AI & Behavior Trees
Unreal Engine AI — Behavior Trees, Blackboard, AIController, NavMesh, AI Perception, EQS, and techniques for debugging and scaling AI characters.
12 questions
JuniorTheoryVery commonWhat is a Behavior Tree in Unreal AI, and how does it drive decisions?
What is a Behavior Tree in Unreal AI, and how does it drive decisions?
A Behavior Tree is a visual asset that drives AI decision-making. It runs from the root through Composite nodes (Selector, Sequence) to Task leaves, reading state from the Blackboard to choose actions.
Common mistakes
- ✗Confusing the Behavior Tree (logic) with the Blackboard (data)
- ✗Mixing up
Selector(run until success) andSequence(run until failure) - ✗Putting decision logic in Tasks instead of using
Decoratornodes
Follow-up questions
- →How does a
Selectordiffer from aSequencecomposite node? - →What role does a
Servicenode play inside a Behavior Tree?
MiddleCodeVery commonHow do you make an AI chase the player in Unreal Engine?
How do you make an AI chase the player in Unreal Engine?
Store the player as an Object Blackboard key (set via perception), then a Behavior Tree Sequence runs a MoveTo task targeting that key. MoveTo uses the NavMesh to path each tick, and a Decorator aborts the branch when the key clears.
Common mistakes
- ✗Teleporting via
SetActorLocationinstead of pathing throughMoveTo - ✗Caching the player's position once instead of re-reading the live key
- ✗Forgetting a NavMesh, so
MoveTosilently fails to move the pawn
Follow-up questions
- →How does
AcceptanceRadiusonMoveToaffect when the chase task succeeds? - →How would you make the AI predict and intercept the player's movement?
MiddleTheoryVery commonHow do Behavior Trees work in Unreal Engine?
How do Behavior Trees work in Unreal Engine?
A Behavior Tree is run by UBehaviorTreeComponent, which ticks nodes top-down left-to-right. Composites (Selector, Sequence) route flow, Task leaves act, Decorator conditions gate branches, and Service nodes update data on a schedule.
Common mistakes
- ✗Thinking nodes run in parallel rather than via ordered top-down traversal
- ✗Confusing
Service(periodic data updates) withTask(leaf actions) - ✗Believing
Decoratorconditions are evaluated only once at startup
Follow-up questions
- →What does the
Observer Abortssetting on aDecoratordo? - →How does a
Tasknode signal success or failure back to its parent composite?
JuniorTheoryCommonWhat is the AIController responsible for in Unreal Engine?
What is the AIController responsible for in Unreal Engine?
An AIController is the non-player brain that possesses a Pawn. It runs the Behavior Tree, owns the Blackboard and AIPerception components, issues pathfinding moves via MoveTo, and persists separately from the pawn it controls.
Common mistakes
- ✗Thinking the
AIControlleris a component attached to the pawn rather than a separate actor - ✗Believing the pawn runs its own Behavior Tree without a controller
- ✗Assuming the controller is destroyed when its pawn dies or is unpossessed
Follow-up questions
- →What is the difference between
AutoPossessAIand callingPossessmanually? - →Why can an
AIControlleroutlive thePawnit was controlling?
JuniorTheoryCommonWhat is the AI Perception system in Unreal, and what does it let an AI do?
What is the AI Perception system in Unreal, and what does it let an AI do?
AI Perception lets an AI sense the world. The AIPerceptionComponent registers senses like Sight, Hearing, and Damage, and fires events when it detects or loses stimuli so the AI can react.
Common mistakes
- ✗Forgetting to register a sense config so no stimuli are ever detected
- ✗Confusing AI Perception (sensing) with the NavMesh (movement)
- ✗Assuming
Sightignores line-of-sight blocking by walls
Follow-up questions
- →How does the
OnTargetPerceptionUpdatedevent help drive AI reactions? - →What is the difference between the
SightandHearingsenses?
JuniorTheoryCommonWhat is the Blackboard in Unreal AI?
What is the Blackboard in Unreal AI?
The Blackboard is a key-value memory shared by the Behavior Tree and its AIController. Typed keys (Object, Vector, Bool, Enum) hold AI state; Decorator and Service nodes read and write them so the tree reacts to changing data.
Common mistakes
- ✗Confusing the Blackboard (data) with the Behavior Tree (logic)
- ✗Assuming one Blackboard is shared globally across all AI instances
- ✗Thinking Blackboard keys are untyped and accept any value
Follow-up questions
- →How can a
Decoratorreact automatically when a Blackboard key changes? - →Why is each AI pawn given its own Blackboard instance at runtime?
MiddleCodeCommonHow do you implement patrol behavior for an AI in Unreal Engine?
How do you implement patrol behavior for an AI in Unreal Engine?
Place patrol-point actors, expose them on the pawn, and use a Behavior Tree Sequence: a custom Task picks the next point into a Blackboard Vector/Object key, MoveTo walks there, then Wait pauses before the index advances and loops.
Common mistakes
- ✗Assuming a Behavior Tree cannot loop — a
Sequenceunder a looping parent does - ✗Putting waypoint coordinates in the tree instead of a Blackboard key
- ✗Forgetting a
Waittask, so the AI snaps between points with no pause
Follow-up questions
- →How would you switch from patrol to chase when perception sees the player?
- →Should patrol points loop, ping-pong, or be chosen randomly — and how?
MiddleTheoryCommonHow do you implement enemy perception by sight and hearing in Unreal?
How do you implement enemy perception by sight and hearing in Unreal?
Add an AIPerceptionComponent to the AIController and configure AISense_Sight and AISense_Hearing configs. Bind OnTargetPerceptionUpdated; when a stimulus fires, write the perceived actor into a Blackboard key so the Behavior Tree reacts.
Common mistakes
- ✗Putting the
AIPerceptionComponenton the pawn instead of theAIController - ✗Forgetting targets need an
AIPerceptionStimuliSourceto be sensed - ✗Thinking one sense covers both sight and hearing without separate configs
Follow-up questions
- →What does the
Dominant Sensesetting on the perception component control? - →How do you make an AI 'forget' a target after losing sight of it?
MiddleDebuggingOccasionalHow do you debug AI behavior in Unreal Engine?
How do you debug AI behavior in Unreal Engine?
Press the apostrophe key in play to open the AI Debugger (Gameplay Debugger): it overlays the active Behavior Tree path, live Blackboard values, perception cones, NavMesh, and the EQS query. Categories toggle on the numpad to isolate the failing system. Here only navigation is bad: with no NavMesh under the bot, MoveTo can never build a path and hangs In Progress — fix navigation coverage, not the tree.
Common mistakes
- ✗Not knowing the apostrophe key opens the Gameplay Debugger live
- ✗Relying on logs alone instead of the visual Behavior Tree overlay
- ✗Forgetting to check the NavMesh (
Pkey) whenMoveTosilently fails
Follow-up questions
- →How do you use the standalone Behavior Tree visual debugger to replay a session?
- →What does the Visual Logger add beyond the live Gameplay Debugger?
SeniorTheoryOccasionalWhat is the Environment Query System (EQS) in Unreal Engine?
What is the Environment Query System (EQS) in Unreal Engine?
EQS is a spatial reasoning system: a query generates candidate items (points or actors), then Test nodes score and filter them by distance, line-of-sight, etc. It returns the best item — used for cover, flanking, or positioning decisions.
Common mistakes
- ✗Confusing EQS (scoring positions) with NavMesh pathfinding (routing)
- ✗Thinking EQS replaces the Behavior Tree rather than feeding it data
- ✗Running heavy queries every tick instead of on demand or throttled
Follow-up questions
- →What is the difference between a
Testused to filter versus to score items? - →How do you run an EQS query as a
Tasknode inside a Behavior Tree?
SeniorTheoryOccasionalHow do you optimize a scene with many AI characters in Unreal Engine?
How do you optimize a scene with many AI characters in Unreal Engine?
Use Significance Manager to scale work by relevance, raise Behavior Tree tick intervals for distant AI, throttle perception and EQS, enable LOD and animation budgeting, and consider Mass Entity for crowds. Cull or sleep off-screen agents.
Common mistakes
- ✗Ticking every AI's Behavior Tree at full rate regardless of distance
- ✗Assuming AI cost is GPU/poly-bound when it is mostly CPU logic
- ✗Running perception and EQS every frame for off-screen agents
Follow-up questions
- →When would you migrate crowd AI to the Mass Entity framework?
- →How does the
Significance Managerdecide how much work each AI gets?
