TO-Agents: A Multi-Agent AI Pipeline for Preference-Guided Topology Optimization

2026-05-23 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Expert, extended

Summary

The TO-Agents framework introduces a multi-agent AI pipeline designed to bridge natural-language design intent with iterative topology optimization. This system translates human-provided problem descriptions into validated solver inputs, executes a topology optimization solver like pyFANTOM, renders the resulting 3D structure, and employs a separate judge agent for multiview vision-language reasoning to critique and revise solver parameters. Evaluated on a cantilever beam benchmark and a phone-stand product design, where the goal was hierarchically branched structures, TO-Agents achieved at least one preference-aligned design in 60% of trials for each case study. This represents up to 6x more successful outcomes compared to an ablated pipeline lacking visual or historical feedback. The system utilizes Qwen2.5-VL-7B-Instruct for vision and Gemma-3-27b-it for judging, demonstrating its ability to identify effective parameter levers and recover from poor revisions.

Key takeaway

For AI Engineers and Research Scientists aiming to operationalize computational design tools with qualitative intent, TO-Agents offers a robust framework. You should consider integrating multi-agent systems with vision-language models to autonomously refine designs based on high-level preferences. This approach shifts focus from low-level parameter tuning to higher-level specification, but requires careful safeguards against failure modes like overshooting or selective memory.

Key insights

Multi-agent AI can autonomously translate qualitative design intent into optimized structures through iterative visual and historical feedback.

Principles

• Iterative feedback loops improve autonomous design alignment.
• Multi-agent systems learn effective parameter manipulation.
• Visual reasoning enhances 3D structural optimization.

Method

The TO-Agents pipeline converts natural language intent to validated solver inputs, runs topology optimization, renders 3D results, and uses a judge agent for visual-language critique and parameter revision over multiple cycles.

In practice

• Use multi-agent systems for complex design tasks.
• Implement vision-language models for visual feedback.
• Incorporate judge agents for objective evaluation.

Topics

• Topology Optimization
• Multi-Agent Systems
• Vision-Language Models
• Generative Design
• Engineering Design Automation
• Additive Manufacturing

Code references

• bellastewart/TO-Agents

Best for: AI Scientist, AI Engineer, Research Scientist

Related on AIssential

• A Lightweight Multi-Agent Framework for Automated Concrete Barrier Design — A multi-agent framework improves LLM-based engineering design by integrating generation, evaluation, and optimization, achieving high accuracy with smaller models.
• IterCAD: An Iterative Multimodal Agent for Visually-Grounded CAD Generation and Editing — IterCAD enables closed-loop, iterative CAD generation and editing by integrating multi-modal feedback and self-correction.
• HyEvo for Topological Reasoning Graph Optimization — A self-evolving system autonomously designs optimal hybrid LLM-code agent workflows, eliminating manual architecture design.
• Prompt Engineering for Agentic AI — Agentic AI prompting requires architectural context engineering, not just better phrasing, to ensure reliable multi-step behavior.
• Agentic AI Vision System: Object Segmentation with SAM 3 and Qwen — Agentic AI systems combine VLMs and segmentation models for iterative, self-correcting visual understanding and object segmentation.
• AlphaJet: Automated Conceptual Aircraft Synthesis via Disentangled Generative Priors and Topology-Preserving Evolutionary Search — AlphaJet automates conceptual aircraft design using generative AI and evolutionary search for real-time 3D synthesis.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.