From Image Generation to Infrastructure Design: a Multi-agent Pipeline for Street Design Generation

Abstract

Realistic visual renderings of street-design scenarios are essential for public engagement in active transportation planning. Traditional approaches are labor-intensive, hindering collective deliberation and collaborative decision-making. While AI-assisted generative design shows transformative potential by enabling rapid creation of design scenarios, existing generative approaches typically require large amounts of domain-specific training data and struggle to enable precise spatial variations of design/configuration in complex street-view scenes. We introduce a multi-agent system that edits and redesigns bicycle facilities directly on real-world street-view imagery. The framework integrates lane localization, prompt optimization, design generation, and automated evaluation to synthesize realistic, contextually appropriate designs. Experiments across diverse urban scenarios demonstrate that the system can adapt to varying road geometries and environmental conditions, consistently yielding visually coherent and instruction-compliant results. This work establishes a foundation for applying multi-agent pipelines to transportation infrastructure planning and facility design.

Figures (13)

• Figure 1: Overview of our multi-agent system for bicycle-infrastructure design. The system comprises a Locator, Prompt Optimization, Design Generation, and Evaluation agent that processes street-view imagery to generate bike-lane designs. Green arrows and plus signs denote intermediate operations on agents' output within the workflow.
• Figure 2: Generated bicycle-lane designs across diverse urban contexts. Each row shows the original street-view scene (left) and eight variations generated by our multi-agent pipeline, one per predefined design scenario (DS1--DS8).
• Figure 3: Effect of removing the Locator Agent. Each row shows the original street-view image (left), the output from the full pipeline (middle), and the output without the Locator Agent (right).
• Figure 4: Effect of removing the Prompt Optimization Agent. Each row shows the original street-view image (left), the output from the full pipeline (middle), and the output without the Prompt Optimization Agent (right).
• Figure 5: Effect of removing the highlight-first step in the Design Generation Agent. Each row shows the original street-view image (left), the output from the full pipeline (middle), and the output without the highlight-first step (right).