Embodied CAD: Solver-Grounded LLM Agents for Parametric B-Rep Assembly Modeling

2026-06-30

AI Summary

Current large language models (LLMs) can generate syntactically correct CAD scripts, yet they struggle to meet the stringent demands of industrial-grade parametric B-Rep assemblies—particularly in geometric precision, editability, and solver compatibility. This work proposes the first framework that integrates solver feedback into an LLM-driven modeling loop. It employs a hierarchical CAD skill library (L0–L4) to guide an agent in iteratively selecting actions, which are mapped to typed geometric operations, executed in a CAD backend, and refined via solver feedback for planning, repair, and policy learning. By combining action grammar constraints, deterministic parameter parsing, and GRPO-style optimization, the approach achieves high executability in long-horizon, editable assembly modeling. Experiments demonstrate significant success rate improvements on multi-step mechanical, industrial equipment, and mold design tasks, while also highlighting the tension between tool invocation efficiency and long-horizon strategic accuracy.

Abstract

Large language models can write plausible CAD scripts, but reliable industrial CAD modeling requires more than syntactically valid code: every feature, placement, and assembly relation must be accepted by an exact geometric kernel while remaining editable as parametric boundary representation geometry. We present Embodied CAD, solver-grounded LLM agents for parametric B-Rep assembly modeling. Instead of generating a complete script in one pass, the agent iteratively selects actions from a stratified L0-L4 CAD skill library, resolves them into typed geometric operations, executes them in a CAD backend, and uses solver feedback to plan, repair, and learn. The framework combines action grammar constraints, deterministic parameter resolution, and solver-derived rewards for supervised warm-up and GRPO-style refinement. We evaluate Embodied CAD on multi-step mechanical, industrial equipment, and mold-oriented assembly tasks using solver-aligned metrics: executable rate, skill accuracy, operation-family accuracy, exact policy accuracy, and task completion success. The results show that solver-grounded planning executes all strong-planner workflows in the current benchmark, while learned controllers reach high executable rates and expose the remaining gap between valid tool calls and exact long-horizon policy prediction.

Problem

• parametric B-Rep
• assembly modeling
• geometric kernel
• CAD scripting
• embodied agents

Innovation

• solver-grounded
• parametric B-Rep
• LLM agents
• iterative action planning
• geometric kernel feedback