Prompts used by generate_online_problem.py and the runtime observation prompt in run_interactive_executor.py.
Identifies groups of grounded goal predicates that should be mutually exclusive during goal enumeration.
Selects the grounded predicates that are plausible goal atoms for the current instruction.
Selects which predicate schemas are relevant to the natural-language goal instruction.
Removes semantically invalid grounded goal predicates before assignment enumeration.
Checks whether one enumerated grounded goal assignment really satisfies the task instruction.
Finds exactly-one deterministic groups and picks the single true predicate inside each group.
Selects which predicate schemas should be treated as location predicates and therefore grouped by movable object for joint judging.
Jointly judges all candidate location predicates for one target movable object.
Judges the truth value of an observable from the initial scene image so the uncertain initial belief can be calibrated from visual evidence.
Judges whether one grounded predicate is true or false in the initial online-planning scene.
Groups uncertain predicates into binary, mutex, or correlated belief factors before observation-aware calibration.
Judges the truth values of applicable observation literals from sampled execution frames during online interaction.
Extracts the visible scene objects that should become problem objects for online planning.