Aligning Constraint Generation with Design Intent in Parametric CAD

ICCV 2025

Evan Casey
Tianyu Zhang
Shu Ishida
William P. McCarthy
Amir Khasahmadi
Joseph George Lambourne
Pradeep Kumar Jayaraman
Karl D.D. Willis

Autodesk Research
[arxiv]




We post-train constraint generation models using solver feedback with the goal of aligning them to remove all degrees of freedom (fully-constrained) while avoiding geometric distortion, over-constraining, and solve failures -- a key first step towards practical auto-constraining tools that preserve design intent.

Abstract

We introduce alignment techniques from reasoning large language models (LLMs) to the task of generating engineering sketch constraints in computer-aided design (CAD) models. Engineering sketches are composed of geometric primitives (such as points and lines) connected by constraints (such as perpendicularity and tangency) that define their relationships. For a design to remain easily editable, these constraints must accurately capture design intent, ensuring that geometry updates predictably as parameters change. While current methods can generate CAD designs, aligning model outputs with design intent—what we call "design alignment"—remains an open challenge. A crucial first step is to generate constraints that fully constrain all geometric primitives without over-constraining or distorting the sketch geometry. By training an existing constraint generation model with alignment techniques and feedback from a constraint solver, we achieve a 93% rate of fully-constrained sketches, compared to 34% using a naïve supervised fine-tuning (SFT) baseline and only 8.9% without SFT. Our approach is model-agnostic and paves the way for further research bridging alignment strategies between language and design domains.


Background




Parametric CAD sketches require constraints that capture design intent. Traditional approaches focus on geometric correctness, but our work emphasizes aligning generated constraints with what designers actually intend, ensuring sketches remain editable and predictable as parameters change.

Method




We introduce a novel framework that adapts alignment techniques from large language models to parametric CAD constraint generation. Our method employs iterative training with constraint solver feedback, combining supervised fine-tuning (SFT) with reinforcement learning from human feedback (RLHF) techniques including Direct Preference Optimization (DPO), Expert Iteration (ExIT), and Reinforce Leave-One-Out (RLOO).


BibTeX

@inproceedings{casey2024aligningconstraintgeneration,
  title={Aligning Constraint Generation with Design Intent in Parametric CAD}, 
  author={Evan Casey and Tianyu Zhang and Shu Ishida and William P. McCarthy and Amir Khasahmadi and Joseph George Lambourne and Pradeep Kumar Jayaraman and Karl D.D. Willis},
  booktitle={ICCV},
  year={2025},
}