Researchers at the Toyota Research Institute (TRI) in Los Altos, California, have introduced an innovative technique that combines engineering constraints with generative AI in the vehicle design process. This groundbreaking approach allows vehicle designers to incorporate factors such as aerodynamic drag and vehicle dimensions into the generative AI process seamlessly.
Traditionally, designers employ publicly available text-to-image generative AI tools early in their creative process. However, Toyota’s new method enables designers to input initial design sketches and engineering requirements into this process, significantly reducing the number of iterations required to align design and engineering considerations.
Avinash Balachandran, Director of TRI’s Human Interactive Driving (HID) Division, explained, “Generative AI tools are often used as inspiration for designers, but cannot handle the complex engineering and safety considerations that go into actual car design. This technique combines Toyota’s traditional engineering strengths with the state-of-the-art capabilities of modern generative AI.”
The researchers leveraged principles from optimization theory, extensively used in computer-aided engineering, to enhance text-to-image-based generative AI. The resulting algorithm empowers designers to optimize engineering constraints while maintaining text-based stylistic prompts for the generative AI process.
The first paper, titled “Drag-Guided Diffusion Models for Vehicle Image Generation,” introduces physics-based guidance that allows performance metric optimization during the image generation process. By adding drag guidance to Stable Diffusion, the researchers could generate images of novel vehicles while simultaneously minimizing their predicted drag coefficients.
In the second paper, “Interpreting and Improving Diffusion Models Using the Euclidean Distance Function,” researchers explored denoising techniques to enhance results. This led to the generation of high-quality samples in latent diffusion models.
Using Toyota’s solution, designers can input a generative AI text prompt to request various designs based on an initial prototype sketch. They can also specify stylistic properties like “sleek,” “SUV-like,” and “modern,” all while optimizing quantitative performance metrics such as aerodynamic drag or other design constraints.
Charlene Wu, Senior Director of TRI’s Human-Centered AI (HCAI) Division, emphasized the potential of this tool to accelerate Toyota’s vehicle design process by allowing engineers and designers to incorporate engineering constraints directly into the creative process.
By Impact Lab