Performance

Documentation Unreal Engine AI Performance

What SEC costs, the three knobs that cut it, and how to read stat SEC on your own hardware.

Movement direction sampling is where the plugin's time goes. Three properties cut it, and stat SEC tells you whether you need to.

Three knobs, in order

All three sit on the Movement Evaluator component, on the AI controller. Num Samples heads the section itself; Enable Avoidance and Enable Nav Aware Sampling head the Avoidance and Nav Sampling groups nested under it.

Knob	Default	What changing it buys
Num Samples	16	Directions scored each pass. Drop it to 8 and the scoring loop, the nav probes and the hazard scoring all halve together. Range 4 to 32.
Enable Avoidance	on	Untick it and the sphere overlap for neighbouring pawns stops. Worth it for solo bosses and sparse fields where enemies do not pack together.
Enable Nav Aware Sampling	on	Untick it and the navmesh probes stop. The ledge guard and hazard-area avoidance go with them, so enemies will strafe off edges.

Details

Movement Evaluator Component

Movement Evaluator

Num Samples

Avoidance

Enable Avoidance

Nav Sampling

Enable Nav Aware Sampling

Advanced

Nav Check Interval

0.1

Nav Stagger Stride

Num Samples scales the direction-scoring loop. Avoidance and nav sampling both refresh at 10 Hz by default.

Start with Num Samples. It sets the loop count for both direction scoring and the nav probes, so halving it moves the two largest numbers on the profile together. Setting Avoidance Weight to 0 skips the overlap query as well, without unticking anything.

Turn the debug overlays off before you profile. SEC.Debug.All 1 draws a text block above every pawn within SEC.Debug.MaxDistance (4000 cm by default), and that draw cost lands in your frame while showing up nowhere in stat SEC. Run SEC.Debug.All 0 first, then measure.

What runs when

Cadence is how often the work happens. The cost column is what one pass does.

System	Runs	Cost per pass
Movement	Every tick that a brain or an AI Movement task asks for it	Scores Num Samples directions and picks a winner
Avoidance	Inside the movement pass, pawn overlap refreshed 10 times a second	One sphere overlap, then a penalty per neighbour found
Nav sampling	Inside the movement pass, cache refreshed every Nav Check Interval (0.1 s), spread across Nav Stagger Stride (4) frames	Projects the sampled directions onto the navmesh
Detour check	Every Detour Recheck Interval (0.5 s) while the enemy sits inside Hybrid Switch Distance	One path query weighed against the straight-line distance
Action scoring	Every Evaluation Cooldown (0.1 s), and only while the enemy is fighting	Hard gates, then a score for every action still standing
Reaction scoring	Once per trigger rule that an incoming stimulus matches	Gates and scores the reactions that rule allows
Awareness	On a timer at Evaluation Interval (0.1 s)	Advances a meter per remembered actor and ages the rest
Brain	Every tick, with action scoring gated inside it	Picks a phase, then drives movement and action scoring
Role assignment	Every Role Reassignment Interval (8 s)	One fitness score per combatant per candidate role
Threat detection	Every tick	One angle check against the focus actor
Melee tracing	Every tick between the start and the end of a trace window	One swept trace per socket, sub-stepped by how fast the weapon moved
Approach window	Every tick while an approach window is open	Steers one pawn toward its approach target
Territory	Only when something asks	Route arithmetic, or one navmesh query for a wander spot
Combat tokens	On claim and release	Slot bookkeeping on the target

Idle enemies are cheap on purpose. Territory, tokens and reactions do nothing until something calls them. The brain skips its movement pass while an enemy walks a patrol leg, waits out a post's dwell, or looks around a spot it reached, and it skips action scoring outright unless the enemy is in combat. Melee tracing, action execution and the approach window each keep their tick switched off until they have work.

Match the tuning to the crowd

What dominates changes with the crowd, and that decides where tuning pays off.

Scenario	Active enemies	What dominates	What to do
Soulslike duel	1 to 5	Nothing worth tuning	Leave the defaults alone.
Action RPG pack	10 to 20	Direction sampling, with avoidance climbing as they close on you	Lower Avoidance Radius (100 cm) before cutting samples, so fewer neighbours reach the penalty loop.
Horde	50 to 100	Direction sampling across the pack	Cut Num Samples to 8. This is the largest single lever.
Stress test	150+	Direction sampling and nav probes	Cut samples, raise Nav Stagger Stride (under Nav Sampling, behind the Advanced expander) so fewer enemies sweep on the same frame, and raise Evaluation Cooldown on enemies out of the fight.

Output Log

Active SEC enemies20 enemies

1 duel50 horde150 stress

Illustrative scaling. Run stat SEC in your level for real numbers

MovementEvaluator - Total

0.070 ms

MovementEvaluator - Direction Scoring

0.063 ms

MovementEvaluator - Avoidance

0.004 ms

RoleSubsystem - Role Distribution

0.001 ms

ThreatDetection - Tick

0.008 ms

Estimated SEC overhead

0.083 ms

Share of 60 fps frame budget (16.6 ms)0.50%

Low share at this count

The slider models linear scaling from an illustrative per-enemy figure. It shows the shape; your machine gives you the numbers, through stat SEC.

Cost scales roughly with the number of enemies running a movement pass every frame. An enemy standing at a post, waiting out a dwell, or dead costs a fraction of one in a fight.

Debug it

Open the console (~) and run:

stat SEC

Stat	Covers
`MovementEvaluator - Total`	The full movement pass, containing the four below it
`MovementEvaluator - Direction Scoring`	The sample loop
`MovementEvaluator - Avoidance`	Throttled overlap scoring
`MovementEvaluator - Nav Sampling`	The nav validity cache refresh
`MovementEvaluator - Detour Query`	The path-length detour check
`ActionEvaluation - Scoring`	Gates and scores every candidate action
`ReactionEvaluation - Scoring`	Gates and scores reactions for one stimulus
`Awareness - Evaluate`	One pass over remembered actors
`Brain - Tick`	The native brain tick, containing action scoring and the movement pass
`RoleSubsystem - Role Distribution`	The global role assignment pass
`ThreatDetection - Tick`	The per-tick threat level

Two of these rows are totals, not separate costs. MovementEvaluator - Total contains the four movement rows under it, and Brain - Tick contains ActionEvaluation - Scoring and the whole movement pass whenever the native brain drives the enemy. Adding the column up counts that work two or three times. Under a StateTree the same work runs with no enclosing brain tick, so Brain - Tick reads zero and the rest stand on their own.

Melee tracing and the approach window carry no stat line. Measure those in Unreal Insights.

For log-based diagnosis, see Debugging & Troubleshooting.

AdvancedFiltering an Unreal Insights capture

The same scopes publish CPU trace events, so an Insights capture can be filtered down to the plugin:

SEC_MovementEvaluator, SEC_DirectionScoring, SEC_Avoidance, SEC_NavSampling, SEC_DetourQuery, SEC_ActionScoring, SEC_ReactionScoring, SEC_Awareness, SEC_BrainTick, SEC_RoleDistribution, SEC_ThreatDetection.

They nest the same way the stats do, and Insights draws that nesting for you, which is the easier read when the totals confuse you.

Slow down the thinking

Two cadences decide how often an enemy makes a decision, and both are safe to widen for enemies that are background dressing rather than the fight.

Evaluation Cooldown (0.1 s) spaces action picks. It sits on the SEC Brain component for a native brain, and on the Poll Action task for a StateTree. Raise it to 0.25 for grunts that do not need ten decisions a second.

Role Reassignment Interval (8 s) is how often global role assignment re-runs. It lives in Project Settings > Plugins > Soulslike Enemy Combat > Combat Roles > Timing, so it is one value for the whole project. Set it to 0 and the periodic pass does not start, leaving role assignment to whatever asks for it.

AdvancedThe rest of the role timing

Timing

Role Reassignment Interval

Min Time In Role

Reevaluate Targets On Reassignment

Min Time In Role holds an enemy in its role for a minimum stretch, which stops role thrashing without touching the pass cost.

Reevaluate Targets On Reassignment adds a target-selection pass for every registered AI to each cycle. It costs nothing in a single-target setup and is worth watching in a multi-target one.

Threat detection ticks every frame by default and does one angle check, so it rarely shows up. Where it does, Set Tick Interval on the Threat Detection component skips frames.

What perception costs

Awareness runs its own bookkeeping on a timer, but the sensing underneath it is Unreal's perception system, and that is the part that scales with your level.

Sight Radius (1500 cm) is the biggest lever, because it decides how many stimulus sources the sight sense considers each update. Cut it to what the encounter needs. Peripheral Vision Half Angle (90 degrees, a 180 degree cone) narrows what qualifies before a trace happens.

Untick Enable Sight, Enable Hearing or Enable Damage Sense on the Awareness Config to drop a sense from the perception component the config creates. An enemy that reacts only to being hit needs neither of the first two. The three toggles grey out when Auto Create Perception Component is off, because a perception component you place yourself keeps its own sense configuration.

AdvancedTrimming awareness memory

Evaluation Interval (0.1 s) is how often the component advances its meters and ages entries. Raise it and detection ramps in coarser steps; the meter maths compensates, so a slower interval changes responsiveness rather than the outcome.

Track Last Known Location records a position and velocity per remembered actor. Untick it in memory-tight scenarios, at the cost of the investigate behaviour that reads those values.

Sight Max Age (5 s) and Hearing Max Age (4 s) set how long Unreal's perception system holds a stimulus after the target leaves. Shorter ages mean fewer remembered actors for the awareness pass to walk each interval.

What melee tracing costs

Tracing enables its own tick when a swing starts and disables it when the swing ends, so an enemy that is not attacking costs nothing here.

While a swing runs, each registered socket sweeps every frame, split into sub-steps so a fast weapon cannot skip past a thin target between frames. Sub Step Distance (10 cm) is the longest gap one sweep step may cover, and Max Sub Steps (20) caps the count however many the speed asks for.

Raise Sub Step Distance to trade precision for fewer sweeps. The cost multiplier that matters more is socket count: three registered sockets on a weapon is three sweeps per sub-step. See Melee Trace.

AdvancedWhat the plugin already throttles for you

None of these need setting up. They are the reason the default numbers hold at a pack size.

Sample cache. The trigonometry for the direction fan is computed once and rebuilt only when Num Samples changes.
Throttled avoidance. The pawn overlap query runs 10 times a second, not every frame. The scoring in between reads the cache.
Staggered nav sweeps. Nav Check Interval and Nav Stagger Stride spread full nav sweeps across frames, so a whole pack does not sweep on the same one. Both sit in the Nav Sampling group behind its Advanced expander.
Timer-based roles. Role assignment runs off a timer on its interval, outside the frame loop.
Hybrid movement. Beyond Hybrid Switch Distance (800 cm) the enemy pathfinds instead of sampling directions, so distant enemies skip the expensive loop.
Conditional ticks. Melee tracing, action execution and the approach window each start with their tick disabled and enable it only for the window they need.
Phase gating. The brain skips its movement pass while it owns the pawn for a patrol leg, a dwell or a look-around, and skips action scoring entirely outside combat.

AlternativeRunning the movement pass on your own schedule

The Movement Evaluator component does not tick. Something has to ask it for a pass, which is what the SEC Brain component and the StateTree AI Movement task do each tick.

Drive it yourself and the cadence is yours: call Evaluate And Move from a Blueprint timer at whatever rate the enemy deserves. A background enemy on a 0.2 s timer costs a fifth of one evaluating every frame, and looks it, so reserve this for enemies the player is not looking at.

Suspend Movement takes a duration and pauses the pass outright, which is cheaper still for an enemy that should hold position.