AI Glossary / Graphs and symbolic AI / Advanced
Ignore preconditions heuristic
Definition
The ignore preconditions heuristic estimates remaining planning effort using a relaxed problem in which actions no longer require their original preconditions. Every action in the grounded action set becomes applicable in every state. The relaxation removes restrictions on when actions may occur; it does not, by itself, remove their effects.
What it means
A precondition specifies something that must already hold before an action is permitted. A delivery action might require a loaded vehicle at the destination; a processing action might require a prepared input. Dropping such requirements lets the relaxed planner consider actions earlier than the original problem allows.
This can provide optimistic guidance for search. It does not make the resulting sequence operationally valid. Preconditions must still be enforced when searching for or validating an executable plan in the original model.
The number of unsatisfied goals is not generally the exact relaxed distance. One action can achieve several goals, while another action can destroy a goal already achieved. Both interactions matter.
How it works
Start with a classical planning model containing states, actions, goals, and action costs. Create a second model with the same actions and effects but empty precondition sets. The original transitions remain possible, and additional transitions may become available.
With unchanged nonnegative action costs, the optimal relaxed cost cannot exceed the original optimal cost. Computing that exact value therefore supplies an admissible lower bound. A convenient approximation is not automatically guaranteed to preserve this property.
A further relaxation can retain only positive effects that help achieve positive goal facts and discard interference. The problem then becomes selecting actions whose useful effects cover the outstanding goals. This is a set-cover formulation. Finding its minimum can be hard, and a greedy cover is not necessarily a lower bound because it may use more actions than the optimal cover.
Be explicit about which relaxation a heuristic implements. Dropping prerequisites, dropping delete effects, counting goals, and approximating set cover are different operations with different consequences.
Three practical examples
- Delivery planning: An estimate can temporarily ignore whether a package has been loaded before considering its delivery effect. The actual plan still needs a physically valid sequence of loading, movement, and unloading operations.
- Production workflows: A heuristic may ignore setup prerequisites to estimate how directly available operations cover required outputs. Real execution must restore all preparation, equipment, and material conditions.
- Planning-system comparison: Researchers compare several relaxations to determine which gives useful guidance at acceptable computational cost.
Think of it this way
Imagine estimating how many activities could deliver a project's outputs while temporarily pretending that every activity is ready to start. That removes preparation barriers. It does not guarantee that each activity delivers only one output or that completing one activity leaves every earlier output intact.
Worked example
Consider a task whose initial state contains no facts and whose goals are ReportReady and ArchiveReady. There are two actions, each costing 1. Prepare has no preconditions and adds Prepared. PublishBundle requires Prepared and adds both ReportReady and ArchiveReady. Neither action deletes anything.
In the original model, the shortest plan is Prepare followed by PublishBundle, costing 2. PublishBundle cannot be applied initially because Prepared is false.
After dropping preconditions, PublishBundle can execute immediately. The optimal relaxed cost is 1, because that one action achieves both goals. Counting two unsatisfied goals would give 2, so goal count is not the exact relaxed distance.
The distinction also affects admissibility. Consider the original state after Prepare has already completed. Both goals remain unsatisfied, but only PublishBundle is needed. The true remaining cost is 1. A heuristic that simply reports the two unsatisfied goals would overestimate it and therefore fail the admissibility requirement.
For a second, independent example, start with neither A nor B true and require both. Action MakeA has no preconditions, adds A, and deletes B. Action MakeB has no preconditions, adds B, and deletes A. These are the only actions.
Ignoring preconditions changes nothing in this second problem, because the actions already have none. No sequence can achieve both goals simultaneously: whichever action executes last removes the other goal. Achieving each goal individually in one step does not establish that their conjunction is reachable.
If delete effects are removed as an additional relaxation, MakeA followed by MakeB achieves both in two steps. That is a different relaxed problem.
Illustrative case study
Fictional Cedar Workflow uses the count of unfinished deliverables to prioritize search. The team labels the count an exact ignore-preconditions estimate and assumes it is always admissible.
A reviewer constructs a bundled publishing action that creates two deliverables at once. The count overestimates the remaining work after preparation. A separate example reveals that two actions can undo each other's effects even after their prerequisites are removed.
The team renames the simple count, records its limitations, and checks any optimality claims against the heuristic actually used. It also replays operational plans using the original preconditions and effects.
Common misconceptions and mistakes
“Removing preconditions removes delete effects.” Effects remain unless another relaxation removes them.
“Every goal costs one action, so add the goals.” Shared achievers can make that count too large.
“Every goal can be achieved individually, so all can be achieved together.” Actions can interfere.
“A greedy relaxed solution is automatically admissible.” Its cost can exceed the relaxed optimum and may exceed the original optimum.
“An optimistic estimate is an executable plan.” Original prerequisites still determine legal execution.
Business use
Record the exact model transformation and the quantity the heuristic estimates. If the system promises minimum-cost plans, verify the admissibility or other required properties of its actual search configuration.
Test small cases with bundled effects, conflicting effects, missing achievers, and already satisfied goals. Compare estimates with exact solutions where feasible.
People also ask
Does removing preconditions guarantee a solution?
No. Required facts may have no achiever, or remaining effects may conflict.
Can some preconditions be removed while others remain?
Yes. That defines a different relaxation and can preserve more useful structure.
Is goal count always useless?
No. It can guide search, but its limitations and guarantees must be stated correctly.
Check your understanding
Why does PublishBundle cost only one step in the relaxed initial state?
Answer: Its preparation precondition has been removed.
Why can goal count overestimate?
Answer: A single action can satisfy both outstanding goals.
Why is the MakeA/MakeB problem still impossible?
Answer: Each action deletes the goal established by the other.
Related terms
- Precondition
- Planning heuristic
- relaxation
- goal count
- delete effect
- admissibility
- set cover
- action interference
Learn next
Build a tiny action model and compare four estimates: unfinished-goal count, exact precondition relaxation, delete relaxation, and a goal-cover approximation. Identify which transformations account for shared achievers and which ignore interference.
More in Graphs and symbolic AI
- Identification in the limit - Identification in the limit is a criterion for learning success in which a learner eventually settles on a correct hypothesis and never changes it afterward.
- Identity uncertainty - Identity uncertainty is uncertainty about whether different observations, names, or records refer to the same underlying entity.
- Ignore delete lists
- Imitation learning - Imitation learning is a family of methods that learn how to perform a task from demonstrations of another agent's behavior.
- Imperfect information - Imperfect information is a condition in a game where a player making a decision cannot distinguish all relevant situations that may have occurred.
- Implementation level - The implementation level describes the concrete mechanisms that make an information-processing system operate.
Sources
AIMA glossary, ignore preconditions heuristic: https://github.com/aimacode/aima-glossary/blob/master/glossary.md Carnegie Mellon University, Planning II, precondition relaxation and set cover: https://www.cs.cmu.edu/~arielpro/15381f16/c_slides/781f16-6.pdf University of Pisa, AI Fundamentals planning lecture: https://elearning.di.unipi.it/mod/resource/view.php?id=3697
Previous: Ignore delete lists Next: Ill-conditioning