This is a sel -a chi ed e sion o an o iginal a icle. This e sion
may di e om he o iginal in pagina ion and ypog aphic de ails.
Au ho (s):
Ti le:
Yea :
Ve sion:
Copy igh :
Righ s:
Righ s u l:
Please ci e he o iginal e sion:
CC BY-NC-ND 4.0
h ps://c ea i ecommons.o g/licenses/by-nc-nd/4.0/
Enhancing Holonic A chi ec u e wi h Na u al Language P ocessing o Sys em o
Sys ems
© 2024 he Au ho s
Accep ed e sion (Final d a )
Ash aq, Muhammad; Sadik, Ahmed; Mikkonen, Tommi; Waseem, Muhammad;
Mäki alo, Niko
Ash aq, M., Sadik, A., Mikkonen, T., Waseem, M., & Mäki alo, N. (2024). Enhancing Holonic
A chi ec u e wi h Na u al Language P ocessing o Sys em o Sys ems. In H.-G. Fill, F. J.
Domínguez Mayo, Sinde en, Ma en an, & L. Maciaszek (Eds.), ICSOFT 2024 : P oceedings o
he 19 h In e na ional Con e ence on So wa e Technologies - Volume 1 (pp. 427-433).
SCITEPRESS Science And Technology Publica ions. h ps://doi.o g/10.5220/0012787300003753
2024
Enhancing Holonic A chi ec u e wi h Na u al Language P ocessing o
Sys em o Sys ems
Muhammad Ash aq1∗, Ahmed R. Sadik2, Tommi Mikkonen1,
Muhammad Waseem1, and Niko M¨
aki alo1∗
1Uni e si y o Jy ¨
askyl¨
a, Jy ¨
askyl¨
a, Finland
2Honda Resea ch Ins i u e Eu ope, Ge many
∗Co esponding au ho : [email p o ec ed]
Keywo ds: Sys em o Sys ems, Holon Communica ion, Holonic a chi ec u e, Na u al Language P ocessing, Con e sa-
ional Gene a i e In elligence, In e ope abili y, Human-Sys em In e ac ion.
Abs ac : The e e -g owing complexi y and dynamic na u e o mode n Sys em o Sys ems (SoS) necessi a e e icien
communica ion mechanisms o ensu e in e ope abili y and collabo a i e unc ioning among cons i uen sys-
ems (CS), e e ed o as holons in he holonic a chi ec u e o SoS. This pape p oposes a no el app oach o
enhance humand- o-holon and holon- o-holon communica ion wi hin he holonic a chi ec u e h ough he in-
eg a ion o Na u al Language P ocessing (NLP) echniques. Ou p oposed amewo k u ilizes ad ancemen s
in NLP, speci ically La ge Language Models (LLMs), enabling holons o unde s and and ac on na u al lan-
guage ins uc ions. This enables mo e in ui i e holon- o-holon and human- o-holon in e ac ions, leading o
be e coo dina ion among di e se sys ems. The amewo k’s p ac ical applica ion is demons a ed h ough an
Unmanned Vehicle Flee (UVF) case s udy, showcasing i s po en ial in enhancing communica ion and coo di-
na ion in complex SoS. Addi ionally, we p opose e alua ion s a egies o assess he e iciency and e ec i eness
o his amewo k, and iden i y a eas o imp o emen . This wo k se s he s age o u u e explo a ion and p o-
o ype implemen a ion, pa ing he way o u he ad ancemen s in SoS communica ion and collabo a ion.
1 INTRODUCTION
A Sys em o Sys em (SoS) is a collec ion o sys-
ems unc ioning oge he o achie e a common
goal (Nielsen e al., 2015). These SoSs a e comp ised
o mul iple Cons i uen Sys ems (CS), each unc-
ioning independen ly wi h i s managemen s uc u e.
CSs wi hin an SoS can be geog aphically dispe sed,
u he highligh ing he need o e ec i e communi-
ca ion and coo dina ion. When in eg a ed, he o e all
SoS capabili ies a e a mo e han ha o he indi id-
ual CSs o ming he SoS. Mo eo e , an SoS should
suppo e olu iona y de elopmen allowing CSs o
join o lea e he SoS a un ime o mee he desi ed
needs. SoS inds nume ous applica ions in eal-li e
domains, including ene gy g ids, ai a ic manage-
men , de ense, and obo ics swa ms (Jamshidi, 2008).
Managing he complexi y o SoS is a majo chal-
lenge. T adi ional app oaches o en s uggle due
o he inhe en au onomy and he e ogeneous na-
u e o CSs. Holonic a chi ec u es (Blai e al.,
2015) o e a p omising solu ion by decomposing
he SoS in o smalle , sel -go e ning en i ies called
holons (Koes le , 1968). This duali y acili a es a e-
cu si e sys em a chi ec u e, allowing o sel - eliance
alongside coope a ion wi h o he holons o ming
a hola chy—a hie a chy o holons ope a ing au-
onomously ye in coo dina ion o achie e common
objec i es. The holonic app oach aligns wi h key SoS
a chi ec u al p inciples, such as in e ope abili y, scal-
abili y, and adap abili y. These p inciples mo i a e
esea che s o ep esen he CS o SoS as ‘holons’.
This ep esen a ion enables unc ionali ies such as CS
disco e y and dynamic SoS composi ion (Elhabbash
e al., 2024).
While he holonic a chi ec u e o e s a p omising
app oach o SoS enginee ing, i can ace subs an-
ial in e ope abili y challenges. The he e ogeneous
CSs, o holons, o en adhe e o dis inc da a o -
ma s, communica ion p o ocols, and in e ac ion pa -
e ns. This di e si y c ea es in e ope abili y hu dles,
hinde ing seamless in o ma ion sha ing, command in-
e p e a ion, and ask collabo a ion equi ing special-
ized knowledge o SoS unde s anding and imple-
men a ion. Mo eo e , he dynamic na u e o SoS,
whe e CSs can join o lea e, necessi a es adap i e
and lexible communica ion mechanisms o handle
e ol ing SoS composi ions. Finally, while in ope -
a ion, SoS should ha e he capabili y o in e ac wi h
humans, necessi a ing communica ion o expand o-
wa ds a o m ha is immedia ely unde s ood by hu-
mans.
Empowe ing holons wi h Na u al Language P o-
cessing (NLP) capabili ies p esen s a ans o ma i e
app oach o o e coming he a o emen ioned commu-
nica ion hu dles. Such capabili ies can enable holons
o in e p e and espond o na u al language ins uc-
ions, he eby simpli ying he in e ac ion be ween
holons and humans and educing he eliance on in-
e nal sys em knowledge. Fu he mo e, his app oach
acili a es holon- o-holon communica ion by encod-
ing and decoding machine-execu able commands in o
na u al language. Thus, he NLP laye ac s as a
communica ion laye among humans and holons, ag-
nos ic o unde lying CS he e ogenei y and p o ocols,
he eby enhancing o e all collabo a ion and adap -
abili y in SoS.
Recen esea ch explo es in eg a ing NLP ech-
nologies, especially La ge Language Models (LLMs),
in o obo s (Koubaa, 2023), leading o imp o ed
human- obo collabo a ion. Howe e , hese s udies
do no inco po a e mul i- obo unc ionali y, which is
essen ial o obo - o- obo communica ion. Fu he -
mo e, hei scope is limi ed o obo ics a he han o
SoS and holonic a chi ec u e.
In his pape , we p opose ex ending he holonic a -
chi ec u e by in eg a ing he NLP capabili ies di ec ly
in o he holons. We p esen a concep ual amewo k
o NLP-enhanced SoS, enabling na u al language in-
e ac ion and decision-making. Th ough he u iliza-
ion o ad anced NLP echnologies, such as LLMs,
ou app oach acili a es seamless communica ion and
collabo a ion wi hin SoS. The amewo k’s p ac ical
applica ion is demons a ed h ough an Unmanned
Vehicle Flee (UVF) case s udy. To he bes o ou
knowledge, his is he i s wo k o i s kind o explo e
NLP-enhanced holonic a chi ec u es wi hin he SoS
domain.
The emainde o his pape is o ganized as ol-
lows. Sec ion 2 p o ides backg ound in o ma ion on
NLP and holonic a chi ec u e. Sec ion 3 e iews he
s a e o he a in his opic. Sec ion 4 de ails he p o-
posed concep ual amewo k inco po a ing NLP in o
he holonic a chi ec u e. Sec ion 5 demons a es he
applica ion o he amewo k o he UVF case s udy.
Sec ion 6 p esen s ideas o e alua ing he amewo k.
Sec ion 7 discusses ou indings and hei implica-
ions. Finally, Sec ion 8 concludes he pape by d aw-
ing inal conclusions and ou lining po en ial a enues
o u u e wo k.
2 BACKGROUND
2.1 Na u al Language P ocessing
Na u al Language P ocessing (NLP) has become a
co ne s one o a i icial in elligence, acili a ing com-
munica ion be ween humans and compu e s (Khu-
ana e al., 2023). I encompasses a ious echniques
o enabling compu e s o unde s and, in e p e , and
gene a e human language. Among hese echniques,
La ge Language Models (LLMs) ha e e olu ionized
he capabili ies o machines in p ocessing and gene -
a ing human-like ex . These LLMs, such as BERT
(Bidi ec ional Encode Rep esen a ions om T ans-
o me s) and GPT (Gene a i e P e- ained T ans-
o me ), a e ypically based on complex neu al ne -
wo ks ained on ex ensi e ex da ase s (Zhao e al.,
2023). Th ough aining, hese neu al ne wo ks in i-
ca e language ea u es, including pa e ns, s uc u es,
con ex , and seman ics, enabling hem o pe o m ad-
anced such as ex classi ica ion, sen imen analysis,
ansla ion, and ques ion-answe ing (Rad o d e al.,
2019; B own e al., 2020). LLMs ha e ound ap-
plica ions in di e se domains, including so wa e de-
elopmen asks like p og amming and code gene a-
ion (Sadik e al., 2023b). Howe e , hei in eg a ion
in o b oade human-sys em, in e -sys em, and in a-
sys em in e ac ions is s ill in i s ea ly s ages. This
pape ocuses on explo ing his po en ial o b oade
in eg a ion.
2.2 Holonic A chi ec u e o Sys em o
Sys ems
Holons a e au onomous ye connec ed en i ies ha
possess independen unc ionali ies while con ibu -
ing o a la ge sys em (Koes le , 1968). Due o hei
dual na u e, holons a e excellen o modeling he
he e ogeneous CSs, accu a ely e lec ing hei inde-
penden unc ions and con ibu ions o he o e all
SoS (Blai e al., 2015).
3 RELATED WORK
Nundloll e al. (2020) u ilized his concep o using
holons o model IoT sys ems and in oduced a ame-
wo k ha desc ibes holons using on ologies. Elhab-
bash e al. (2024) adap ed his amewo k o he SoS
domain and p oposed an SoS a chi ec u e whe e CSs
a e modeled as on ological holons. This a chi ec u e
allows CSs o eason abou and unde s and each o he ,
acili a ing CS disco e y, ad-hoc scalabili y, and dy-
namic SoS composi ion.
2
Howe e , his a chi ec u e assumes ha he on o-
logical desc ip ions o he holons a e manually p o-
ided by endo s o sys ems enginee s. Add essing
his limi a ion, Zhang e al. (2023) p opose an NLP-
based app oach ha au oma ically ex ac s on ologi-
cal desc ip ions o IoT de ices by sc aping web da a.
While his app oach o e s au oma ion, he holonic a -
chi ec u e o SoS s ill lacks capabili ies o holon-
o-en i onmen communica ion (Halba e al., 2021),
human- o-holon in e ac ion, and communica ion wi h
unknown holons.
4 CONCEPTUAL FRAMEWORK
Ou p oposed amewo k aims o add ess he com-
munica ion challenges inhe en in SoS by le e aging
NLP echniques wi h a pa icula ocus on enabling
holon- o-holon and holon- o-human communica ion.
Figu e 1 illus a es he amewo k depic ing i s com-
ponen s and hei in e ac ions.
4.1 O e iew
The amewo k in ol es a human ope a o who p o-
ides na u al language ins uc ions o he holons ep-
esen ing he CS o an SoS. The ins uc ions can ange
om b oad, high-le el goals o speci ic asks.
The CSs o he SoS a e ep esen ed by holons
(e.g., Holon A and Holon B in Figu e 1). Each holon
possesses speci ic capabili ies o se ices, which a e
he esou ces in eg a ed in o hem. In addi ion, each
holon is also equipped wi h a localized NLP mod-
ule. These goals usually exceed he capabili ies o
a single holon and equi e he collabo a ion o mul i-
ple holons. The sys em hen iden i ies ele an holons
based on hei capabili ies and o ches a es hei col-
labo a ion o achie e he ope a o ’s objec i es.
4.2 NLP In eg a ion Module
This module is he co e o he amewo k. This
module comp ises h ee componen s: on ology-based
p omp enginee ing, an LLM, and an NLP communi-
ca ion in e ace.
4.2.1 On ology-based P omp Enginee ing
This componen is esponsible o c a ing p ecise
and con ex ually ele an p omp s by inco po a ing
domain-speci ic on ologies. This ensu es ha he op-
e a o ’s inpu s a e ailo ed accu a ely o he speci ic
needs o he SoS domain, he eby enhancing he e -
ec i eness and p ecision o he LLM’s esponse (Sec-
ion 4.2.2).
4.2.2 La ge Language Model (LLM)
This componen u ilizes NLP capabili ies o unde -
s and and gene a e na u al language in e ac ions. I
in e p e s he e ined inpu om he p omp enginee -
ing componen (Sec ion 4.2.1), p ocesses i , and con-
e s i in o a syn ax comp ehensible o o he he e o-
geneous holons o he SoS.
4.2.3 NLP Communica ion In e ace
This in e ace ansla es he p ocessed ins uc ions
om he LLM in o ac ionable commands o he
holons. These commands can ac i a e o deac i-
a e he holons’ capabili ies o se ices as needed o
achie e he o e a ching goal.
The NLP module p o ides eedback o he human
ope a o in he o m o con i ma ion messages, cla i-
ying ques ions, o summa ies o he in ended ac ions.
This allows he ope a o o e i y he sys em’s unde -
s anding and make any necessa y co ec ions o e-
inemen s.
4.2.4 Holon-le el NLP
Each holon is equipped wi h a localized NLP mod-
ule, including a domain-speci ic LLM, enabling i o
unde s and and gene a e na u al language ins uc ions
o bo h human-holon and holon-holon in e ac ions.
The human-holon communica ion occu s when he
NLP In eg a ion Module in e ac s wi h he localized
NLP module o holons. The holon- o-holon in e ac-
ion acili a es he holons o coo dina e and exchange
sys em desc ip ions o hei encapsula ed sys em wi h
o he holons using na u al language (NL). As appa -
en in he Figu e 1, he holon- o-holon in e ac ion is
possible e en wi hou he NLP In eg a ion module.
This dis ibu ed app oach enhances au onomy and e-
duces eliance on a cen alized module.
4.3 Holon Composi ion and
Collabo a ion
Holons, u ilizing hei embedded NLP capabili ies,
in e ac wi h each o he and he cen al NLP mod-
ule o de e mine hei capabili ies and ele ance o
he goal. Th ough an i e a i e nego ia ion p ocess,
holons commi o p o iding speci ic se ices, while
o he s may be deemed i ele an . The ele an holons
hen o m a holon composi ion, collabo a ing o ul ill
he gi en mission.
3
Na u al Language In e ac ion
Human
Ope a o
Re ined p omp
On ology-based P omp
enginee ing
NLP Communica ion
In e ace
Capabili ies
/
Se ices
Sys em o Sys ems
Holon A
Capabili ies
/
Se ices
Holon B
T ansla es o/ om ac ionable commands and holon desc ip ions
NL Communica ion
La ge Language Model
LLM
Gene a e s uc u ed commands
Re ine Ins uc ions (i needed)
NLP In eg a ion Module
Figu e 1: The concep ual amewo k showing NLP in eg a ion wi h he Holonic A chi ec u e
This app oach s eamlines he p ocess o achie -
ing complex goals wi hin an SoS. Ins ead o equi ing
in-dep h knowledge o each cons i uen sys em, he
ope a o can simply communica e hei in en in na -
u al language, and he amewo k handles he es .
5 CASE STUDY:
NLP-INTEGRATED UVF FOR
URBAN MOBILITY
5.1 O e iew
This case s udy explo es he in eg a ion o he p o-
posed NLP-enhanced holon communica ion ame-
wo k (Sec ion 4) in o sma ci y anspo a ion us-
ing Unmanned Vehicle Flee s (UVF). The UVF, op-
e a ing wi hin a dynamic en i onmen , exempli ies an
SoS wi h complex in e ac ion dynamics and scalabil-
i y challenges such as e ol ing missions, expanding
ange and capaci y demands, ehicle ailu es, and ba -
e y limi a ions (Sadik e al., 2023a). I comp ises
se e al au onomous en i ies, including Unmanned
G ound Vehicles (UGVs) and Unmanned Ae ial Ve-
hicles (UAVs), each unc ioning as a holon, ope a ing
independen ly as well as collabo a ing o achie e he
lee ’s o e all objec i es (Tchappi e al., 2020).
5.2 Scena io
Conside a scena io whe e a esiden in a sma ci y
needs anspo a ion om Posi ion A o Posi ion B.
The goal is o na iga e he complex ci yscape quickly
and e icien ly, conside ing no- ly zones, a ic condi-
ions, and oad layou s. This scena io p esen s se e al
challenges:
•Complex U ban En i onmen : Na iga ing
h ough a densely popula ed u ban a ea wi h
a ying al i udes and no- ly zones o UAVs.
•Dynamic Rou ing: Adap ing in eal- ime o a ic
and en i onmen al condi ions o ensu e he as es
and sa es ou e.
•Vehicle Coo dina ion: Seamlessly ansi ioning
be ween UAVs and UGVs while main aining a
consis en and com o able expe ience.
•Communica ion: Ensu ing clea and e icien
communica ion be ween he use , UVs, and he
con ol cen e o manage expec a ions and adap
o any changes in he mission.
5.3 NLP-enhanced UVF
Communica ion
Figu e 2 illus a es he in eg a ion o NLP module
(Sec ion 4.2) in o he UVF communica ion ame-
wo k. This in eg a ion enhances bo h human- o- lee
and in a- lee communica ions.
5.3.1 Human-UV In e ac ion
The use communica es hei des ina ion o he u ban
mobili y se ice using a na u al language in e ace.
The NLP module’s on ology-based p omp enginee -
ing componen (Sec ion 4.2.1) e ines his inpu , en-
su ing i ’s ailo ed o he UVF domain. The LLM
(Sec ion 4.2.2) hen in e p e s he eques and ini ia es
mission planning.
4

NLP In eg a ion
Figu e 2: Human-UV In e ac ion and Planning
5.3.2 In a-Flee Nego ia ion
To op imize e iciency, ep esen a i e UVs om bo h
he UGV and UAV swa ms a e selec ed o in e ac
wi h he use and he u ban mobili y se ice. These
ep esen a i es use hei localized NLP modules (Sec-
ion 4.2.3) o communica e wi h he es o he lee ,
nego ia ing oles, pa hs, and iming based on indi-
idual capabili ies, cu en s a us, and en i onmen-
al ac o s. This nego ia ion p ocess aligns wi h Sec-
ion 4.2.4 and Sec ion 4.3 pa s o he amewo k.
Fo example, he UAV ep esen a i e communi-
ca es i s es ima ed ime o a i al a he no- ly zone
and a ailable landing zones o he UGV ep esen a-
i e. The UGV ep esen a i e hen analyzes po en ial
ou es, conside ing a ic condi ions and oad lay-
ou s, and p oposes a sui able landing zone and en-
dez ous ime. The UAV and UGV ep esen a i es i -
e a e on his p ocess un il a mu ually ag eeable plan
is eached.
5.4 Resul ing UVF Composi ion
Following he nego ia ion p ocess, a UVF is o med
o accommoda e he no- ly zone (P1– P2in Figu e 3).
This composi ion includes wo UAVs o ae ial seg-
men s and one UGV o g ound anspo a ion, show-
casing he dynamic SoS composi ion capabili y en-
abled by he amewo k. The UAV, upon eaching he
no- ly zone, communica es wi h he UGV in na u al
language, I am app oaching he no- ly zone. Please
p epa e o ecei e he passenge a he designa ed
landing zone (Sec ion 4.2.4). The UVs coo dina e
seamlessly, ensu ing a smoo h ansi ion o he use
a he landing and launch zones.
Upon eaching he p ede e mined landing zone
close o he use ’s loca ion (Posi ion A), he i s UAV
communica es wi h he UGV o p epa e o a smoo h
ansi ion o he no- ly zone. The use is hen ans-
po ed by he UGV o a launch zone close o his des-
ina ion, whe e he second UAV akes o e o com-
ple e he inal leg o he jou ney.
A
B
P1
P2
Figu e 3: Resul ing UVF o wo UAVs and one UGV (P1—
P2is no- ly zone)
5.5 Mission Success C i e ia
The success o he mission will be e alua ed based on
he ime aken o comple e he mission, he numbe
o passenge s success ully anspo ed, he numbe o
success ul nego ia ions among UVs, and he o e all
sa is ac ion o he human ope a o wi h he sys em’s
pe o mance. These me ics will p o ide aluable in-
sigh s in o he e ec i eness and usabili y o he NLP-
enhanced holonic a chi ec u e in eal-wo ld scena -
5
ios.
6 EVALUATION
While his pape p ima ily ocuses on p esen ing a
concep ual amewo k, we acknowledge he impo -
ance o e alua ing i s e ec i eness in eal-wo ld sce-
na ios. Fu u e wo k will in ol e a igo ous e alua ion
o he p oposed NLP-enhanced holonic a chi ec u e
in a simula ed en i onmen , simila o he app oach
used by (Sadik e al., 2023a). The implemen a ion
can be done in by de eloping a mul i-agen simula ion
using a sui able amewo k (e.g., JADE), o a mul i-
obo en i onmen using ROS 2 and Gazebo o model
he in e ac ions be ween he human ope a o , he NLP
module, and he holons.
Po en ial e alua ion me ics could include:
•Task Comple ion Ra e: The pe cen age o asks
comple ed by he SoS using he NLP in e ace
compa ed o adi ional me hods.
•Communica ion E iciency: The educ ion in
communica ion o e head (e.g., message olume,
bandwid h usage) and ime o comple e asks
achie ed h ough NLP-based in e ac ion.
•Communica ion E ec i eness: The pe cen age
o co ec ly in e p e ed use eques s and holon-
gene a ed ins uc ions.
•Adap abili y: The abili y o he SoS o dynam-
ically econ igu e and adap o changes in he
en i onmen o mission objec i es, acili a ed by
NLP-based nego ia ion and coo dina ion.
•Usabili y: Quali a i e assessmen o he ease o
use, in ui i eness, and use sa is ac ion o he NLP
in e ace o bo h human ope a o s and holons.
By quan i ying hese me ics, we can assess he
impac o NLP in eg a ion on SoS pe o mance and
iden i y a eas o u he imp o emen .
7 DISCUSSION
The in eg a ion o NLP wi hin he holonic a chi-
ec u e o e s se e al ad an ages ha con ibu e o
he imp o ed e iciency, adap abili y, and usabili y o
SoS.
Fi s ly, by enabling na u al language communi-
ca ion be ween he human ope a o and he holons,
he amewo k educes he cogni i e load on he op-
e a o , who no longe needs o be amilia wi h he
speci ic syn ax o p o ocols o each cons i uen sys-
em. This s eamlines he in e ac ion p ocess and al-
lows o mo e e icien ask assignmen and coo dina-
ion. T adi ionally, achie ing a goal in an SoS would
equi e he ope a o o be amilia wi h he in e nal
wo kings o each CS. The ope a o would hen need
o o ches a e hese CSs oge he o design an SoS ha
accomplishes he goal.
Secondly, he use o NLP enables he holons o
dynamically nego ia e and adap hei oles based on
he gi en goal and he capabili ies o o he holons.
This adap abili y is c ucial in complex and dynamic
SoS en i onmen s whe e he composi ion o holons
may change o e ime. Finally, he in ui i e na u e
o na u al language in e ac ion enhances he o e all
usabili y o he SoS, making i mo e accessible o a
wide ange o use s, including hose wi hou special-
ized echnical knowledge.
8 CONCLUSION AND FUTURE
WORK
This pape in oduces a no el app oach o enhance
holon communica ion wi hin SoS h ough Na u al
Language P ocessing, aiming o b idge he communi-
ca ion gap be ween human ope a o s and holons, and
among holons hemsel es. The p oposed amewo k
demons a es he po en ial o NLP o imp o e he e -
iciency, adap abili y, and usabili y o SoS, pa ing he
way o mo e in ui i e and e ec i e sys em-le el col-
labo a ion.
O e all, his ield ep esen s a p omising a ea o
ongoing esea ch, wi h u u e de elopmen s expec ed
o u he e ine and alida e he p oposed model in
p ac ical SoS applica ions. In he sho e m, le e -
aging he ROS2 pla o m (Dauba is e al., 2023),
we a e cu en ly implemen ing he p oposed ame-
wo k using he Holon P og aming Model (Ash aq
e al., 2024) and sys em a chi ec u es o au onomous
obo s (M¨
aki alo e al., 2021).
While he in eg a ion o LLMs in o he holonic
a chi ec u e holds p omise o enhancing in e ope -
abili y and adap abili y, u u e esea ch should also
add ess po en ial e hical conce ns, such as p i acy,
sa e y, and con lic s o in e es (Rousi e al., 2023;
Le inson e al., 2024). Fu u e wo k should also o-
cus on add essing he challenges associa ed wi h na -
u al language ambigui y, such as implemen ing cla -
i ica ion dialogs. This can in ol e explo ing he use
o domain-speci ic on ologies, con olled na u al lan-
guages, cla i ica ion dialogs, and con ex -awa e in e -
p e a ion. Addi ionally, he e is a need o in eg a e
eedback mechanisms and alida ion echniques o
6
ensu e obus and eliable communica ion wi hin he
SoS, enabling con inuous lea ning and imp o emen
o he NLP module.
REFERENCES
Ash aq, M., Sadik, A. R., Mikkonen, T., Waseem, M., and
M¨
aki alo, N. (2024). The holon p og amming model o
he e ogeneous and adap i e sys em o sys ems. TechRxi
p ep in 10.36227/ ech xi .171387729.91629072/ 2.
Blai , G., B ombe g, Y.-D., Coulson, G., Elkha ib, Y.,
R´
e eill`
e e, L., Ribei o, H. B., Ri i`
e e, E., and Ta¨
ıani, F.
(2015). Holons: Towa ds a sys ema ic app oach o com-
posing sys ems o sys ems. In 14 h In e na ional Wo k-
shop on Adap i e and Re lec i e Middlewa e, pages 1–6.
B own, T., Mann, B., Ryde , N., Subbiah, M., Kaplan, J. D.,
Dha iwal, P., Neelakan an, A., Shyam, P., Sas y, G.,
Askell, A., e al. (2020). Language models a e ew-sho
lea ne s. Ad ances in neu al in o ma ion p ocessing sys-
ems, 33:1877–1901.
Dauba is, P., Linkola, S., Kan osalo, A., and M¨
aki alo, N.
(2023). Ge ing s a ed wi h ROS2 de elopmen : A
case s udy o so wa e de elopmen challenges. In 2023
IEEE/ACM 5 h In e na ional Wo kshop on Robo ics So -
wa e Enginee ing, pages 37–44.
Elhabbash, A., Elkha ib, Y., Nundloll, V., Ma co, V. S., and
Blai , G. S. (2024). P incipled and au oma ed sys em o
sys ems composi ion using an on ological a chi ec u e.
Fu u e Gene a ion Compu e Sys ems, 157:499–515.
Halba, K., G i o , E., Lba h, A., and Dahbu a, A. (2021).
A amewo k o he composi ion o IoT and CPS capa-
bili ies. In IEEE 45 h Annual Compu e s, So wa e, and
Applica ions Con e ence, pages 1265–1272.
Jamshidi, M. (2008). Sys ems o sys ems enginee ing: P in-
ciples and applica ions. CRC p ess.
Khu ana, D., Koli, A., Kha e , K., and Singh, S. (2023).
Na u al language p ocessing: S a e o he a , cu en
ends and challenges. Mul imedia ools and applica-
ions, 82(3):3713–3744.
Koes le , A. (1968). The ghos in he machine. Macmillan.
Koubaa, A. (2023). ROSGPT: Nex -gene a ion human-
obo in e ac ion wi h Cha GPT and ROS. P ep in s.o g,
2023:2023040827.
Le inson, L., Die ich, M., Sa kisian, A., Sabano ic, S., and
Sma , W. D. (2024). P i acy awa e obo ics. In Com-
panion o he 2024 ACM/IEEE In e na ional Con e ence
on Human-Robo In e ac ion, pages 1335–1337.
M¨
aki alo, N., Linkola, S., Lau inen, T., and M¨
annis ¨
o, T.
(2021). Towa ds no el and in en ional coope a ion o di-
e se au onomous obo s: An a chi ec u al app oach. In
ECSA 2021 Companion, olume 2978 o CEUR Wo k-
shop P oceedings.
Nielsen, C. B., La sen, P. G., Fi zge ald, J., Woodcock, J.,
and Peleska, J. (2015). Sys ems o sys ems enginee ing:
Basic concep s, model-based echniques, and esea ch
di ec ions. ACM Compu ing Su eys, 48(2):1–41.
Nundloll, V., Elkha ib, Y., Elhabbash, A., and Blai , G. S.
(2020). An on ological amewo k o oppo unis ic
composi ion o IoT sys ems. In IEEE in e na ional con-
e ence on in o ma ics, IoT, and enabling echnologies,
pages 614–621.
Rad o d, A., Wu, J., Child, R., Luan, D., Amodei, D.,
Su ske e , I., e al. (2019). Language models a e un-
supe ised mul i ask lea ne s. OpenAI blog, 1(8):9.
Rousi, R., Samani, H., M¨
aki alo, N., Vakku i, V., Linkola,
S., Kemell, K.-K., Dauba is, P., F onza, I., Mikkonen,
T., and Ab ahamsson, P. (2023). Business and e hi-
cal conce ns in domes ic con e sa ional gene a i e ai-
empowe ed mul i- obo sys ems. In In e na ional Con-
e ence on So wa e Business, pages 173–189. Sp inge
Na u e Swi ze land Cham.
Sadik, A. R., Bolde , B., and Subasic, P. (2023a). A sel -
adap i e sys em o sys ems a chi ec u e o enable i s ad-
hoc scalabili y: Unmanned ehicle lee -mission con ol
cen e case s udy. In 7 h In e na ional Con e ence on In-
elligen Sys ems, Me aheu is ics & Swa m In elligence,
pages 111–118.
Sadik, A. R., Ce a ola, A., Joublin, F., and Pa a, J. (2023b).
Analysis o Cha GPT on sou ce code. a Xi p ep in
a Xi :2306.00597.
Tchappi, I. H., Galland, S., Kamla, V. C., Kamgang, J. C.,
Mualla, Y., Najja , A., and Hilai e, V. (2020). A c i ical
e iew o he use o holonic pa adigm in a ic and ans-
po a ion sys ems. Enginee ing Applica ions o A i icial
In elligence, 90:103503.
Zhang, Z., Elkha ib, Y., and Elhabbash, A. (2023). NLP-
based gene a ion o on ological sys em desc ip ions o
composi ion o sma home de ices. In IEEE In e na-
ional Con e ence on Web Se ices, pages 360–370.
Zhao, W. X., Zhou, K., Li, J., Tang, T., Wang, X., Hou,
Y., Min, Y., Zhang, B., Zhang, J., Dong, Z., Du, Y.,
Yang, C., Chen, Y., Chen, Z., Jiang, J., Ren, R., Li,
Y., Tang, X., Liu, Z., Liu, P., Nie, J.-Y., and Wen, J.-
R. (2023). A su ey o la ge language models. a Xi
p ep in a Xi :2303.18223.
7