VLM-Based Component-Level Task Planning Framework

This document is a technical report that presents a two-phase framework for component-level task planning in construction robotics, specifically for bricklaying. The framework integrates design-stage 3D models and 2D shop drawings to generate task specifications that are executable by robots. The first phase, task formalization, involves extracting assembly logic from 2D drawings into an Assembly Representation (AsmRep) and obtaining geometric constraints from a 3D model into a Building Element Representation (ElemRep). The second phase, task compilation, utilizes a deterministic compiler to combine both representations into robot-executable specifications, detailing per-unit dimensions, 6D poses, and assembly sequences. The report discusses the limitations of existing BIM-based planning methods, which often rely on heuristic decomposition or exhaustive pre-modeling, leading to brittleness under design changes. Experimental results indicate that the proposed framework can effectively formalize assembly rules and adapt to changes, supporting scalable task planning in dynamic construction environments.