CRAFT: Corrective and Robust Multi-Agent Framework for Text-to-Parametric CAD

Text-to-CAD generation has the potential to make mechanical CAD more accessible, but existing approaches face a trade-off between data dependence and output fidelity. CRAFT combines semantic understanding, parametric planning, compilation, rendering, self-correction, and component-level verification within a layered recovery process, using multi-view visual feedback to detect and repair geometric errors.

Across the NopSCADlib, ABC, and Slice-100K datasets, CRAFT produces models that stay genuinely editable — averaging 13.1 editable parameters per model (compared to 0.1 for GPT-4o) with 66.8% symbolic-expression preservation — while achieving a 0.2226 CLIP score and a 95.11 FID score.

Project page: idealab-isu.github.io/CRAFT  ·  Code: idealab-isu/CRAFT

Authors: Mohammed Musthafa Rafi, Anushrut Jignasu, Mahdi Saraeian, Chinmay Hegde, Aditya Balu, Adarsh Krishnamurthy

Keywords: Text-to-CAD, Multi-Agent LLM Systems, Parametric CAD, Self-Correction, Visual Feedback, Program Synthesis