Planning Questions and Answers

Answer: A

A state is a set of fluents, and everything not listed is false by the closed world assumption, which is a factored representation and is exactly what lets a planner reason about actions without enumerating states. Atomic tokens would hide the structure, distributions belong to probabilistic planning and action sequences are plans rather than states. Choosing a representation in which an action's effect can be described locally is what makes planning different from ordinary search.

Answer: B

Listing only what is true keeps a state description short, since the alternative would be to record the truth or falsity of every atom. It is appropriate because a planner's model of the world is complete by construction. Solvability, uncertainty and goal testing are unaffected. An assumption adopted because the domain genuinely satisfies it, rather than for convenience, is the legitimate use of the closed world reading.

Answer: C

A schema such as move with parameters states what must hold for it to apply and what becomes true and false afterwards, and grounding it gives the individual actions. Internal steps are below the level of the model, probabilities belong to probabilistic planning and costs are an optional addition. Describing a family of actions by one parameterised schema is what keeps a domain description short enough to write by hand.

Answer: D

The representation simply stipulates that an action changes only what its effects mention, which sidesteps the frame problem by convention rather than solving it logically. That convention is what makes the formalism usable. Explicit listing is what the convention avoids, successor state axioms are the logical alternative and recomputation is not performed. Adopting a convention that makes a hard problem disappear within a restricted formalism is a legitimate move provided its scope is understood.

Answer: A

Progression applies actions whose preconditions hold and moves forward through states, which is ordinary state space search with a planning representation. Regression works backwards, partial order planning searches plans rather than states, and refinement is hierarchical planning. Naming the direction of search correctly determines which heuristics apply and what the branching factor will be.

Answer: B

In a realistic domain hundreds of actions apply in every state and almost none advances the goal, so unguided forward search is hopeless. Good heuristics derived from relaxed problems changed that entirely and made forward search competitive again. The initial state is complete by assumption, actions are applicable regardless of the goal, and state spaces are enormous. A method dismissed as hopeless and revived by better heuristics is a reminder that a search strategy is only as good as its guidance.

Answer: C

Regression takes a set of goal conditions and asks which actions could have achieved one of them, replacing it by that action's preconditions. It considers only relevant actions, which is its advantage, but the states it manipulates are partial descriptions rather than complete ones. The other options describe different techniques. Searching backwards restricts attention to what could possibly matter, which is exactly what forward search lacks. Working one numerical case through both cases side by side fixes the distinction better than any rule of thumb.

Answer: D

Working from the goal means an action is considered only if it achieves something wanted, which prunes the irrelevant actions that swamp forward search. The price is that the descriptions manipulated are partial and consistency checking is harder. Branching depends on the domain and heuristics remain essential. Restricting attention by relevance rather than by applicability is the structural difference between the two directions. Naming the cost of a choice alongside its benefit is what turns a fact into a design decision.

Answer: A

Dropping the negative effects means nothing ever becomes false, so the relaxed problem is solvable in polynomial time and its solution length is an admissible estimate. That single idea produced the heuristics that revived forward search. Fewer objects, weaker goals and approximation are different simplifications. Deriving a heuristic from a relaxation whose optimal solution is cheap to compute is the same recipe used throughout heuristic search.

Answer: B

Because nothing is ever deleted, the set of achievable facts grows monotonically and can be computed by a fixed point iteration, so no choice among actions can ever be regretted. Action counts and preconditions are unchanged and the goal is not automatically met. Monotonicity removing the need for search is the same property that makes forward chaining over definite clauses linear, appearing here in a different guise.