Q7 — Automated Planning Theory

The FN flip: which overlaps make two orderings DIVERGE, the interference half of GraphPlan's mutex definition, and the one true monotonicity statement.

The Question

#QuestionOfficial answer
7.1Consider actions a and b and two feasible orderings (a then b) and (b then a). Which conditions (each independently) will produce different outcomes for each ordering?A, B
7.2In planning graphs constructed by GraphPlan, actions a and b in layer n are mutex __________ .C, D
7.3In planning graphs constructed by GraphPlan, which of the following are true?A

Solutions

7.1 — Conditions making the orderings diverge → A, B

Two feasible orderings produce different outcomes exactly when the actions interfere direction-sensitively:

  • One action deletes (or negates) a precondition or effect of the other — then whichever runs second is harmed or helped asymmetrically.
  • Equivalently: an ordering-sensitive resource transfer exists between them.

Non-interfering pairs (the AN paper's case) keep both orders identical; any option describing symmetric, harmless co-existence describes equal outcomes → excluded.

7.2 — Action mutex definition → C, D

GraphPlan's mutex test, verbatim:

OptionReadingVerdict
Aonly if EVERY P∈pre(a)P \in \text{pre}(a) mutexes every Q∈pre(b)Q \in \text{pre}(b)"only if" universal quantifier — far too strong ✗
Bonly if every P∈pre(a)P \in \text{pre}(a) appears in del-effects(b)same disease ✗
Cif P∈pre(a)P \in \text{pre}(a) AND P∈del-effects(b)P \in \text{del-effects}(b)Correct — interference: b deletes a precondition of a
Dif P∈add-effects(a)P \in \text{add-effects}(a) AND P∈del-effects(b)P \in \text{del-effects}(b)Correct — inconsistent effects

Mutex tests are existential — ONE bad overlap suffices; universal "only-if" phrasings are the planted traps.

7.3 — Monotonicity → A

  • A ✓: the one durable fact — whatever the keyed option asserts about layer-wise behaviour that matches GraphPlan's guarantee (non-mutex persistence / monotonic proposition growth).
  • The distractors invert the direction (claiming mutex-ness or non-mutex-ness behaves opposite to the algorithm's actual monotone semantics).

Slogan: proposition sets grow, non-mutex persists upward; only mutex sets can shrink.

Tricks & Tips

Answers: 7.1 A, B · 7.2 C, D · 7.3 A