Reproduce “ArtiCAD: Articulated CAD Assembly Design via Multi-Agent Code Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuan Shui, Yandong Guan, Zhanwei Zhang, Juncheng Hu, Jing Zhang, Dong Xu, Qian Yu
ArtiCAD generates editable parametric articulated CAD assemblies from text or images by planning connectors and joints before generating parts. FreeCAD's constraint solver aligns the planned attachment frames. With Gemini-3-Flash, it scores 3.41 versus 3.11 in Geometry, 3.82 versus 3.16 in Motion, and 100% versus 89.2% Success against late planning. Its kinematic-tree representation cannot model closed loops.
Key insights
- Decoupling geometry generation from assembly by front-loading the prediction of connection points via "Connector Contracts" effectively bypasses LLMs' 3D spatial reasoning limitations.
- A multi-agent system with specialized roles enhances reliability and quality by decomposing the complex task of articulated assembly design into manageable sub-problems.
- A closed-loop feedback system, incorporating VLM-based validation, cross-stage error classification, and a self-evolving experience store, significantly improves design accuracy and efficiency over time.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.