Planning Questions and Answers

Answer: C

A contingent plan includes sensing actions and branches on their results, so the executive follows a tree rather than a sequence. It is more powerful than a conformant plan and much more expensive to compute. Backup plans and multi agent cooperation are different arrangements. Whether sensing is available during execution determines which of the three planning settings applies, and it should be settled before any planner is chosen. The shared properties explain why these belong to one family, and the differences explain why more than one exists.

Answer: D

An executive compares what it observes with what the plan expected and detects when an action has failed or the world has changed unexpectedly. Detecting the divergence is what triggers replanning. Timing, prior verification and observing others are different activities. Closing the loop between what was planned and what actually happens is what distinguishes an agent from a plan generator, and it is where most practical difficulty lies. Reading the definition literally, rather than recalling a formula, is the reliable route through questions of this shape.

Answer: A

Action monitoring verifies the immediate preconditions and catches a failure at once, while plan monitoring checks whether the remainder of the plan can still succeed and catches it earlier. The second is more expensive and more useful. Timing and ordering are different checks. Detecting a problem as early as possible, rather than when it finally blocks progress, is worth the extra checking in any domain where recovery is costly. A limit that survives making every external component ideal is an intrinsic one, and worth knowing for that reason.

Answer: B

Enumerating every possible outcome produces an enormous tree, almost all of which is never executed, so planning again when something actually goes wrong is usually far cheaper. Contingent planning is reserved for cases where recovery would be impossible or the stakes are high. Feasibility, plan length and monitoring are secondary. Deferring work until it is known to be needed is the same lazy evaluation argument that appears throughout computing.

Answer: C

A spacecraft manoeuvre or any situation where communication is impossible and a failure must be handled immediately justifies preparing the branches in advance. Otherwise replanning on demand is cheaper. Plan length, absence of uncertainty and computing power are secondary considerations. Asking whether the opportunity to think again will exist, before deciding how much to think now, is the deciding question.

Answer: D

An MDP gives each action a distribution over successor states and each state or transition a reward, and the solution is a policy rather than a sequence. It is the standard model for planning under uncertainty with full observability. Partial observability gives the partially observable variant, and durations and multiple agents are further extensions. Moving from a sequence to a policy is the structural change that uncertainty forces. Working one numerical case through both cases side by side fixes the distinction better than any rule of thumb.

Answer: A

Since an action may lead to any of several states, the agent cannot know in advance which it will occupy and must be prepared with an action for each. A fixed sequence would be wrong as soon as an unexpected outcome occurred. Goals are fixed, size is a separate issue and policies are usually larger. Uncertainty in outcomes converting a sequence into a mapping is the same distinction that separates plans from policies generally.

Answer: B

The property means the history beyond the current state carries no additional information, which is what allows a policy to depend on the state alone. Where it fails, the state description must be enriched until it holds. Reachability, reward independence and termination are different conditions. An assumption that licenses a compact representation, and can be restored by enlarging the state, is worth checking explicitly whenever such a model is adopted.

Answer: C

Temporal planning gives each action a duration and allows concurrent execution, with conditions that must hold at the start, at the end or throughout. Deadlines are one kind of constraint expressible within it. Search time and ordering are different notions. Adding duration to a formalism that assumed instantaneous actions requires the whole treatment of conditions and effects to be reconsidered, which is why it is a substantial extension rather than a detail.

Answer: D

Actions running at the same time may each require or destroy a condition the other needs, so mutual exclusion between overlapping actions must be checked rather than assumed. That is the same interference the planning graph's mutex relations record. Execution by one agent, representation and goal timing are different issues. Interference between simultaneous actions is the fundamental difficulty concurrency adds, in planning as in any other concurrent setting.