03 — Core Concepts¶

System¶

FFuzzySystem contains:

Inputs — input variables;
Outputs — output variables;
Rules — rule text in authoring order;
Settings — the mathematical inference parameters.

A system can have multiple outputs. The same inputs and rules can simultaneously drive aggression, speed, and altitude.

Range¶

FFuzzyRange defines a value domain [Min, Max]. Max must be greater than Min. An input outside the range is clamped to the nearest bound before fuzzification.

The range should match the physical meaning of the quantity: Health 0…100, Temperature -40…120, Aggression 0…1. A range that's too wide compresses the useful part of the curves; one that's too narrow often saturates values at the edges.

Variable¶

FFuzzyVariable has a name, a range, a default value, and a set of fuzzy sets.

A name starts with a letter and contains only letters, digits, and _.
Names must be unique among all inputs and outputs.
Default Value is used for an input if code hasn't supplied a value.
For an output, Default Value doesn't participate in inference.

Fuzzy Set¶

FFuzzySet pairs a name like Near with a membership function. The function returns a number μ(x) within [0,1]:

0 — the value doesn't belong to the set at all;
1 — it fully belongs;
an intermediate number — partial membership.

For a single input, several sets can be true at once. For example, a distance of 700 might have Near = 0.3 and Mid = 0.7.

Rule¶

A rule links fuzzy statements about inputs to a single output set:

IF Distance IS Near AND Health IS Low THEN Aggression IS Cautious WITH 0.8

The premise's result is the rule's firing strength. WITH 0.8 additionally caps its influence.

Fuzzification¶

Crisp inputs are converted into degrees of membership. The result is stored in InputMemberships with keys of the form Variable.Set, e.g. Distance.Near.

Aggregated Surface¶

Each rule shapes part of the output set. FFuzzyAggregatedSurface is a discretized curve obtained after combining all such parts for a single output. Most strategies convert this curve into a number.

Defuzzification¶

Defuzzification returns a crisp result from the aggregated surface. The most common method, Centroid, computes the center of gravity of the area under the curve.

Result¶

FFuzzyInferenceResult contains not just Outputs but also an explanation of the decision:

Field	Content
`bSuccess`	Whether inference completed without errors
`InputMemberships`	Degrees of membership of the inputs
`OutputActivations`	Activations of the output sets
`Outputs`	Crisp numbers by output name
`RuleActivations`	Strength of each authored rule
`Diagnostics`	Errors, warnings, and info

Data Asset and Inline System¶

UFuzzySystemAsset stores a single system as Unreal content. This is the recommended option when the behavior is used by several actors or edited by a designer.

UFuzzyLogicComponent also has an Inline System. It's useful for a one-off prototype. If a System Asset is assigned, it takes priority over the inline data.

Detailed shape selection is described in 04 — Membership Functions.