Research map / Large language models / Agents, code and math
Aligning Constraint Generation with Design Intent in Parametric CAD
Evan Casey, Tianyu Zhang, Shu Ishida, John Roger Thompson, Amir Khasahmadi, Joseph George Lambourne, Pradeep Kumar Jayaraman, Karl D.D. Willis
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add supervised fine-tuning on solver-verified ideal sketches | Vitruvion · SketchGraphs test | % fully constrained: 8.87 -> 34.24 | Improved |
| add RLOO post-training with solver rewards | SFT Vitruvion · SketchGraphs test | % fully constrained: 34.24 -> 93.05 | Improved |
| add GRPO post-training with solver rewards | SFT Vitruvion · SketchGraphs test | % fully constrained: 34.24 -> 91.59 | Improved |
| add ReMax post-training | SFT Vitruvion · SketchGraphs test | % fully constrained: 34.24 -> 79.84 | Improved |
| add Iterative DPO | SFT Vitruvion · SketchGraphs test | % fully constrained: 34.24 -> 64.91 | Improved |
| add Expert Iteration | SFT Vitruvion · SketchGraphs test | % fully constrained: 34.24 -> 71.70 | Improved |
| add ReMax post-training | SFT Vitruvion · SketchGraphs test | % stable (bins=4): 92.48 -> 75.77 | Hurt |
| add Iterative DPO | SFT Vitruvion · SketchGraphs test | % not solvable: 3.85 -> 7.64 | Hurt |
| use RLOO instead of Expert Iteration (online vs offline) | Vitruvion · SketchGraphs test | Pass@1: 64.09 -> 83.57 | Improved |
| increase samples from 1 to 8 (ablation) | RLOO · SketchGraphs test | success rate (Pass@K): 83.57 -> 84.96 | No clear effect |
| increase samples from 1 to 8 (ablation) | SFT · SketchGraphs test | success rate (Pass@K): 33.32 -> 42.62 | Improved |
| add RLOO post-training | SFT Vitruvion · SketchGraphs test (diversity) | % Unique@8: 46.71 -> 32.11 | Hurt |
Closest papers by meaning
- GeoCAD: Local Geometry-Controllable CAD Generation with Large Language Models (NeurIPS 2025 · similarity 0.86)
- MiCADangelo: Fine-Grained Reconstruction of Constrained CAD Models from 3D Scans (NeurIPS 2025 · similarity 0.86)
- FlexCAD: Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models (ICLR 2025 · similarity 0.86)
- CAD-Llama: Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generation (CVPR 2025 · similarity 0.85)
- CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward (NeurIPS 2025 · similarity 0.85)
- CAD-Editor: A Locate-then-Infill Framework with Automated Training Data Synthesis for Text-Based CAD Editing (ICML 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
