Tutorial 08 - Perception-driven tactical AI

TL;DR: Add an LCM Nav3D AI Perception Bridge next to your agent's UAIPerceptionComponent → its perceived threats auto-feed the LCM Nav3D | Threats From AI Perception EQS context → a true-3D LCM Nav3D EQS query (volumetric candidate points + 3D-LOS / cover / threat-exposure tests) picks a flanking or cover position → your BT/StateTree moves the agent there. Mass crowds get the same via the LCM Nav3D Mass Perception Agent trait, which writes a perceived-threats fragment.

Why this exists

Stock UE tactical AI flows through EQS, but the stock generators/tests are 2.5D (navmesh-projected). A flying or multi-level NPC can't reason about cover, line-of-sight, or flanking in true 3D with them. LCM Nav3D ships true-3D EQS - candidate points anywhere in the navigable volume, scored by volumetric SVO line-of-sight, cover, and threat exposure - and a one-click bridge so the threats those tests score against come straight from UE's own perception system.

Prerequisites

LCMNav3D plugin enabled, an ALcmNavigationManagerSVO with a built SVO in the level.
An AI agent with an AAIController and a UAIPerceptionComponent configured with an AISense_Sight config (the standard UE setup - LCM Nav3D doesn't change this).
Threats set up so perception can sense them (registered stimuli source / pawns on a hostile team), per your normal UE perception configuration.

Part A - UActor agent (BT / StateTree)

Step 1 - Add the perception bridge

On your agent (the pawn or its controller - wherever the UAIPerceptionComponent lives), Add Component → "LCM Nav3D AI Perception Bridge" (ULcmAIPerceptionBridge).

It auto-finds the owner's UAIPerceptionComponent on BeginPlay and subscribes to OnTargetPerceptionUpdated.
DecayWindowSeconds (default 5) - how long a lost-sight threat stays queryable before it fades. Tune to taste.
No other wiring: successfully-sensed actors are tracked; lost ones decay out.

That's the entire perception-to-LCM Nav3D link.

Step 2 - Build the tactical EQS query

Create an EQS query (e.g. EQS_FindFlankingApproach) and assemble it from these nodes:

Generator (pick one):

Node	Use it for
LCM Nav3D \| Points In SVO Volume	general 3D candidate cloud around the agent
LCM Nav3D \| Hemispherical Ring 3D	sniper-perch / high-ground / orbit candidates (set Upper Hemisphere Only)
LCM Nav3D \| Points Along Path	candidates strung along the LCM Nav3D path to a goal

Tests (add several; set each one's Context to LCM Nav3D | Threats From AI Perception where it scores against threats):

Node	Purpose	Typical setup
LCM Nav3D \| Threat Exposure	prefer low summed danger toward threats	Score, prefer less
LCM Nav3D \| Cover Score 3D	prefer positions occluded from threats	Score, prefer greater
LCM Nav3D \| Line Of Sight 3D	filter to (or away from) visible-to-threat	Filter
LCM Nav3D \| Reachability 3D	drop positions the agent can't path to	Filter (Context = Querier)
LCM Nav3D \| Path Distance 3D	prefer near / far by LCM Nav3D path length	Score (Context = Querier)
LCM Nav3D \| Height Advantage	favour high ground over threats	Score, prefer greater

A solid "flank to cover" query: Points In SVO Volume → Reachability 3D (filter) → Cover Score 3D (score↑) → Threat Exposure (score↓) → Path Distance 3D (score↓, mild weight).

The threats every threat-scoring test sees come from the bridge via LCM Nav3D | Threats From AI Perception - no manual tagging. (For non-perception workflows, LCM Nav3D | Known Threats returns actors tagged ApexThreat instead.)

Step 3 - Wire it into BT or StateTree, then verify

Behavior Tree: add a Run EQS Query service/task that runs your query and writes the result Vector/Actor to the blackboard, then a MoveTo (or an LCM Nav3D FlyTo) to that key.
StateTree: run the query from a task that stores the chosen location, then transition into your LCM Nav3D move task (see Tutorial 05).

Verify in PIE: the agent should pick positions that are reachable, in cover from, and low-exposure toward the threats it currently perceives - and should stop reacting to a threat ~5 s after losing sight of it (the decay window). Walk a threat out of the sight cone and watch the chosen position shift once it decays.

Part B - Mass crowd variant

For crowd-scale tactical perception, use the native LCM Nav3D Mass perception (Tutorial 06 covers Mass setup):

Add the LCM Nav3D Mass Perception Agent trait (ULcmMassPerceptionTrait) to your Mass entity config - set FactionId, SightRadius, FOVDegrees. Agents perceive different-faction agents within range + FOV + true-3D SVO line of sight.
The ULcmMassPerceptionProcessor writes each agent's closest visible hostile into its FLcmMassPerceivedThreatsFragment (ClosestThreatPos, bHasThreat, VisibleCount, LastSeenSeconds).
Read that fragment in your Mass logic (a StateTree task via UMassStateTreeProcessor, a steering processor, etc.) to drive evade / engage / regroup behaviour.

Mass agents consume the perceived-threats fragment directly - they don't issue EQS queries (there is no Mass-native EQS querier). EQS tactical queries are the UActor path (Part A). A hero-NPC-as-UActor querying with Apex: Known Threats can include crowd-relevant threats by tagging, giving a practical hybrid.

What you build by hand

The nodes above are all shipped and usable. What is not shipped is prebuilt query assets - you author your own from the generators and tests, which Tutorial 07 walks through end to end.

Four nodes need live flow-field or macro-graph state and are not available: the flow-field-directed generator, the flow-field-alignment test, the active-flow-field context, and the portal-proximity generator.

Tutorial 04 - StateTree, the move-task side of step 3.
Tutorial 06 - Mass Entity, for the Part B agent setup.
Tutorial 07 - Tactical EQS, authoring queries from these nodes.
ULcmTacticalQueryLibrary - the Blueprint tactical primitives (Is Segment LOS Clear, Find Best Cover, ...) the EQS tests wrap.