COMP9414 Chap.5 STRIPS Planning and Goal Interaction
STRIPS Planning and Goal Interaction
Planning opens the successor function by representing states as fact sets and actions through preconditions, add effects and delete effects. This chapter shows how schemas become grounded actions, how the STRIPS transition preserves unmentioned facts, how BFS searches symbolic states and why goals can interfere through shared preconditions. It finishes by separating formal plan validity from real execution reliability.
What this chapter covers
- 01
Closed-world symbolic states and ground facts
- 02
Operator schemas and grounded actions
- 03
Preconditions as an applicability gate
- 04
Add and delete effects in the state transition
- 05
Forward state-space planning with BFS
- 06
Interacting goals, monitoring and replanning
Worked example · free
Repair a sample-delivery plan
- +1Apply pickup_sample and delete both the sample-at-room fact and handempty while adding carrying-sample.
- +1Identify pickup_token as the first illegal action because its handempty precondition is false.
- +1Under a separate token-storage fact, collect the token first, then obtain the sample and unlock the door.
- +1Replay every successor and verify that the final drop adds the sample-at-lab goal fact.
Key terms
- STRIPS
- A restricted planning representation using action preconditions plus add and delete effects.
- Ground fact
- A predicate statement whose variables have been replaced by named objects.
- Action schema
- A variable-based description of a family of planning actions.
- Frame assumption
- The rule that facts not named by an action effect persist into the successor state.
- Regression
- Backward reasoning from a goal through actions that could make the required facts true.
STRIPS Planning and Goal Interaction FAQ
What makes a planning action applicable?
Every grounded precondition must be present in the current fact set under the tutorial's closed-world convention. If even one required fact is absent, the action cannot execute. Diagnosis should name the first illegal action and the specific false precondition.
How are add and delete effects applied?
Remove the facts in the delete list, insert the facts in the add list and keep all other facts unchanged. This order prevents contradictory state descriptions, such as a robot being both empty-handed and carrying the same parcel after pickup.
Why can two goals interfere?
An action that achieves one goal can delete a fact needed to achieve or preserve another. Writing the goal-producing actions beside their preconditions and delete effects exposes the conflict and helps choose a valid interleaving rather than solving subgoals independently.
Exam move
Translate every planning action into three lists: preconditions, add effects and delete effects. Ground one schema with concrete objects, test applicability against a printed state and compute the successor by crossing out deletes and inserting adds. Next validate a complete plan, stopping at the first false precondition and recording the exact state that caused it.
Practise repairing the order without changing hidden capacity or storage assumptions. For interacting goals, regress each goal through its likely final action and compare the prior subgoals for shared resources or threatened facts. Finally, add one execution boundary: identify an unmodelled or changing condition and choose monitoring, conditional planning, replanning or motion feasibility as the response.
This combination preserves both formal precision and awareness of the abstraction. Create a state ledger with one row per grounded action and columns for missing preconditions, deleted facts, added facts and preserved facts. Deliberately remove one delete effect and describe the impossible state that results, then remove one precondition and show which unsafe action becomes legal.
Translate a compact PDDL action into the three-list form and back again so syntax never hides semantics. For goal interaction, draw arrows from each goal-producing action to the facts it requires and mark any fact deleted by the other goal's route. Validate the repaired sequence forward even when regression suggested it.
Run a final qualification pass that separates static map facts from changeable status facts and symbolic feasibility from geometric feasibility. Finish by stating which observation should trigger replanning during execution and which part of the existing plan can remain valid.
Build one adversarial domain variant in which a useful fact is consumed by an action, then predict the earliest point at which a formerly valid plan fails. Repair the model and the sequence separately so you do not confuse representation errors with ordering errors.
As a final recall test, describe the state before and after a grounded action without looking at PDDL syntax, then verify every changed and preserved fact against the operator lists.
Working through STRIPS Planning and Goal Interaction in COMP9414? Sia is AskSia’s AI Computer Science tutor — ask any COMP9414 STRIPS Planning and Goal Interaction question and get a clear, step-by-step explanation grounded in how COMP9414 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
