CME-CAD: Heterogeneous Collaborative Multi-Expert Reinforcement Learning for CAD Code Generation

2025-12-29

AI Summary

Industrial CAD modeling automation faces three core challenges: high geometric fidelity, parametric editability, and minimal reliance on human annotations. Existing sketch-, image-, or text-driven approaches suffer from geometric distortion, non-adjustable parameters, and heavy dependence on labor-intensive ground-truth annotations. To address these, we propose a novel heterogeneous multi-expert collaborative reinforcement learning paradigm, featuring a two-stage training framework (MEFT + MERL) that integrates CADQuery code generation, orthogonal projection-based modeling, dimensional constraint injection, and chain-of-thought prompting. We introduce CADExpert—the first open-source benchmark for CAD code generation—comprising 17,299 diverse, real-world industrial instances. Experiments demonstrate that our method achieves state-of-the-art performance on CADExpert, significantly outperforming prior works in geometric accuracy, constraint compliance, and model editability. It enables end-to-end generation of executable, parameter-modifiable CAD code suitable for industrial deployment.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Computer-Aided Design (CAD) is essential in industrial design, but the complexity of traditional CAD modeling and workflows presents significant challenges for automating the generation of high-precision, editable CAD models. Existing methods that reconstruct 3D models from sketches often produce non-editable and approximate models that fall short of meeting the stringent requirements for precision and editability in industrial design. Moreover, the reliance on text or image-based inputs often requires significant manual annotation, limiting their scalability and applicability in industrial settings. To overcome these challenges, we propose the Heterogeneous Collaborative Multi-Expert Reinforcement Learning (CME-CAD) paradigm, a novel training paradigm for CAD code generation. Our approach integrates the complementary strengths of these models, facilitating collaborative learning and improving the model's ability to generate accurate, constraint-compatible, and fully editable CAD models. We introduce a two-stage training process: Multi-Expert Fine-Tuning (MEFT), and Multi-Expert Reinforcement Learning (MERL). Additionally, we present CADExpert, an open-source benchmark consisting of 17,299 instances, including orthographic projections with precise dimension annotations, expert-generated Chain-of-Thought (CoT) processes, executable CADQuery code, and rendered 3D models.

Problem

• Automates high-precision, editable CAD model generation
• Overcomes non-editable, approximate outputs from sketch-based methods
• Reduces manual annotation reliance in text/image inputs

Innovation

• Heterogeneous collaborative multi-expert reinforcement learning paradigm
• Two-stage training with fine-tuning and reinforcement learning
• Open-source benchmark with orthographic projections and CAD code