Lernzettel: Planning I — Classical Planning

Methods of AI — SoSe 2026 · 1-page exam sheet

For more depth: STRIPS (full merged hub: 3 problems, Sit. Calc., STRIPS, PDDL, heuristics, algorithms with pseudocode) · quiz_planning_30-04-26 (covers Planning I + II) · sibling: lernzettel_planning-ii_30-04-26 (MDP)

Core Ideas

  • Planning = find a sequence of actions transforming an initial state into a goal state (fully observable, deterministic, static).
  • Unlike local search (don’t care about path) or classical search (opaque states): planning exploits the internal structure of states (sets of properties).
  • Three challenges: Frame, Qualification, Ramification.
  • Two classical formalisms (lecture): Situation Calculus (deductive/logical) and STRIPS (state-based). PDDL = modern STRIPS extension. (MDPs → Planning II.)

Mini-glossary

TermMeaning
Frame problemHow to efficiently represent what does not change after an action
Qualification problemActions may have endless preconditions we can’t enumerate
Ramification problemActions may have unintended side-effects
FluentPredicate that changes over time; in Sit. Calc. carries an extra S arg
SituationTerm in Sit. Calc.; atomic (S1) or complex (put(A,B,S₁))
PRE / ADD / DELSTRIPS operator: preconditions / atoms to add / atoms to remove
CWAClosed-World Assumption — atoms not in the state are false (STRIPS yes, Sit. Calc. no)
Effect axiomSit. Calc.: describes what becomes true after an action
Frame axiomSit. Calc.: describes what stays unchanged (≈ O(actions × fluents) of them)
Ignore-PRE-DELHeuristic: drop PRE & DEL → set-cover (NP-hard, not admissible)
Ignore-DELHeuristic: drop only DEL → no op undoes goals (NP-hard, not admissible)

⚠️ Frame ≠ Qualification ≠ Ramification — most-confused trio. Have one concrete example per problem ready.

Formulas / Notation

SymbolMeaning
Fluent(args, S)Time-varying predicate in situation S
put(X, Y, S)Lecture’s situation term (not R&N’s do(a,S))
PRE(o) ⊆ sSTRIPS applicability
s' = (s \ DEL(o)) ∪ ADD(o)STRIPS state update
G ⊆ sSTRIPS goal test
(:requirements :strips :typing)PDDL flags actually used in slides

Common Exam Traps ⚠️

  • Frame ≠ Qualification ≠ Ramification. Know each by an example.
  • STRIPS uses CWA; Situation Calculus does not.
  • STRIPS “solves” the frame problem only by convention — it doesn’t dissolve it. Sit. Calc. pays explicitly via O(actions × fluents) frame axioms.
  • DEL contains positive atoms that will be removed; don’t write ¬clear(B) into ADD.
  • Ignore-PRE-DEL and Ignore-DEL are NOT admissible → A* loses optimality guarantee.
  • Lecture writes put(A,B,S₁), not R&N’s do(a, S).
  • PDDL in the slides = :strips, :typing, numeric fluents, derived predicates. :adl, conditional effects, action costs, durative actions are NOT lecture material.
  • Plan-space search (POP) is NOT in the slides — only forward / backward state-space search.
  • “Syntactic vs. ontological” frame-problem framing is external philosophy of AI — not the lecture’s framing.
  • PSPACE-completeness of classical planning is R&N background, not in the lecture.

Quick Comparison Table

FormalismState rep.Frame problemCWA?Planner
Situation CalculusLogical fluents (FOL)Explicit frame axioms (O(actions × fluents))NoTheorem prover
STRIPSSet of ground atomsImplicit (only DEL changes things)YesClassical state-space search
PDDLSTRIPS + types + numeric fluents + derived predicatesImplicit (inherits STRIPS convention)YesModern classical planners (IPC)

Quiz

Targeted exam questions in Questions for Methods of AI

  • Q30–39 (basics: Sit. Calc., STRIPS, PDDL) · Q86 (Frame/Qualification/Ramification) · Q98 (Fluents) · Q99 (PDDL) · Q114–116 (deep / exam-trap: STRIPS and the Frame Problem)

Atomic notes

See also

Tags: methods-of-ai lernzettel planning ai-generated
Full reference: STRIPS (merged hub — formerly planning-i_30-04-26)
Quiz: quiz_planning_30-04-26
Sibling: lernzettel_planning-ii_30-04-26
Superlink: Methods of AI Lecture
Questions hub: Questions for Methods of AI

Created: 30/04/26