Exploiting 3D Variational Autoencoders for Interactive Vehicle Design
DS 116: Proceedings of the DESIGN2022 17th International Design Conference
Year: 2022
Editor: Mario Štorga, Stanko Škec, Tomislav Martinec, Dorian Marjanović
Author: Sneha Saha (1), Leandro L. Minku (2), Xin Yao (2,3), Bernhard Sendhoff (1), Stefan Menzel (1)
Series: DESIGN
Institution: 1: Honda Research Institute Europe GmbH, Germany; 2: University of Birmingham, United Kingdom; 3: Southern University of Science and Technology, China
Section: Artificial Intelligence and Data-Driven Design
Page(s): 1747-1756
DOI number: https://doi.org/10.1017/pds.2022.177
ISSN: 2732-527X (Online)
Abstract
In automotive digital development, 3D prototype creation is a team effort of designers and engineers, each contributing with ideas and technical evaluations through means of computer simulations. To support the team in the 3D design ideation and exploration task, we propose an interactive design system for assisted design explorations and faster performance estimations. We utilize the advantage of deep learning-based autoencoders to create a low-dimensional latent manifold of 3D designs, which is utilized within an interactive user interface to guide and strengthen the decision-making process.
Keywords: data-driven design, collaborative design, 3D modelling