Reproduce “DeciCAD: Harness-Grounded Learning of Action Policies for CAD Agents” 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
Yufeng Peng, Yixin Zhang, Ziyi Liu, Zhikun Xu, Jiarong Yang, Sijie Wang, Wenjie Feng, Qiyu Kang
DeciCAD uses typed CAD operations bound to a persistent document, with compact feedback and targeted inspection to guide repairs. On P3D-Text-to-3D, it reports F1 of 43.63 versus 40.44 for Claude Code using the same frontier model, with geometry for 100% of tasks in both; output tokens average 10.54 thousand versus 85.57 thousand per task, not total computation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.