RA-CAD: State-Aware Agent for Text-to-CAD Generation

Researchers have introduced RA-CAD (ReAct Agent for CAD), a state-aware agent designed to improve text-to-CAD generation by learning post-execution critiques. The system operates through a Generate-Execute-Critique-Rewrite loop, where it executes generated CAD code, observes the outcome, and produces an explicit critique as an intermediate policy action. This approach addresses a gap in existing methods that rely on fixed or externally supplied critique mechanisms, which may not optimize feedback interpretation. The work is detailed in a paper on arXiv (2608.05714), submitted as a new announcement. The method aims to reduce the expertise needed for parametric CAD modeling by translating natural-language design intent into editable and executable code. The agent conditions its critique on the design instruction, current code, and execution feedback, enabling more effective corrective actions throughout the generation process. The research contributes to the field of AI-driven design automation, with potential applications in architecture, engineering, and product design.