CADFather Reconstructs Parametric CAD Programs Using Coordinated Tools

Key points

• A vision-language assistant inspects renders of targets and intermediate states to decide which candidate programs to extend, which tools to run, proposal quantities, and completion timing.1
• The workflow pairs learned and algorithmic proposal tools with numerical parameter optimization, executing and evaluating candidate programs to provide iterative feedback.1
• The agent tracks alternative candidate programs per part, retains the best valid result, and operates using pretrained generation and assistant models without extra training.1
• Evaluation includes reconstruction quality, execution validity, and computational cost trade-offs across DeepCAD, Fusion360, MCB, CADENA-Bench, CADBench, and BenchCAD benchmarks.1

Why it matters

• CADFather enables parametric CAD reconstruction from 3D meshes using off-the-shelf pretrained models without task-specific training, while offering measurable trade-offs between compute cost and reconstruction quality.1

Sources (2)

• [1]CADFather: Autonomous CAD Reconstruction through Coordinated Tool Use
  Hugging Face Daily Papers · Oct 6, 12:00 AM

  Reconstructing an editable CAD model from a 3D shape remains a challenging engineering task. Existing methods can propose CAD operations, but no single source of proposals works equally well across different part geometries and stages of reconstruction. We introduce CADFather, an autonomous agentic system that coordinates complementary tools to recover parametric CAD programs from 3D meshes.

  A vision-language assistant inspects renders of the target and intermediate reconstructions, then decides which candidate CAD programs to extend, which tools to invoke, how many proposals to generate, and when to finish. Learned and algorithmic tools propose CAD operations, while numerical optimization refines the parameters of existing programs. Proposed or refined programs are executed and evalu

• [2]CADFather: Autonomous CAD Reconstruction through Coordinated Tool Use
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 09:20 PM · same content

AI summary (gemini-3.8-flash); 100% of sentences verified against the sources.