Page 21 - IJAMD-1-3

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

Figure 8. The ergonomic analysis utilizing human digital twin and the
generated objects
Abbreviation: REBA: Rapid entire body assessment.

During the interaction simulation, data on the human
body segment positions are collected and analyzed using the
ergonomics branch to assess the musculoskeletal disorder
risk on different body parts. The ergonomic assessment
involves evaluating the comfort and accessibility of the
generated objects, ensuring that the HDT model can
comfortably reach all necessary controls and interfaces,
and that all displays are within the line of sight. Ease of
use is assessed by simulating the HDT operating various
controls and using comfort metrics such as seat pressure
distribution and back support. Potential ergonomic risks,
such as awkward postures, repetitive movements, and
excessive force requirements are identified. An automated
feedback loop uses performance metrics to suggest
Figure 7. The implementation details of text-to-3D generative model design modifications, which are then used to update the
Abbreviation: CLIP: Contrastive language-image pre-training.
generative model and create new iterations of the vehicle
measures taken to ensure the removal of any duplicated seat design. This iterative process ensures continuous
test samples from the training set. To augment the diversity refinement and validation of the design through virtual
of the training dataset, each object was rendered from 100 simulations and real-world testing, optimizing the vehicle
distinct random perspectives. seat for user comfort and safety. The ergonomic analysis
process utilizing a vision-based HDT approach and the
4.2.3. Human factors engineer agent generated vehicle seat from a text-to-3D generative model
is shown in Figure 8.
To conduct an automated ergonomic analysis using an
HDT model interacting with a generated vehicle seat, the 5. Limitations and future work
process initiates with the creation of a virtual environment.
This entails developing a comprehensive 3D model of the Despite the promising potential of the CockpitGemini
framework, several limitations need to be addressed
smart vehicle cockpit, encompassing the generated vehicle in future research. First, the technology for generating
seat and other relevant components such as the steering wheel, 3D objects using large generative models is still in its
dashboard, and control interfaces. Detailed anthropometric early stages. The current state of 3D generative models
data and REBA scores are obtained from the vision-based does not yet produce outputs that meet product-level
HDT, which replicates the user’s physical characteristics quality standards. This limitation impacts the overall
and behaviors. The digital twin is programmed to perform effectiveness of the personalized design framework, as
various activities related to interacting with the vehicle seat, the generated 3D models may lack the precision and
such as sitting down, adjusting the seat position, reaching detail required for practical application in smart vehicle
for controls, and maintaining different postures, ensuring cockpits. Consequently, further advancements in 3D
that the movements are realistic and grounded in actual generative modeling are necessary to enhance the fidelity
human biomechanics. and usability of the generated designs. Second, the

Volume 1 Issue 3 (2024) 15 doi: 10.36922/ijamd.4220