15 — FAQ¶

Where should I start if I don't know fuzzy logic?¶

Read 01 — Introduction, then work through 02 — Quick Start. For a first system, Triangle, Trapezoid, default settings, and Centroid are enough.

How is Fuzzy Logic different from a Behavior Tree?¶

Fuzzy Logic computes smooth degrees and numeric decisions. A Behavior Tree organizes task sequencing and selection. They combine well: a fuzzy result becomes a score, a weight, or a task parameter.

Why is the output always the midpoint of the range?¶

Likely no rule fired, or no rule writes to that output. Check the Compile Log, the coverage of the input sets, and RuleActivations.

Why does Set Input return false?¶

There's no input with that name in the currently active system. Check the System Asset, the spelling (case doesn't matter — names are matched case-insensitively), and Get Input Names.

Remember that System Asset takes priority over Inline System: if an asset is assigned, the inline system's variables are not the ones being looked up.

Do I need to set all inputs before every Evaluate?¶

No. The component stores supplied values. Missing ones use Default Value. Clear Inputs resets all inputs to their defaults.

Can I have multiple outputs?¶

Yes. Use Evaluate to get the whole map, or Evaluate Output by name. Evaluate Single Output is meant only for a system with one output.

Why aren't the editor's results saved?¶

The Inference sliders are a temporary preview. Change the Default Value in Details if it should be the asset's starting value.

How do I import JSON into a Data Asset?¶

Open the asset and click Load Preset. The import is validated before the data changes and supports Undo. The reverse direction is Save To JSON.

Is JSON needed in a packaged game?¶

No, if the component uses a Data Asset. JSON is only needed when your game deliberately loads external data through the API.

Why is a rule marked Invalid — skipped?¶

The parser couldn't link it to the system. Hover over it or open the Compile Log: the message will point to an unknown token, variable, set, or a mixed operator.

Which defuzzification method should I choose?¶

Start with Centroid. Weighted Average suits fast systems with Singleton or symmetric outputs. The maximum-based methods are needed when the most-supported region matters more than a smooth balance across the whole curve.

Can I add a custom membership function?¶

Yes. Derive a USTRUCT from FFuzzyMembershipFunction and implement EvaluateRaw. A detailed example is in 04 — Membership Functions.

I have a hundred agents on one behaviour. Is a component on each the right thing?¶

Not past a certain size. Each component compiles and holds its own copy of the system, so a hundred of them means a hundred copies. Use UFuzzyLogicSubsystem instead — Get Game Instance Subsystem, class Fuzzy Logic — which compiles each asset once and shares it. See 14 — Architecture And Performance.

Below a few dozen agents this will not show up in a profile, and the component is easier to work with. Measure first.

Can the engine be used from multiple threads?¶

After compilation, an immutable FFuzzyInferenceEngine can be evaluated in parallel. Don't call Compile concurrently with Evaluate on the same instance. The component remains a UObject and typically runs on the Game Thread.

Where can I see a full working example?¶

Open /FuzzyLogic/Demo/FuzzyDroneArena/Maps/FuzzyDroneArena or /FuzzyLogic/Demo/FuzzyTurretDefense/Maps/FuzzyTurretDefense. The structure and how to disable the demo are described in 16 — Demo Content.