Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Zero-shot Sequential Neuro-symbolic Reasoning for Automatically Generating Architecture Schematic Designs

Milin Kodnongbua, Lawrence H. Curtis, Adriana Schulz

TL;DR

This work tackles the challenge of automating architectural schematic design for multifamily buildings by introducing a sequential neuro-symbolic reasoning framework that pairs GPT-4-based neural specification generation with formal symbolic solvers (MILP/MIQP) and feedback loops. The approach decomposes design into building-plan and floor-plan stages, each constrained and refined by neural guidance and solver verification, ensuring feasibility and high space utilization. Ablation studies show the necessity of sequential reasoning and symbolic feedback for feasible, efficient designs, while two case studies demonstrate neighborhood-aware outputs that resemble real-world patterns. The method advances practical design workflows by offering multiple valid design options rapidly, with room for future enhancements in architectural considerations, priors, and expert interaction. The work highlights a promising direction for integrating generative models with rigorous optimization to support early-stage architectural decision-making in real estate development.

Abstract

This paper introduces a novel automated system for generating architecture schematic designs aimed at streamlining complex decision-making at the multifamily real estate development project's outset. Leveraging the combined strengths of generative AI (neuro reasoning) and mathematical program solvers (symbolic reasoning), the method addresses both the reliance on expert insights and technical challenges in architectural schematic design. To address the large-scale and interconnected nature of design decisions needed for designing a whole building, we proposed a novel sequential neuro-symbolic reasoning approach, emulating traditional architecture design processes from initial concept to detailed layout. To remove the need to hand-craft a cost function to approximate the desired objectives, we propose a solution that uses neuro reasoning to generate constraints and cost functions that the symbolic solvers can use to solve. We also incorporate feedback loops for each design stage to ensure a tight integration between neuro and symbolic reasoning. Developed using GPT-4 without further training, our method's effectiveness is validated through comparative studies with real-world buildings. Our method can generate various building designs in accordance with the understanding of the neighborhood, showcasing its potential to transform the realm of architectural schematic design.

Zero-shot Sequential Neuro-symbolic Reasoning for Automatically Generating Architecture Schematic Designs

TL;DR

This work tackles the challenge of automating architectural schematic design for multifamily buildings by introducing a sequential neuro-symbolic reasoning framework that pairs GPT-4-based neural specification generation with formal symbolic solvers (MILP/MIQP) and feedback loops. The approach decomposes design into building-plan and floor-plan stages, each constrained and refined by neural guidance and solver verification, ensuring feasibility and high space utilization. Ablation studies show the necessity of sequential reasoning and symbolic feedback for feasible, efficient designs, while two case studies demonstrate neighborhood-aware outputs that resemble real-world patterns. The method advances practical design workflows by offering multiple valid design options rapidly, with room for future enhancements in architectural considerations, priors, and expert interaction. The work highlights a promising direction for integrating generative models with rigorous optimization to support early-stage architectural decision-making in real estate development.

Abstract

This paper introduces a novel automated system for generating architecture schematic designs aimed at streamlining complex decision-making at the multifamily real estate development project's outset. Leveraging the combined strengths of generative AI (neuro reasoning) and mathematical program solvers (symbolic reasoning), the method addresses both the reliance on expert insights and technical challenges in architectural schematic design. To address the large-scale and interconnected nature of design decisions needed for designing a whole building, we proposed a novel sequential neuro-symbolic reasoning approach, emulating traditional architecture design processes from initial concept to detailed layout. To remove the need to hand-craft a cost function to approximate the desired objectives, we propose a solution that uses neuro reasoning to generate constraints and cost functions that the symbolic solvers can use to solve. We also incorporate feedback loops for each design stage to ensure a tight integration between neuro and symbolic reasoning. Developed using GPT-4 without further training, our method's effectiveness is validated through comparative studies with real-world buildings. Our method can generate various building designs in accordance with the understanding of the neighborhood, showcasing its potential to transform the realm of architectural schematic design.
Paper Structure (31 sections, 4 figures, 2 tables)

This paper contains 31 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Generative AI in Computational Designs
  4. Symbolic Reasoning with PLMs
  5. PLMs in Architecture
  6. Automated Generation of Floor Plans
  7. Methods
  8. Building Plan Generation
  9. Building specification
  10. Solving for a Building Plan
  11. Neural feedback loop
  12. Floor Plan Generation
  13. Floor Specification
  14. Solving for Floor Plans
  15. Symbolic Feedback Loop
  16. ...and 16 more sections

Figures (4)

  • Figure 1: Overview of our method. Given high-level lot specification, it outputs building floor plans. Our method is a two-step process. In Step 1, we generate a building plan consisting of amenities and units to include on each floor, and in Step 2, we generate the building floor plans. In each step, we use neuro reasoning to create specifications which can then be solved using symbolic reasoning. Our method is further enhanced by feedback loops that are based on neuro reasoning in Step 1 and symbolic reasoning in Step 2.
  • Figure 2: Some of the generated floor plans for 360 W 43rd St, New York, NY (1). The generated adjacency constraints group the amenities together and put the rooftop terrace at the boundary of the building. The generated plans also exhibit high utilization of the available space.
  • Figure 3: Generated floor plans using (a) only GPT-4 in a single step (Ablation 1); and (b) only GPT-4 breaking the problem to multiple steps (Ablation 2).
  • Figure 4: Gallery of generated apartment floor plans in five different neighborhoods: Brentwood, Los Angeles, CA; Lakeview, Chicago, IL, Capitol Hill, Seattle, WA; Downtown, Baltimore, MD; and Midtown West, New York, NY. The result shows variations in the building designs and trends for each location.