu e-sinkage ela ionship:
p=kc
b+kφzn(1)
whe e zis he sinkage, pis he p essu e, bis he wid h o
he con ac pa ch, n,kc, and kφa e e ain-speci ic p essu e-
sinkage pa ame e s.
The exponen ial equa ion by Janosi and Hanamo o [50] is
used as shea displacemen - shea s ess ela ionship:
τ=(c+p an φ)1−e−s/Ks(2)
whe e τis he shea s ess, cand φa e, espec i ely, cohesion
and in e nal ic ion angle o he e ain, Ksis he shea
de o ma ion modulus, sand pa e he shea displacemen and
he p essu e a he poin in which he shea s ess is o be
calcula ed.
Al hough hese models a e widely used in e amechan-
ics [51], hey ely on pa ame e s ha mus be de e mined
expe imen ally o each ype o soil, which limi s hei
178054 VOLUME 12, 2024
A. Rana e al.: Towa d Digi al Twin o O -Road Vehicles Using Robo Simula ion F amewo ks
gene al applicabili y ac oss di e en e ains and soil ypes.
Howe e , he al e na i e o classic e amechanics models
is he in eg a ion o a FEM model o he e ain, which
would wo sen he compu a ional bu den o he sys em.
Mo e impo an ly, he mul ibody model o Polibo has been
ex ensi ely alida ed agains expe imen al da a [48], p o ing
ha he dynamics o Polibo can be eplica ed wi h a good
le el o accu acy.
Finally, i is wo h ecalling ha senso da a canno be
simula ed o in eg a ed in MSC Adams, and he e o e he
obo canno in e ac wi h he en i onmen .
V. RESULTS AND DISCUSSION
In his sec ion, we assess he sui abili y o he wo obo ic
simula o s conce ning he implemen a ion o he DT o he
Polibo . To his aim, we simula e a common ag icul u al ask
by means o bo h amewo ks. The ask objec i e is o inspec
a ield o ganized in ows, such as in he con ex o a ineya d.
A. ROW FOLLOWING ALGORITHM
The goal o he simula ed ask is o ollow a pa allel ajec o y
wi h espec o he ow while main aining a cons an dis ance
o p e en collisions wi h and moni o c ops [52],[53],[54].
To his aim, he obo is equipped wi h an RGB-D came a
poin ing on he ow, an IMU, and a GNSS ecei e o es ima e
he ela i e pose wi h espec o he cu en ow. The ela i e
pose o he obo , exp essed by he ela i e o ien a ion and he
dis ance o he obo wi h espec o he cu en ow, is used
o eed he con ol algo i hm. Speci ically, he o ien a ion
o he obo is egula ed by ac ing on he angula eloci y
whose alue is gi en by a weigh ed sum o he dis ance and
o ien a ion e o . On he o he hand, he linea eloci y o he
obo is se a a cons an alue. The con ol law applied is as
ollows:
=σ
ω=k1(d−¯
d)+k2(γ− ¯γ),(3)
whe e and ωa e he linea and angula eloci y, espec-
i ely; dand γa e he dis ance and o ien a ion o he obo
w. . . he ineya d ow; σand ki, o i=1,2, a e cons an s
g ea e han 0; inally, ¯
dand ¯γa e he desi ed dis ance and
o ien a ion o he obo w. . . he ineya d ow. Fo mo e
de ails on hese opics, eade s a e encou aged o consul he
ela ed a icles [55] and [56].
B. SIMULATION SETUP
The same simula ion se up has been implemen ed in bo h
Gazebo and MSC Adams. Speci ically, in Gazebo, we ha e
simula ed he p esence o a eal ineya d ow in he en i on-
men . Addi ionally, he obo simula ion was comple ed by
inco po a ing an app op ia e senso sui e, namely an RGB-D
came a, a GNSS ecei e , and an IMU senso , o alida e and
es he pe o mance o he en i e algo i hm. Fig. 13 depic s
he simula ed en i onmen , including he obo equipped wi h
he senso sui e and he ineya d ow. Each o hese senso s
FIGURE 13. Visualiza ion in Gazebo o he simula ed en i onmen ,
including he obo wi h he senso sui e and he ineya d ow.
FIGURE 14. Visualiza ion in RViz o he obo , he ineya d cap u ed in
Gazebo and he poin cloud gene a ed by he RGB-D came a.
emula es he acquisi ion o ealis ic in o ma ion abou he
en i onmen and can be subs i u ed by he eal da a coming
om he physical asse o he DT. Fo ins ance, as depic ed
in Fig. 14, he simula ed came a gene a es a poin cloud
o he su ounding en i onmen a each acquisi ion ins an ,
p o iding a ealis ic ep esen a ion o he ineya d. Howe e ,
simula ing acked obo s in Gazebo p esen s challenges,
wi h modeling wheel/g ound in e ac ion dynamics being
hinde ed by he limi a ion o no simula ing de o mable
e ains. The e ain included in he simula ed en i onmen
eplica es a u al e ain solely in appea ance, no i s physical
p ope ies. None heless, in ou case s udy, he modeling
o a acked obo , namely he Polibo , was add essed by
subs i u ing he acks wi h wo equi alen wheels on each
side. Fu he mo e, knowing he eal loca ion o he ineya d,
i was possible o o ien a e he i ual ineya d wi h he same
o ien a ion, i.e., 12.5◦w. . . he geog aphic No h, which
co esponds o he x-axis o he Wo ld e e ence sys em in
Gazebo. The geog aphical e e ence posi ion o he obo ,
ins ead, is de ined by longi ude =18.34581509◦and la i ude
=40.05999619◦. This speci ic pa ame e is con igu able
wi hin he simula ed GNSS ecei e plugin. In addi ion,
o simula e he sys em unde quasi- eal condi ions, he ine ial
pa ame e s we e se ela i e o he eal ones. In pa icula , he
mass and he ine ial ma ix a e speci ied as ollows:
m=125 kg,(4)
I=
14.264 7.21E−05 7.21E−05
7.21E−05 11.628 3.09E−05
7.21E−05 3.09E−05 10.811
,(5)
whe e mis exp essed in kg, whe eas Iin kg ·m2.
VOLUME 12, 2024 178055
