Planning Questions and Answers

Answer: A

Counting actions is the simplest cost measure, and weighting each action by its own cost generalises it to problems where actions differ in expense. Planning time is a separate measure of the search rather than the plan, and the other options are different quantities. Distinguishing the cost of the plan from the cost of finding it is essential, since a planner may trade one against the other deliberately.

Answer: B

A route with three long flights may cost far more than one with five short journeys, so minimising action count and minimising cost are different objectives. Planners that accept action costs optimise the second. Computation time, unrelatedness and post hoc measurement are different claims. Being explicit about which quantity is being minimised, before running an optimiser, prevents the familiar outcome of optimising the wrong thing well. A single measurement chosen to respond to the one asymmetry beats several that respond to shared behaviour.

Answer: C

A domain describes the actions and vocabulary once, and each problem instance supplies objects, an initial state and a goal. That separation lets one domain serve many problems and is what makes benchmark libraries possible. Instances, goals and physical environments are different things. Separating what is common to a class of problems from what varies between them is a modelling discipline that pays whenever more than one instance will be solved.

Answer: D

Given a domain, an instance names the objects, describes the initial state completely and states the goal condition. Grounding the schemas over the objects produces the actions. Schemas belong to the domain, and heuristics and length limits are planner settings. Knowing which half of a specification each piece of information belongs to is what allows domains to be reused and instances to be generated automatically.

Answer: A

A schema with three parameters over fifty objects gives a hundred and twenty five thousand ground actions, and many planners must enumerate them before search begins. Reachability analysis prunes most of them. Goal knowledge, representation and initial states are not the difficulty. A preprocessing step whose cost grows polynomially with a parameter that is large in practice can dominate the whole computation, which is worth checking before assuming the search is the bottleneck.

Answer: B

A reachability analysis computes which atoms can ever become true and discards every ground action requiring an unreachable precondition, which typically removes the great majority. Changing the schemas alters the domain, goal actions alone are insufficient, and solving first is circular. Using a cheap over-approximation to eliminate the impossible, before doing anything expensive, is a preprocessing pattern that recurs throughout combinatorial search.

Answer: C

Blocks world, logistics and their successors are shared domains against which competing planners are measured in periodic competitions, which has driven the field's progress and also shaped what it optimises for. Guaranteed success, industrial fidelity and internal testing are different notions. A shared benchmark accelerating progress while narrowing what is worked on is a trade every empirical field makes.

Answer: D

Domains chosen for their tractability and their tradition may share structural features that real applications lack, so a planner excellent on them may disappoint in practice. The competitions have responded by broadening the domain set. Difficulty, volatility and reproducibility are lesser concerns. A measure that becomes a target ceases to be a good measure, and the field's response has been to keep changing the target.

Answer: A

The blocks world has a handful of predicates and one or two actions, yet exhibits goal interactions that defeat naive planners, which is why it has served as an illustration for decades. Its very simplicity is what makes it useful pedagogically and limited as a benchmark. The other options misread the name. A minimal example that nonetheless contains the essential difficulty is worth more for teaching than a realistic one that obscures it. The shared properties explain why these belong to one family, and the differences explain why more than one exists.

Answer: B

The anomaly shows a goal conjunction whose parts cannot be achieved one after another without undoing the first, which defeated planners that assumed goal independence. It motivated interleaved planning and partial order approaches. Insolubility, bugs and non-termination are different phenomena. A small counterexample that refutes an assumption an entire generation of algorithms rested on is worth remembering for what it demonstrates rather than for the puzzle itself.