A Formal Grammar for Auditable and Verifiable 3D Concrete Printing via multimodal Large Language Model Suggestion-to-Action Translation

The variability of material properties and process execution in 3D Concrete Printing (3DCP) constitutes a significant barrier to its standardisation and industrial adoption. While Adaptive 3DCP with real-time sensing and ad-justment is a promising pathway to standardisation, its implementation is challenged by the integration of AI-based orchestrators. A gap exists be-tween the semantic, natural language output of Large Language Models (LLMs) and the deterministic commands required for machines to operate reliably. This paper proposes a formal grammar and parsing framework to enable LLM-driven Adaptive 3DCP by creating a reliable bridge between AI reasoning and robotic execution. The LLM output is structured into a standardised vocabulary, converting suggested adjustments directly into ac-tionable commands. The formal grammar is implemented in a computational digital twin proof-of-concept in Grasshopper, demonstrating a consistent, closed-loop response to process deviations, with corresponding sugges-tions and actions. This structured AI-machine dialogue serves as a critical enabler of standardisation, helping make 3DCP workflows reliable, audita-ble, and verifiable.