Page 13 - IJAMD-1-3

International Journal of AI for
Materials and Design
Smart cockpit design with generative models

Generative models have made significant advancements space and simulate human biological responses to support
in interpreting human context and creating meaningful their decision-making and design iterations.
content. Concurrently, MAS can integrate generative
models with domain expertise to simulate collaboration 2.4. Research gaps
design processes and autonomously produce design results. In summary, despite technological advancements and
Therefore, this paper investigates the application of MAS, increasing user expectations, there remains a lack of
enhanced by multiple generative models, in achieving systematic personalized design approaches for proactive
personalized design solutions for smart vehicle cockpits. user engagement in immersive environments. To address
this research gap, we propose CockpitGemini, a generative
2.3. HDT-enabled design model-based MAS and HDT-enabled personalized design
With the increasing popularity of DTs across various framework for smart vehicle cockpits in immersive design
domains, 31-33 research on HDT has swiftly emerged, space.
incorporating humans as the foundational physical
34
entities to broaden the conventional scope of DTs. The 3. Methods
creation of HDT mainly relies on the dynamic collection of This section demonstrates the generative model-based
human-related data (e.g., external data, physiological data, MAS and HDT-empowered generic personalized
behavioral data, social interaction data, and environment design framework, main functions, and its specific
data) in the physical world, which is then sensed and implementation, respectively.
transmitted to the digital realm. This data transmission
35
facilitates feedback mechanisms that can influence human 3.1. Generic personalized design framework
physical entities directly or indirectly, such as by aiding The proposed personalized design framework, integrating
decision-making processes. 34 generative model-based MAS and HDT technologies, is
Furthermore, HDT has garnered much attention across shown in Figure 1. This framework efficiently accomplishes
various disciplines. For instance, Okegbile et al. explored fundamental design tasks, provides preliminary design
36
the opportunities and challenges of HDT for personalized results, and enables design simulation and iteration within
healthcare. Chen et al. proposed a mobile AI-generated an immersive environment. It primarily consists of four
37
content (AIGC)-based HDT system architecture and components: the physical human, the digital human, the
demonstrated its feasibility through case studies on automatic design process, and the design simulation.
personalized surgery planning and medication. Wang A physical user equipped with a virtual reality (VR)
et al. conducted a survey reviewing the development and headset can voice input their design requirements to obtain
10
future perspectives of HDT in the industrial sector. Fan design results through a multi-agent collaborative automatic
et al. leveraged a vision-based HDT modeling approach to design process. This process achieves personalized design
38
recognize human status for human-centric manufacturing. by assigning specific roles to each agent. In addition, the
Aboulsafa et al. presented an educational HDT model user must wear various physical sensors, including eye
39
for personalized e-learning. He et al. summarized the trackers, inertial measurement units (IMUs), temperature
40
progress from digital human modeling to HDT and further sensors, and others, to collect multidimensional individual
examined HDT from the perspective of human factors. data, which serves as the basis for constructing the HDT.
However, the implementation of HDT remains in The automatic design process is performed by
its early stages, despite exhibiting key characteristics multi-agent internal collaboration, enabled by multi-
of DT, such as real-time response, high fidelity, and modal generative models or algorithms such as GPT4,
interoperability. Furthermore, HDT can represent Midjourney, StyleGAN, and Dreamfusion. Each agent is
36
44
43
humans at various levels, ranging from individual endowed with capabilities for perception, decision-making,
body parameters to the entire human being. Given its and execution, allowing them to fulfill distinct roles and
41
inherent complexity, developing a comprehensive HDT execute specific tasks. For example, the consumer agent
encompassing all aspects of the human being is currently functions as an automated user analyzer, extracting user
unfeasible. Consequently, HDT should be tailored to requirement keywords from natural language inputs. The
42
specific contexts and scenarios, focusing on a particular designer agent then utilizes these extracted requirements
aspect of humans, such as specific body components, to engage modality-specific generative models, producing
physiological features, or cognitive properties. In this personalized design solutions. The engineer agent
34
paper, we develop a physiology-related HDT to extend evaluates the generated conceptual designs and optimizes
the users’ activity range for design experiences in a virtual the details based on design constraints. Finally, the UX

Volume 1 Issue 3 (2024) 7 doi: 10.36922/ijamd.4220