TECHNIQUES FOR GENERATIVE DESIGN BASED ON LARGE LANGUAGE MODELS

Smart overview of the Invention

The patent application introduces methods for generative design using large language models. The process begins by receiving design prompts and generating multiple design tokens through a large language model. These tokens are then used to create a conceptual design layout, which is subsequently optimized to produce a compliant design layout. This innovation aims to enhance the diversity and performance of design outputs while addressing various constraints and requirements.

Background

Generative design in architecture and engineering involves creating diverse and innovative designs that adhere to specific constraints. Traditional methods utilize computational techniques to produce architectural layouts, building structures, and urban planning solutions. These methods are employed in developing sustainable buildings, optimizing interior spaces, and creating functional urban environments. Conventional approaches like heuristic-based methods and machine learning models have certain limitations, including lack of diversity in outputs and heavy reliance on large datasets.

Challenges with Existing Methods

Heuristic-based methods such as genetic algorithms often require extensive manual intervention and fail to provide sufficient diversity in design outputs. These approaches can be computationally intensive and may not scale well for complex projects requiring high detail and compliance with numerous constraints. Machine learning models, while capable of generating innovative designs, depend heavily on large datasets and substantial computational resources. Additionally, these models might produce designs that do not fully comply with practical standards without further refinement.

Advantages of the Proposed Techniques

The disclosed techniques offer several advantages over conventional methods. By integrating algorithms like WFC (Wave Function Collapse) and genetic algorithms for design selection and mutation, these techniques reduce dependency on extensive datasets and computational resources. They enable the generation of diverse, high-performing designs that address a wide range of goals and constraints without requiring significant manual fine-tuning. This approach enhances the efficiency and effectiveness of the generative design process.

System Implementation

The system architecture supporting these techniques includes a central processing unit (CPU), system memory, display processor, and input/output devices. The CPU manages operations, running software applications that leverage the generative design techniques. Display processors handle visual outputs to display devices, while system disks provide storage for applications and data. This configuration supports various implementations, including augmented reality systems or mobile devices, facilitating the practical application of the disclosed generative design methods.