The Planner Module Flashcards

What optional predicate can be defined to enable informed (heuristic-guided) search in Picat's planner?

Answer: heuristic/2

Defining `heuristic(State, H)` provides the planner with a cost-to-goal estimate H, enabling A*-style best-first search.

For `heuristic/2` to guarantee optimal solutions, what property must the heuristic value H satisfy?

Answer: H must never overestimate the true remaining cost (admissible)

An admissible heuristic never overestimates the true cost to goal, which guarantees that best_plan returns an optimal solution.

Which `best_plan` variant also enforces an upper bound on the search cost?

Answer: best_plan/3 (best_plan(State, Limit, Plan))

`best_plan/3` takes an explicit Limit argument and returns the cheapest plan whose cost does not exceed that limit.

What mechanism does the Picat planner use to avoid re-expanding already-visited states?

Answer: Tabling (memoization of visited states)

The planner automatically tables visited states so that the same state is never expanded twice, preventing infinite loops and redundant work.

In a `best_plan` search, what happens when two plans have the same total cost?

Answer: The planner returns whichever it finds first during its search

The planner returns the first optimal-cost plan it discovers; tie-breaking among equal-cost plans is not specified and depends on search order.

What is the minimum number of `action/4` clauses needed for a valid Picat planning problem?

Answer: At least one

At least one `action/4` clause must be defined so the planner has at least one possible transition to explore.

Can a Picat `action/4` clause have a Cost of 0?

Answer: Yes, zero-cost actions are allowed and represent free transitions

Zero is a valid non-negative cost; zero-cost actions are treated as free moves that do not increase the accumulated plan cost.