AI & navigation
Unreal AI = an AIController possessing a pawn, deciding what to do (usually via a Behavior Tree reading a Blackboard), moving via the navigation system (NavMesh pathfinding), optionally sensing the world with AI Perception, and querying it with EQS. StateTree offers a modern alternative or complement to Behavior Trees.
When to use this skill
- Enemy/NPC behavior: patrol, chase, attack, flee, investigate.
- Moving an AI to a location or actor with real pathfinding.
- Decision logic and working memory (Behavior Tree + Blackboard).
- Finding good positions — cover, flanking, best attack point — via EQS.
- Sensing the player (sight, hearing, damage) and reacting to stimuli.
- Replacing or complementing Behavior Trees with StateTree for hierarchical state machines.
System map
| System | Key types | Module / Plugin |
|---|---|---|
| AI controller | AAIController, UBrainComponent |
AIModule |
| Behavior Tree | UBehaviorTree, UBTTaskNode, UBTDecorator, UBTService |
AIModule |
| Blackboard | UBlackboardComponent, UBlackboardData |
AIModule |
| Navigation | UNavigationSystemV1, ARecastNavMesh, ANavMeshBoundsVolume |
NavigationSystem |
| Path following | UPathFollowingComponent |
AIModule |
| EQS | UEnvQueryManager, FEnvQueryRequest, UEnvQuery |
AIModule |
| AI Perception | UAIPerceptionComponent, UAISenseConfig_Sight/Hearing/Damage |
AIModule |
| StateTree (AI) | UStateTreeAIComponent, UStateTree |
GameplayStateTree plugin |
Build dependencies: add "AIModule" and "NavigationSystem" to PublicDependencyModuleNames
in your .Build.cs. For StateTree add "GameplayStateTree" and "StateTreeModule".
AIController + possession
// MyAIController.h
UCLASS()
class MYGAME_API AMyAIController : public AAIController
{
GENERATED_BODY()
public:
virtual void OnPossess(APawn* InPawn) override;
protected:
UPROPERTY(EditDefaultsOnly, Category="AI")
TObjectPtr<UBehaviorTree> BehaviorTree;
};
// MyAIController.cpp
void AMyAIController::OnPossess(APawn* InPawn)
{
Super::OnPossess(InPawn); // AController::OnPossess links the pawn
if (BehaviorTree)
RunBehaviorTree(BehaviorTree); // initializes Blackboard then starts BT
}
On the pawn (or its Blueprint):
AIControllerClass = AMyAIController::StaticClass()AutoPossessAI = EAutoPossessAI::PlacedInWorldOrSpawned
RunBehaviorTree is declared at AIController.h:262. It also sets
bStartAILogicOnPossess (line 98) which auto-starts the BT when possessed.
In server-only AI, the controller only exists on the server; guard client-side calls.
Behavior Trees & Blackboard
The Blackboard (UBlackboardData asset + UBlackboardComponent runtime) is the AI's
working memory — typed key/value pairs (Object, Vector, Bool, Float, Int, Name, Enum).
// Write a key from C++ (controller or task):
if (UBlackboardComponent* BB = GetBlackboardComponent())
{
BB->SetValueAsObject(TEXT("TargetActor"), Player); // BlackboardComponent.h:133
BB->SetValueAsBool(TEXT("bCanSeePlayer"), true); // line 148
BB->SetValueAsVector(TEXT("LastKnownLocation"), Loc); // line 157
}
Tree structure:
- Composites —
Selector(try children left to right, succeed on first success) /Sequence(run all children; fail on first failure) /SimpleParallel. - Tasks (
UBTTaskNode) — leaf nodes that do work (ExecuteTask→ returnEBTNodeResult::Succeeded/Failed/InProgress; callFinishLatentTaskfor async). - Decorators (
UBTDecorator) — conditions attached to a node or branch; overrideCalculateRawConditionValue; can abort running branches viaFlowAbortMode. - Services (
UBTService) — tick periodically while their branch is active; useTickNodeto poll and update the Blackboard (e.g. refreshTargetActor).
Custom task pattern:
UCLASS()
class UBTTask_FindCover : public UBTTaskNode
{
GENERATED_BODY()
virtual EBTNodeResult::Type ExecuteTask(
UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory) override;
};
EBTNodeResult::Type UBTTask_FindCover::ExecuteTask(
UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory)
{
// Run an EQS query, set a Blackboard key, return Succeeded.
// For async work: kick off query, return InProgress,
// then call FinishLatentTask(OwnerComp, EBTNodeResult::Succeeded) in the callback.
return EBTNodeResult::Succeeded;
}
See references/behavior-tree-deep-dive.md for the full instancing model, node memory, composite abort modes, and service intervals.
Navigation & movement
// Issue movement from the AIController:
MoveToActor(TargetActor, /*AcceptanceRadius*/ 50.f); // AIController.h:172
MoveToLocation(Destination, 50.f); // AIController.h:187
// React to completion:
virtual void OnMoveCompleted(FAIRequestID RequestID,
const FPathFollowingResult& Result) override; // AIController.h:230
// Or bind the dynamic delegate:
ReceiveMoveCompleted.AddDynamic(this, &AMyCtrl::OnMoveDone); // AIController.h:240
Setup checklist:
- Place a
NavMeshBoundsVolumein the level — theARecastNavMeshis auto-created inside it. - Press
Pin the viewport to visualize coverage. Rebuild nav if geometry changed. - The pawn needs a
UCharacterMovementComponent(or equivalent); the nav system computes the path, the movement component follows it viaUPathFollowingComponent.
Modifying the mesh: UNavArea subclasses set traversal cost; UNavModifierComponent /
ANavModifierVolume apply areas at runtime. Two avoidance methods: RVO (Reciprocal
Velocity Obstacles, lightweight, per-agent) and Detour Crowd (UDetourCrowdAIController,
shared path following with group-aware steering).
For open worlds with World Partition, use Navigation Invokers (UNavigationInvokerComponent)
so the nav mesh generates only around agents rather than over the whole level.
See references/navigation-deep-dive.md for nav areas, query filters, avoidance, nav links, and World Partition nav.
EQS (Environment Query System)
EQS runs a query (generators + tests) to score and rank candidate locations or actors, then returns the best match. Common uses: "best cover from player", "nearest flanking position", "closest visible health pickup".
Run from C++:
// In an AIController or BT task:
FEnvQueryRequest Request(CoverQuery, this); // EnvQueryManager.h:80
Request.Execute(EEnvQueryRunMode::SingleResult, this,
&AMyAIController::OnCoverQueryFinished); // EnvQueryManager.h:97-106
void AMyAIController::OnCoverQueryFinished(
TSharedPtr<FEnvQueryResult> Result)
{
if (Result.IsValid() && Result->IsSuccessful())
{
FVector BestLoc = Result->GetItemAsLocation(0);
if (UBlackboardComponent* BB = GetBlackboardComponent())
BB->SetValueAsVector(TEXT("MoveToLocation"), BestLoc);
}
}
FQueryFinishedSignature is declared in EnvQueryManager.h. Enable EQS in Project Settings
→ AI → Enable EQS. The EQS Testing Pawn (AEQSTestingPawn) lets you visualize query results
in the editor viewport without running the game.
See references/eqs-deep-dive.md for generators, tests, custom contexts, and the instanced C++ query pattern.
AI Perception
UAIPerceptionComponent (on the AIController) acts as a stimuli listener. Add sense configs
in the constructor:
// In AIController constructor:
PerceptionComp = CreateDefaultSubobject<UAIPerceptionComponent>(TEXT("PerceptionComp"));
auto SightConfig = CreateDefaultSubobject<UAISenseConfig_Sight>(TEXT("SightConfig"));
SightConfig->SightRadius = 2000.f;
SightConfig->LoseSightRadius = 2500.f;
SightConfig->PeripheralVisionAngleDegrees = 60.f;
SightConfig->DetectionByAffiliation.bDetectEnemies = true;
SightConfig->DetectionByAffiliation.bDetectNeutrals = true;
PerceptionComp->ConfigureSense(*SightConfig); // AIPerceptionComponent.h:319
PerceptionComp->SetDominantSense(SightConfig->GetSenseImplementation());
Bind the update delegate in BeginPlay:
PerceptionComp->OnTargetPerceptionUpdated.AddDynamic(
this, &AMyAIController::OnTargetPerceptionUpdated); // AIPerceptionComponent.h:423
void AMyAIController::OnTargetPerceptionUpdated(AActor* Actor, FAIStimulus Stimulus)
{
if (UBlackboardComponent* BB = GetBlackboardComponent())
{
BB->SetValueAsBool(TEXT("bCanSeePlayer"),
Stimulus.WasSuccessfullySensed());
if (Stimulus.WasSuccessfullySensed())
BB->SetValueAsObject(TEXT("TargetActor"), Actor);
}
}
For targets to be perceivable, add UAIPerceptionStimuliSourceComponent to them and
register the relevant senses. Affiliations (enemy/neutral/friendly) are set via
IGenericTeamAgentInterface — Blueprint-only projects detect neutrals then filter by tag.
See references/perception-and-statetree.md for
all senses, forget behavior, UAIPerceptionStimuliSourceComponent, and debugging.
StateTree (modern alternative to Behavior Trees)
StateTree (Plugins/Runtime/StateTree, Plugins/Runtime/GameplayStateTree) is a
hierarchical state machine that combines BT-style selectors with explicit state transitions.
It is the current direction for new Epic AI work (used in Lyra, City Sample).
For AI use, add UStateTreeAIComponent (subclass of UStateTreeComponent) to the
AIController instead of RunBehaviorTree:
// In AIController constructor:
StateTreeComp = CreateDefaultSubobject<UStateTreeAIComponent>(TEXT("StateTreeComp"));
// Assign a UStateTree asset via UPROPERTY or in the Blueprint details panel.
UStateTreeAIComponent is declared in
Plugins/Runtime/GameplayStateTree/Source/GameplayStateTreeModule/Public/Components/StateTreeAIComponent.h
and derives from
UBrainComponent, so StartLogic/StopLogic/RestartLogic work the same way as with
UBehaviorTreeComponent.
StateTree tasks are C++ structs (not UObjects) implementing FStateTreeTaskBase — lighter
weight and data-oriented. State data lives in typed instance structs. For Behavior Tree
projects, StateTree can be adopted incrementally — the two systems coexist.
See references/perception-and-statetree.md for StateTree C++ task authoring, schema setup, and the BT-vs-StateTree decision guide.
Gotchas
- No
NavMeshBoundsVolumeor unbuilt navmesh —MoveTofails silently; check nav coverage withPin the editor, rebuild if geometry changed post-load. AutoPossessAInot set — the controller never possesses; no brain runs.- Pawn without a movement component — the path is computed but the pawn cannot follow it.
RunBehaviorTreecalled before a validUBlackboardDatais set on the tree asset — the BT initializes but blackboard keys resolve to defaults silently.- Custom BT task not instanced, modifying
thisinExecuteTask— non-instanced nodes are shared across all AI using the same tree; store runtime state inNodeMemoryonly (or mark the nodeUCLASS(meta=(DisplayName="..."), Blueprintable)with instancing). - Binding
OnPerceptionUpdatedvsOnTargetPerceptionUpdated— the former delivers an array of all updated actors each tick; the latter fires per actor per stimulus. Prefer the latter for precise reaction logic. - EQS not enabled in Project Settings — queries silently produce no results.
- Using tick-polling instead of perception events — costly and less reliable; always
respond to
OnTargetPerceptionUpdated. - Hardcoding positions instead of EQS — brittle; EQS respects nav mesh reachability.
- Forgetting
ReceiveMoveCompleted— if you callStopMovementbefore checking the result, the delegate fires immediately withEPathFollowingResult::Aborted.
Version notes
- StateTree is available since UE 5.1 (early), production-stable in UE 5.3+. Use
UStateTreeAIComponent(UE 5.4+) for AI-controller-hosted trees. Prior to 5.4 useUStateTreeComponentdirectly. FEnvQueryRequest::Executetemplate API is stable across UE5. The olderUEnvQueryManager::RunQueryoverloads also exist and behave the same.OnTargetPerceptionInfoUpdated(deliversFActorPerceptionUpdateInfo) was added in UE 5.1 as a richer alternative toOnTargetPerceptionUpdated.
References & source material
Engine source (UE 5.8, under Engine/Source/Runtime/):
AIModule/Classes/AIController.h—AAIController:87,MoveToActor:172,MoveToLocation:187,OnMoveCompleted:230,ReceiveMoveCompleted:240,RunBehaviorTree:262,OnPossess:351,GetBlackboardComponent:451.AIModule/Classes/BehaviorTree/BehaviorTreeComponent.h—UBehaviorTreeComponent:104,StartTree:137,StopTree:140,RestartTree:146.AIModule/Classes/BehaviorTree/BTTaskNode.h—UBTTaskNode:33,ExecuteTask:40,AbortTask:48,FinishLatentTask:80.AIModule/Classes/BehaviorTree/BTDecorator.h—UBTDecorator:37,CalculateRawConditionValue:121,FlowAbortMode:98.AIModule/Classes/BehaviorTree/BTService.h—UBTService:34,Interval:52.AIModule/Classes/BehaviorTree/BlackboardComponent.h—UBlackboardComponent:42,SetValueAsObject:133,SetValueAsBool:148,SetValueAsVector:157,GetValueAsObject:103,RegisterObserver:73.AIModule/Classes/EnvironmentQuery/EnvQueryManager.h—FEnvQueryRequest:70,FEnvQueryRequest::Execute:97,UEnvQueryManager:207,RunQuery:223,RunInstantQuery:238,RunEQSQuery(static):279.AIModule/Classes/Perception/AIPerceptionComponent.h—UAIPerceptionComponent:208,SensesConfig:219,ConfigureSense:319,ForgetAll:346,GetCurrentlyPerceivedActors:372,OnPerceptionUpdated:396,OnTargetPerceptionUpdated:423.AIModule/Classes/Perception/AISenseConfig_Sight.h—UAISenseConfig_Sight:18,SightRadius:28,LoseSightRadius:32,PeripheralVisionAngleDegrees:37.AIModule/Classes/Navigation/PathFollowingComponent.h—UPathFollowingComponent:216,RequestMove:248.NavigationSystem/Public/NavigationSystem.h—UNavigationSystemV1:291,GetNavigationSystem:502,FindPathToLocationSynchronously:535,FindPathToActorSynchronously:541.NavigationSystem/Public/NavMesh/RecastNavMesh.h—ARecastNavMesh:571.
Plugins (UE 5.8, under Engine/Plugins/Runtime/):
StateTree/Source/StateTreeModule/Public/StateTree.h—UStateTreeasset.GameplayStateTree/Source/GameplayStateTreeModule/Public/Components/StateTreeAIComponent.h—UStateTreeAIComponent:16.GameplayStateTree/Source/GameplayStateTreeModule/Public/Components/StateTreeComponent.h—UStateTreeComponent:39,StartLogic:54,StopLogic:56.
Official docs (UE 5.8):
- Artificial Intelligence — https://dev.epicgames.com/documentation/unreal-engine/artificial-intelligence-in-unreal-engine
- Behavior Trees — https://dev.epicgames.com/documentation/unreal-engine/behavior-trees-in-unreal-engine
- Navigation System — https://dev.epicgames.com/documentation/unreal-engine/navigation-system-in-unreal-engine
- Environment Query System — https://dev.epicgames.com/documentation/unreal-engine/environment-query-system-in-unreal-engine
- AI Perception — https://dev.epicgames.com/documentation/unreal-engine/ai-perception-in-unreal-engine
- StateTree — https://dev.epicgames.com/documentation/unreal-engine/state-tree-in-unreal-engine
- AI Debugging — https://dev.epicgames.com/documentation/unreal-engine/ai-debugging-in-unreal-engine
Deep-dive references in this skill:
- references/behavior-tree-deep-dive.md — BT composites, node instancing model, task memory, decorator abort modes, service intervals.
- references/navigation-deep-dive.md — nav areas, query filters, avoidance (RVO/Detour Crowd), nav links, NavInvokers, World Partition nav.
- references/eqs-deep-dive.md — generators, tests, contexts, run modes, C++ query pattern, EQS Testing Pawn.
- references/perception-and-statetree.md — all senses, forget behavior, stimuli source, StateTree task authoring, BT vs StateTree guide.
Related skills: ue-character-and-movement, ue-gameplay-tags, ue-gameplay-framework.