Is a fluent always Boolean?

No. Relational fluents are truth-valued, while functional or numerical fluents can represent locations, quantities or other values.

Does a fluent require clock time?

Not necessarily. Situation calculus can index it by an action history rather than a numerical timestamp.

Is an unchanged property no longer a fluent?

No. A fluent may remain constant in a particular execution while still being capable of changing under other actions or situations.

AI Glossary / Graphs and symbolic AI / Intermediate

Fluent

Definition

A fluent is a property or quantity whose truth or value can vary between situations or over time in an action model. Examples include whether a door is open, where an object is located, or how much fuel remains. Fluents let a reasoning system describe change while keeping the meaning of the represented property consistent.

What it means

An agent needs to distinguish what is true now from what may become true after an action. Stating Open(DoorA) without context can obscure that difference. A situation-aware form such as Open(DoorA,s) says that the door is open in situation s.

The word fluent here concerns changing world descriptions, not how smoothly an AI speaks. A fluent need not change after every action; it is simply allowed to have different values in different situations. A static property, by contrast, is treated as unchanged within the model's scope. Which properties are static depends on that scope, not on a universal list.

How it works

Choose the relevant state properties and represent them as relational or functional fluents. A relational fluent expresses a truth-valued condition, such as whether a robot is carrying a box. A functional fluent can return a value, such as the robot's location or a numerical resource level.

An action model states when actions are possible and how they affect fluents. In situation calculus, a situation can be represented by the sequence of actions leading to it, and a successor situation results from performing an action. Other formalisms may use explicit time points or intervals.

The model must also describe persistence: what remains unchanged when an action occurs. This is part of the frame problem. Successor-state axioms and other formal mechanisms combine change conditions with persistence conditions. Without them, a system may know an action's direct effect but be unable to infer that unrelated properties remain true.

Three practical examples

1. A robot model uses Carrying(Robot1,BoxA,s) to describe whether the robot holds the box in a particular situation. Pickup and release actions can change that fluent.

2. A workflow model tracks Approved(Request7,s). An approval action and a revocation action affect its truth under defined preconditions, while unrelated actions can leave it unchanged.

3. A planning exercise uses Fuel(VehicleA,s) as a numerical fluent. Travel consumes a modeled amount, and refueling increases it, subject to the action model's assumptions and constraints.

Think of it this way

Think of a status panel whose labels stay the same while their displayed values change as events occur. “Door open” remains the same property even when its indicator switches from true to false. The analogy has limits: a logical fluent is part of a formal model, and its value may be unknown to the agent even though the modeled world has a definite state.

Worked example

Consider one door and three actions: OpenDoor, CloseDoor and RingBell. In this simplified deterministic model, the door starts closed. OpenDoor makes the Open fluent true, CloseDoor makes it false, and RingBell does not affect it. Assume these are the only actions and their preconditions are satisfied when used.

A successor rule can say: Open is true after action a exactly when a is OpenDoor, or Open was already true and a is not CloseDoor. From the initial closed situation, performing OpenDoor gives an open door. Performing RingBell next preserves that truth. Performing CloseDoor afterward makes Open false.

The persistence part matters. If the model said only that OpenDoor makes the door open, it would not by itself explain what RingBell leaves unchanged. The rule also encodes strong assumptions: no wind, automatic closer or failed actuator changes the door unless represented. The trace illustrates modeled change, not a guarantee that a physical door obeys every command.

Illustrative case study

Fictional case study: Maple Robot plans a delivery using a fluent for whether it carries a package. Its initial model updates location after movement but lacks a clear persistence rule for carrying. The planner therefore cannot reliably connect a pickup to a later delivery condition.

The team specifies which actions establish or terminate carrying and which preserve it. It separately models observations used to confirm a successful pickup. The case illustrates the distinction between world-state change, persistence and knowledge of that state, without claiming a completed robot deployment or measured reliability gain.

Common misconceptions and mistakes

A fluent is not an action: OpenDoor is an event or command, while Open is a condition that may result. A fluent also does not have to change at every step. It can persist through many unrelated actions.

The world's property and the agent's belief about that property are different. An unknown door state does not automatically mean closed. Likewise, issuing a command does not prove that its intended effect occurred outside the model. Finally, situation arguments must be kept consistent; mixing assertions from different situations can create conclusions that no single state supports.

Business use

Fluents help make planning and workflow rules explicit about state. Define each property, its possible values, the actions that change it and how persistence is represented. Clarify whether updates describe actual observations, intended effects or inferred consequences.

For operational systems, connect the symbolic model to evidence from execution. Decide how failed actions, concurrent events and outdated observations affect the state description. Test small action sequences that change and preserve each important fluent. This provides a more dependable foundation for planning than treating a list of current labels as if it automatically explained all future changes.

Check your understanding

Why does RingBell leave Open true in the worked trace?

Answer: Open was true beforehand, and RingBell is not the closing action. The stated successor rule explicitly preserves the fluent in that case.

Related terms

Learn next

Write a small action model with one changing property and one unrelated action. Trace its values across situations, then add an action failure or uncertain observation to distinguish the modeled state from what an agent knows.

Sources

Primary sources: Poole and Mackworth, “Situation Calculus”: https://artint.info/html1e/ArtInt_335.html ; John McCarthy, situation-calculus formalism: https://www-formal.stanford.edu/jmc/sitcalc/node2.html . The door rule and fictional robot case are original simplified teaching examples, with their action assumptions stated explicitly.

Previous: Fluency Next: Fluid Communities