Digital Twin and Foundation Models: A New Frontier

Athanasios Trantas

and Paolo Pileggi

Advanced Computing Engineering, Unit ICT, Strategy & Policy, TNO, The Netherlands

Keywords:

Artiﬁcial Intelligence, Digital Twin, Foundation Model.

Abstract:

A Foundation Model (FM) possesses extensive learning capabilities; it learns from diverse datasets. This is

our opportunity to enhance the functionality of Digital Twin (DT) solutions in various sectors. The integration

of FMs into the DT application is particularly relevant due to the increased prevalence of Artiﬁcial Intelli-

gence (AI) in real-world applications. In this position paper, we begin to explain a novel perspective on this

integration by exploring the potential of enhanced predictive analytics, adaptive learning, and improved han-

dling of complex data within DTs — by way of designated purposes. Ultimately, we aim to uncover hidden

value of enhanced reliable decision-making, whereby systems can make more informed, accurate and timely

decisions, based on comprehensive data analytics and predictive insights. Mentioning selected ongoing cases,

we highlight some beneﬁts and challenges, like computational demand, data privacy concerns, and the need

for transparency in AI decision-making. Underscoring the transformative implications of integrating FMs into

the DT paradigm, a shift towards more intelligent, versatile and dynamic systems becomes clearer. We cau-

tion against the challenges of computational resources, safety considerations and interpretability. This step is

pivotal towards unlocking unprecedented potential for advanced data-driven solutions in various industries.

1 INTRODUCTION

The concept of Digital Twin (DT) and the Foundation

Model (FM) represents a conﬂuence of real-world and

digital realms, each with its transformative potential.

Digital twins, as deﬁned by the Digital Twin Consor-

tium, are virtual representations of real-world enti-

ties and processes, synchronised at speciﬁc frequen-

cies and ﬁdelity (Digital Twin Consortium, 2023).

The DT paradigm uses real-time and historical data

to mirror and understand real-world systems and their

processes throughout their entire life-cycle. Utilising

sensors or other data-producing mechanisms, the dig-

ital twin and its real-world counterpart can achieve

synchronisation with a high degree of detail, as de-

picted in Figure 1. The digital twin processes ex-

perimental inputs (such as conﬁguration parameters)

and yields model outputs. Conversely, the real twin

gathers feedback from the real world and can enact

changes in its environment. Synchronisation of the

digital twin occurs via state updates received from

the real twin, while the digital twin can guide the

real twin through state predictions or control com-

mands (Semeraro et al., 2021). Digital twins are in-

https://orcid.org/0000-0001-7109-9210

https://orcid.org/0009-0001-6031-201X

creasingly prevalent across various sectors, including

manufacturing, healthcare, urban planning and envi-

ronmental monitoring (Barricelli et al., 2019). They

serve as synchronised models for real-world systems

or objects, while the DT Applications (DTA) that em-

ploy them serve one or more purposes, like improved

decision-making, predictive maintenance and system

optimisation. This requires the digital twin to interact

with the digital environment and possibly other digital

twins, as the ﬁgure suggests.

FMs represent a paradigm shift in AI. Exempliﬁed

by large-scale Machine Learning models, like Gener-

ative Pre-trained Transformer 3 (Brown et al., 2020),

Bidirectional Encoder Representations from Trans-

formers (Devlin et al., 2018) and Contrastive Lan-

guage Image Pre-training (Radford et al., 2021)) that

are trained on extensive and diverse datasets. A key

characteristic is their ability to adapt to a wide range

of tasks and domains, leveraging their capacity to

generalise from the training data (Yuan, 2023) — gen-

erally using self-supervision (Hinton et al., 2006) at

scale. They have shown remarkable capabilities in un-

derstanding and generating human language, photo-

realistic images and videos; with potential in various

applications — from automated content creation to

complex decision-making (Yang et al., 2023).

988

Trantas, A. and Pileggi, P.

Digital Twin and Foundation Models: A New Frontier.

DOI: 10.5220/0012427000003636

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 3, pages 988-994

ISBN: 978-989-758-680-4; ISSN: 2184-433X

Figure 1: Interaction between the digital twin and its real-

world counterpart.

Following (Bommasani et al., 2021), our work

uses the term Foundation Model to describe a broad

paradigm shift in AI that encompass models like

pre-trained models, self-supervised models, large

language models, language vision models, general

purpose models, multi-purpose models, and task-

agnostic model. This term encompasses their funda-

mental role: While these models are initially incom-

plete, they provide a base for developing numerous

task-speciﬁc models through adaptation. It highlights

the importance of stability, safety and security in their

architecture. The concept of FMs is shown in Fig-

ure 2. Particularly, in this ﬁgure, we position the FM

system concept in the DT paradigm in line with how

DT is presented in Figure 1.

The integration of FMs in the DT paradigm is

promising: It suggests a synergy where the adaptive

and extensive learning capabilities of FMs can en-

hance the predictive accuracy, operational efﬁciency

and decision-making processes in the DTA, as sug-

gested by Figure 1. This integration could lead to

twins that are not only representative of their real-

world counterparts but also capable of anticipating fu-

ture states, adapting to changes with minimal human

intervention and devising plans. In addition, this syn-

ergy can pinpoint a new way to interact and conﬁgure

systems in real-time.

To the best of the authors’ knowledge, there ap-

pears to be no literature on the explicit or direct inte-

gration of FMs with digital twins. While there are ex-

amples of AI being used in DTA for applications like

predictive maintenance and system optimisation (Pi-

leggi et al., 2021), the speciﬁc application of FMs in

such a context is scarce. This intersection offers a rich

area for exploration, posing new opportunity with sig-

niﬁcant challenge.

In Sec. 2, we mention related work. We then pro-

ceed to delve into the potential integration of FMs

with digital twins by highlighting areas of applicabil-

ity with so-called designated purposes in Sec. 3. In

Sec. 4, we explore beneﬁts, challenges and practical

considerations. Finally, in Sec. 5 we conclude with a

discussion and mention future work.

2 RELATED WORK

The integration of AI in the DT pradigm has been a

subject of increasing interest in both academic and in-

dustrial research. While the speciﬁc incorporation of

FMs in DT is relatively uncharted, there is a growing

body of work highlighting the use of AI formal meth-

ods in enhancing the capabilities of DTs (Rathore

et al., 2021). Below, we present a comprehensive and

cohesive narrative on the role of AI in echancing DTs

across various domains.

Aerospace: In the aerospace industry, the syn-

ergistic integration of AI with DTAs represents a

signiﬁcant technological advancement in system de-

sign, maintenance and operational efﬁciency. The

role of AI in this domain is multifaceted, encompass-

ing advanced simulations, predictive analytics and en-

hanced fault diagnostic capabilities within DT frame-

works. This integration facilitates a high-ﬁdelity

representation of aerospace systems, empowering

real-time monitoring and predictive assessments of

structural and functional parameters (Allen, 2021).

The incorporation of AI algorithms into DTs en-

ables the extraction of insightful data from complex

aerospace dynamics, aiding in the optimisation of

design processes, the elevation of safety standards,

and the reﬁnement of strategic decision-making pro-

tocols (H

anel et al., 2020).

Environmental Sciences: AI-enhanced DTs are

also being employed to assist the analysis of com-

plex systems, environmental monitoring and manage-

ment (Pylianidis et al., 2022). For instance, models

that simulate and predict environmental changes, such

as those impacting biodiversity, are increasingly in-

corporating AI for more accurate and timely predic-

tions (Trantas et al., 2023). Projects like Destination

Earth (Nativi et al., 2021) that seek to build a digi-

tal twin of the Earth use advanced data management

systems and data-driven AI technologies to generate

deep insights from complex real-world processes and

interpret them to produce actionable intelligence.

Healthcare: In the this sector, AI-driven DTAs

are being explored for personalised patient care and

treatment. A notable example is the use of DTs

in patient-speciﬁc models for predicting disease pro-

Digital Twin and Foundation Models: A New Frontier

989

Figure 2: An FM system, adapted from (Bommasani et al., 2021), envisioned as a digital twin with the capacity to consolidate

different data modalities from multiple sources, where the model output can be adapted to an array of speciﬁc downstream

tasks.

gression and treatment outcomes. These models

utilise AI algorithms to analyse patient data and pro-

vide personalised treatment recommendations (Kaul

et al., 2023).

Manufacturing: In manufacturing, digital twins

have become instrumental in advancing smart and

ﬂexible manufacturing, fault diagnosis, robotic as-

sembly, quality monitoring and job shop scheduling.

AI algorithms are increasingly being integrated into

DTAs for manufacturing and supply chain optimi-

sation. This involves using AI for analysing pro-

duction data and optimising supply chain logistics

and production plans, seamlessly integrating multiple

topological plant instances and predicting market de-

mands (Mo et al., 2023; Vysko

cil et al., 2023).

Predictive Maintenance: A widely recognised

application of AI in DT is in predictive maintenance.

For instance, a study detailed in the so-called Digi-

tal Twin Primer outlines the use of AI for predictive

maintenance in manufacturing (Borth and Broekhui-

jsen, 2020). By including AI algorithms in DTAs,

manufacturers can anticipate equipment failures and

schedule maintenance, thereby reducing downtime

and extending equipment life (van Dinter et al., 2022).

Smart Cities and Urban Planning: In urban

planning, AI-driven DTAs are used for optimising

city operations and infrastructure management. These

applications employ AI algorithms to analyse data

from various urban systems, enabling city planners

to simulate and test different scenarios for urban de-

velopment and smart mobility. An solution proposed

by (Seuwou et al., 2020), connected autonomous

vehicles are build on top of intelligent transporta-

tion systems and can inﬂuence the growth of digital

economies in large cities and alleviate problems like

excessive trafﬁc jams, road accidents, CO

-emissions

and public health deterioration.

Despite these advancements, the speciﬁc applica-

tion of FMs, which are more generalist and capa-

ble of learning from diverse data sets, in DT is still

emerging. Current research predominantly focuses

on more traditional AI algorithms tailored to speciﬁc

tasks within DT. The potential of FMs to enhance the

adaptability, predictive and autonomy power of DT

represents an exciting, albeit relatively unexplored,

research avenue.

While the role of AI in advancing DTAs is well-

established across various sectors, the integration of

FMs into this domain remains an area ripe for ex-

ploration. This research gap presents an opportunity

to develop more advanced, versatile and efﬁcient DT

solutions by leveraging the broad applicability and

learning capabilities of FMs.

3 DESIGNATED PURPOSES

The integration of FMs into the DT paradigm offers a

new dimension of functionality and potential. These

models can contribute signiﬁcantly to the evolution

of DTs, especially in areas requiring complex data

interpretation, predictive analytics, adaptive learning

and autonomous decision making. Main contribution

areas highlighted by the synergy of FM and digital

twins, which we refer to as designated purposes, are

as follows.

Enhanced Predictive Analytics: FMs, with their

ability to process and learn from vast datasets, can sig-

niﬁcantly enhance the predictive capabilities of DTs.

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

990

For instance, in the manufacturing sector, FMs can

analyze patterns from historical and real-time data

to predict equipment failures or maintenance needs

more accurately than traditional models. This ca-

pability aligns with the ﬁndings in the Digital Twin

Primer mentioned before, emphasising the impor-

tance of predictive maintenance in manufacturing us-

ing DTAs.

Adaptive Learning and Evolution: Unlike tra-

ditional models that require retraining or adjustments

for new scenarios, FMs can continuously evolve by

learning from new data inputs. This characteristic

is particularly beneﬁcial for digital twins represent-

ing complex and dynamic systems, such as urban in-

frastructures (Takeda et al., 2023) or environmental

models, where conditions can change rapidly and to

signiﬁcant extremes. The adaptive nature of FMs can

help digital twins remain accurate and relevant over

time, reﬂecting real-world changes more effectively.

Complex Data Interpretation and Simulation:

FMs are adept at handling and interpreting complex,

unstructured data, which is a common challenge in

DT. In healthcare, for example, DTAs can leverage

FMs to interpret diverse patient data, leading to more

personalised and accurate healthcare solutions. This

use case resonates with the healthcare applications

mentioned in the literature, where AI-driven patient-

speciﬁc models are becoming increasingly prevalent.

For environmental sciences, a recent example is the

Harmonised Landsat and Sentinel-2 Geospatial FM

that can support applications that include tracking

changes in land use, monitoring natural disasters and

predicting crop yield (NASA and IBM, 2023).

Generalisation Across Domains: A signiﬁcant

contribution of FMs is their ability to generalise

across different domains and tasks. This feature

can be incredibly advantageous for digital twins used

in multidisciplinary ﬁelds or applications requiring

cross-domain knowledge. For instance, in robotics an

FM trained on datasets of diverse robot demonstra-

tions can offer a generalist model that controls many

different types of robots, following diverse instruc-

tion, perform basic reasoning about complex tasks

and generalise effectively (Open X-Embodiment Col-

laboration, 2023). In healthcare, (Moor et al., 2023)

proposed Generalist Medical AI – an FM concept that

can offer bedside decision support, grounded radiol-

ogy reports and augmented procedures across numer-

ous medical tasks.

System Veriﬁcation: In contrast with traditional

AI models designed for speciﬁc tasks, FMs can han-

dle a wide range of tasks, making it difﬁcult to an-

ticipate all potential failures. Therefore, developers

and regulators must ensure thorough testing and clar-

ify approved uses. Interfaces should alert users about

”off-label usage” to prevent misinformation. Veriﬁca-

tion of these models requires multidisciplinary panels,

as they process complex inputs and outputs, demand-

ing collaborative efforts for accurate assessment.

Computational Demand and Resource Inten-

sity: The implementation of FMs as DT components

demands signiﬁcant computational resources, which

can be a limiting factor, especially for smaller-scale

applications.

Data Privacy and Security: The vast amount of

data required to train FMs raises concerns about data

privacy, particularly in sensitive domains like health-

care.

Model Interpretability: FMs, particularly be-

cause of their architecture and size, often lack trans-

parency in their decision-making processes, which

can be a critical issue in scenarios where reasoning

and explainability is essential.

4 CASE ANALYSES

The potential integration of FMs into the DT

paradigm can be best understood through speciﬁc

case analyses. Here, we examine three scenarios to

explore the potential beneﬁts and challenges of this

integration. In these examples, FMs are positioned at

the core representation models of the DTA.

Case 1: Manufacturing Process Optimisation

A. Potential Beneﬁts of Integration:

• Enhanced Predictive Maintenance: Leveraging

FMs in manufacturing DTAs can signiﬁcantly

improve predictive maintenance. These models

can analyse complex patterns from machine op-

eration data, identifying potential issues before

they lead to downtime.

• Optimised Production Processes: FMs can be

used to represent various production scenar-

ios simultaneously, using real-time and histor-

ical data, leading to more efﬁcient and cost-

effective manufacturing processes.

B. Integration Challenges:

• Real-Time Data Processing: Manufacturing

DTAs require rapid processing and analysis of

data. FMs, due to their size and complex-

ity, could struggle with real-time data analysis,

leading to delays in crucial decision-making.

• Model Pre-training and Fine-tuning: Manufac-

turing environments are highly speciﬁc, varied

and produce a lot of data from sensors. Cus-

tomising FMs and adapting them to accurately

Digital Twin and Foundation Models: A New Frontier

991

reﬂect the nuances of each manufacturing pro-

cess could be resource-intensive and techni-

cally challenging.

Case 2: Biodiversity Conservation Planning

A. Potential Beneﬁts of Integration:

• Enhanced Data Harmonisation and Analysis:

Using FMs in biodiversity DTAs can increase

the efﬁciency on data fragmentation by con-

solidating and analysing dispersed and multi-

modal data sources. This ability can lead to

more comprehensive and insightful ecological

analyses, enhancing the effectiveness of biodi-

versity research.

• Sophisticated Modelling of Complex Ecologi-

cal Systems: Leveraging FMs can signiﬁcantly

improve the modelling of complex ecological

systems. Their advanced machine learning al-

gorithms and adaptability allow for a deeper

understanding of complex, less understood eco-

logical dynamics. By assimilating these mod-

els with DTs, researchers can develop more

nuanced and accurate representations of eco-

logical systems, addressing the challenges of

complexity and scalable inter-model coordi-

nation. FMs can assist by aligning species-

speciﬁc models within a broader ecological

context, matching various temporal resolutions

and levels of abstraction with real-world eco-

logical processes.

• New ways for User Interfacing: DTAs are

mainly centered around the monitoring, pre-

diction and control elements of the underly-

ing processes than are being modelled, often

lacking effective and interactive user experi-

ence. FMs can complement this part by offer-

ing instruction-based interfacing through text,

voice and visual queues while in the same time

allowing for easier model conﬁguration with

limited expert knowledge.

B. Integration Challenges:

• Real-time Data Processing, Monitoring and

Fine-tuning: Traditional methods, such as

static species distribution maps, lack real-time

updating capabilities. Integrating FMs with

DTAs requires the development of systems ca-

pable of real-time data processing and monitor-

ing to provide up-to-date ecological informa-

tion. This challenge involves not only the tech-

nical aspect of real-time data handling but also

ensuring the continuous ﬂow and integration of

dynamic ecological data into the models. Addi-

tionally, efﬁcient algorithms for ﬁne-tuning the

FMs on the dynamic datasets are required.

• Complexity in Addressing Uncertainties: The

inherent limitations in current methods to

identify uncertainties and knowledge gaps in

ecosystems make integration complex. FMs

need to be sophisticated enough to address

these uncertainties effectively, which can be a

signiﬁcant technical and methodological chal-

lenge.

• Achieving Scalable Inter-Model Coordination:

The integration of species-speciﬁc models

within a generic ecological model is challeng-

ing due to varying research objectives and time

scales. FMs in DTAs must be designed to

match the diverse temporal resolutions and ab-

straction levels of different models with real-

world ecological processes. This requires a

delicate balance between the granularity of the

data, the scope of the models, and the overarch-

ing ecological dynamics they aim to represent.

Case 3: Smart City Infrastructure Management

A. Potential Beneﬁts of Integration:

• Comprehensive Urban Simulation: FMs can

assimilate data from various urban systems

(trafﬁc, utilities, public services) to represent

and predict urban dynamics, aiding in more ef-

fective city planning and management.

• Adaptive Response to Urban Challenges:

These models can help DTs adapt to changing

urban environments, such as ﬂuctuating trafﬁc

patterns or utility usage, ensuring efﬁcient and

sustainable city operations.

B. Integration Challenges:

• Complex Data Integration: Integrating and pro-

cessing data from diverse urban systems pose

signiﬁcant challenges, requiring advanced data

harmonisation and management strategies.

• Ethical and Privacy Concerns: The use of ex-

tensive urban data in FMs raises concerns about

individual privacy and data ethics, necessitating

stringent data governance protocols.

By no means is our analysis of each case exhaus-

tive. They serve as an initial insight into the poten-

tial value and wide range of challenges that arise.

However, in all cases, the integration of FMs in the

DTA seems to offer substantial beneﬁt, particularly

in terms of improved and enhanced representation,

prediction, planning and generalisation capabilities.

Moreover, DTAs can greatly beneﬁt from the FMs

ability to handle complex, multi-modal data. FMs can

offer new ways for end-users to engage with DTAs.

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

992

5 CONCLUSION AND FUTURE

WORK

In this paper, we make an initial attempt at explaining

a novel perspective that makes explicit the direct in-

tegration of FMs in DT. This exploration opens up a

new frontier in the intersection of AI and real-world

applications. In this paper we delve into the poten-

tial of this integration, highlighting its transformative

implications across various sectors – from manufac-

turing and healthcare, to biodiversity and aerospace.

The key value highlights explored further in our case

analyses are as summarised as follows.

• Enhanced Capabilities: The integration of FMs

with digital twins promises enhanced predictive

analytics, adaptive learning capabilities, and supe-

rior handling of complex multi-modal data. This

integration can lead to more accurate, efﬁcient,

and dynamic DT solutions.

• Broad Applicability: Known for their versatility,

FMs can be adapted to diverse DTAs, from opti-

mising manufacturing processes to personalising

healthcare treatments.

• Continuous Evolution: Unlike conventional

models, FMs can help enable digital twins to

evolve continuously, learning from new data and

adapting to changes in their real-world counter-

parts.

Future work should address the following chal-

lenges.

• Computational and Resource Demands: The

implementation of FMs in DTAs is computation-

ally intensive, necessitating signiﬁcant processing

power and specialised expertise.

• Data Privacy and Safety Concerns: In sectors

like healthcare, the use of extensive and sensitive

data in FMs raises critical questions about privacy,

security, and ethical usage.

• Transparency and Interpretability: The often

opaque nature of the FM decision-making pro-

cesses poses challenges in scenarios where ex-

plainability is crucial.

The potential integration of FMs into DTAs is an

exciting development that stands to revolutionise var-

ious sectors by providing more intelligent, adaptable

and efﬁcient systems. However, realising this poten-

tial will require addressing signiﬁcant challenges, in-

cluding managing computational demands, ensuring

data privacy and enhancing model transparency.

As we move forward, it is crucial to continue ex-

ploring this integration with a focus on safe, respon-

sible and ethical AI practices. The journey towards

fully realising the potential of FMs in DT will in-

volve interdisciplinary collaboration, continuous re-

search and a commitment to overcoming the technical

and ethical challenges that lie ahead.

ACKNOWLEDGEMENTS

This study has received funding from the European

Union’s Horizon Europe research and innovation

programme under grant agreement No 101057437

(BioDT project, https://doi.org/10.3030/101057437).

Views and opinions expressed are those of the au-

thor(s) only and do not necessarily reﬂect those of the

European Union or the European Commission. Nei-

ther the European Union nor the European Commis-

sion can be held responsible for them.

REFERENCES

Allen, B. D. (2021). Digital twins and living models at nasa.

https://ntrs.nasa.gov/citations/20210023699.

Barricelli, B. R., Casiraghi, E., and Fogli, D. (2019). A

survey on digital twin: Deﬁnitions, characteristics,

applications, and design implications. IEEE Access,

7:167653–167671.

Bommasani, R., Hudson, D. A., Adeli, E., Altman, R.,

Arora, S., von Arx, S., Bernstein, M. S., Bohg, J.,

Bosselut, A., Brunskill, E., et al. (2021). On the

opportunities and risks of foundation models. arXiv

preprint arXiv:2108.07258.

Borth, M. and Broekhuijsen, J. (2020). Ready for digital

twins. https://downloads.esi.nl/leaﬂets/TNO Digital

Twin Primer SMEplus.pdf.

Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D.,

Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G.,

Askell, A., et al. (2020). Language models are few-

shot learners. Advances in neural information pro-

cessing systems, 33:1877–1901.

Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K.

(2018). Bert: Pre-training of deep bidirectional trans-

formers for language understanding. arXiv preprint

arXiv:1810.04805.

Digital Twin Consortium (2023). Glossary of digi-

tal twins. https://www.digitaltwinconsortium.org/

glossary/glossary/. Accessed: 08-11-2023.

Hinton, G. E., Osindero, S., and Teh, Y.-W. (2006). A

Fast Learning Algorithm for Deep Belief Nets. Neural

Computation, 18(7):1527–1554.

anel, A., Schnellhardt, T., Wenkler, E., Nestler, A., Bro-

sius, A., Corinth, C., Fay, A., and Ihlenfeldt, S.

(2020). The development of a digital twin for ma-

chining processes for the application in aerospace in-

dustry. Procedia CIRP, 93:1399–1404. 53rd CIRP

Conference on Manufacturing Systems 2020.

Digital Twin and Foundation Models: A New Frontier

993

Kaul, R., Ossai, C., Forkan, A. R. M., Jayaraman, P. P., Zel-

cer, J., Vaughan, S., and Wickramasinghe, N. (2023).

The role of ai for developing digital twins in health-

care: The case of cancer care. Wiley Interdisciplinary

Reviews: Data Mining and Knowledge Discovery,

13(1):e1480.

Mo, F., Rehman, H. U., Monetti, F. M., Chaplin, J. C.,

Sanderson, D., Popov, A., Maffei, A., and Ratchev,

S. (2023). A framework for manufacturing system re-

conﬁguration and optimisation utilising digital twins

and modular artiﬁcial intelligence. Robotics and

Computer-Integrated Manufacturing, 82:102524.

Moor, M., Banerjee, O., Abad, Z. S. H., Krumholz, H. M.,

Leskovec, J., Topol, E. J., and Rajpurkar, P. (2023).

Foundation models for generalist medical artiﬁcial in-

telligence. Nature, 616(7956):259–265.

NASA and IBM (2023). Harmonized landsat and sentinel-2

geospatial foundation model. https://www.earthdata.

nasa.gov/news/impact-ibm-hls-foundation-model.

Accessed: 20-10-2023.

Nativi, S., Mazzetti, P., and Craglia, M. (2021). Dig-

ital ecosystems for developing digital twins of the

earth: The destination earth case. Remote Sensing,

13(11):2119.

Open X-Embodiment Collaboration (2023). Open X-

Embodiment: Robotic learning datasets and RT-X

models. https://arxiv.org/abs/2310.08864.

Pileggi, P., Lazovik, E., Snijders, R., Axelsson, L.-U.,

Drost, S., Martinelli, G., de Grauw, M., and Graff, J.

(2021). A lesson on operationalizing machine learning

for predictive maintenance of gas turbines. In Turbo

Expo: Power for Land, Sea, and Air, volume 84966,

page V004T05A006. American Society of Mechani-

cal Engineers.

Pylianidis, C., Snow, V., Overweg, H., Osinga, S., Kean,

J., and Athanasiadis, I. N. (2022). Simulation-assisted

machine learning for operational digital twins. Envi-

ronmental Modelling & Software, 148:105274.

Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G.,

Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark,

J., et al. (2021). Learning transferable visual models

from natural language supervision. In International

conference on machine learning, pages 8748–8763.

PMLR.

Rathore, M. M., Shah, S. A., Shukla, D., Bentafat, E., and

Bakiras, S. (2021). The role of ai, machine learning,

and big data in digital twinning: A systematic litera-

ture review, challenges, and opportunities. IEEE Ac-

cess, 9:32030–32052.

Semeraro, C., Lezoche, M., Panetto, H., and Dassisti, M.

(2021). Digital twin paradigm: A systematic literature

review. Computers in Industry, 130:103469.

Seuwou, P., Banissi, E., and Ubakanma, G. (2020). The fu-

ture of mobility with connected and autonomous ve-

hicles in smart cities. Digital twin technologies and

smart cities, pages 37–52.

Takeda, S., Kishimoto, A., Hamada, L., Nakano, D., and

Smith, J. R. (2023). Foundation model for material

science. Proceedings of the AAAI Conference on Arti-

ﬁcial Intelligence, 37(13):15376–15383.

Trantas, A., Plug, R., Pileggi, P., and Lazovik, E. (2023).

Digital twin challenges in biodiversity modelling.

Ecological Informatics, 78:102357.

van Dinter, R., Tekinerdogan, B., and Catal, C. (2022). Pre-

dictive maintenance using digital twins: A systematic

literature review. Information and Software Technol-

ogy, 151:107008.

Vysko

cil, J., Douda, P., Nov

ak, P., and Wally, B. (2023).

A digital twin-based distributed manufacturing execu-

tion system for industry 4.0 with ai-powered on-the-

ﬂy replanning capabilities. Sustainability, 15(7).

Yang, S., Nachum, O., Du, Y., Wei, J., Abbeel, P., and Schu-

urmans, D. (2023). Foundation models for decision

making: Problems, methods, and opportunities. arXiv

preprint arXiv:2303.04129.

Yuan, Y. (2023). On the power of foundation models. In In-

ternational Conference on Machine Learning, pages

40519–40530. PMLR.

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

994