Towards High-Fidelity CAD Generation via LLM-Driven Program Generation and Text-Based B-Rep Primitive Grounding

The field of Computer-Aided Design (CAD) generation has made significant progress in recent years. Existing methods typically fall into two separate categories: parametric CAD modeling and direct boundary representation (B-Rep) synthesis. In modern feature-based CAD systems, parametric modeling and B-Rep are inherently intertwined, as advanced parametric operations (e.g., fillet and chamfer) require explicit selection of B-Rep geometric primitives, and the B-Rep itself is derived from parametric operations. Consequently, this paradigm gap remains a critical factor limiting AI-driven CAD modeling for complex industrial product design. This paper presents FutureCAD, a novel text-to-CAD framework that leverages large language models (LLMs) and a B-Rep grounding transformer (BRepGround) for high-fidelity CAD generation. Our method generates executable CadQuery scripts, and introduces a text-based query mechanism that enables the LLM to specify geometric selections via natural language, which BRepGround then grounds to the target primitives. To train our framework, we construct a new dataset comprising real-world CAD models. For the LLM, we apply supervised fine-tuning (SFT) to establish fundamental CAD generation capabilities, followed by reinforcement learning (RL) to improve generalization. Experiments show that FutureCAD achieves state-of-the-art CAD generation performance. Code and dataset are available at https://github.com/JohanStackk/FutureCAD.

Computer-aided design (CAD) tools are widely used in engineering and manufacturing, but creating detailed 3D models often requires significant expertise and time. Existing AI methods for CAD generation usually struggle to produce realistic industrial designs because they cannot effectively combine high-level design procedures with detailed geometric structures. In this work, we develop FutureCAD, an AI system that generates editable CAD models directly from natural language descriptions. Our method combines large language models with a geometric grounding module that helps the system identify and manipulate specific parts of a 3D design, such as edges and surfaces, during the modeling process. This allows the generated models to support complex industrial design operations including fillets, chamfers, and shells. We also construct a large-scale dataset of real-world CAD models to train and evaluate the system. Experiments show that our approach produces more accurate and reliable CAD models than previous methods.