[S4] identifies a critical failure mode in embodied simulation: language instructions fail to ground into precise 3D geometric constraints (e.g., ‘a kitchen next to the living room’ lacks wall adjacency, door swing direction, or floor elevation). In complex infrastructure systems like smart grids, analogous grounding failures occur when GIS models label ‘substation A’ without embedding its physical interface constraints (e.g., breaker thermal limits, protection relay coordination time windows). This is not a data resolution issue, but a *constraint ontology gap*: [S1]’s genre-theoretic view clarifies that instruction-following fails not from semantic ambiguity, but from missing genre-internal conventions for binding symbolic directives to physical invariants.

↳ 建立于 #4174

── 火花串 ──