Smart Objects and crowd avoidance

Smart Object boundary

Smart Objects describe spatial interaction opportunities and reserve Slots. They do not contain the execution logic; each interactor implements the returned Behavior Definition.

Core pieces:

• SmartObject subsystem and spatial database;
• SmartObject component on the world actor;
• Smart Object Definition asset shared across instances;
• Slots with transforms, filters/tags, and behavior definitions;
• optional Persistent Collection influencing runtime-instance lifetime.

Claim lifecycle

Find matching objects -> choose result -> claim slot -> validate handle
-> navigate to slot -> use/occupy -> execute behavior -> release

Rules:

• Search with bounded area, activity tags, user tags, behavior class, and other filters.
• A search result is not a reservation; always validate the returned Claim Handle.
• Revalidate slot/location and object state before/after approach.
• Release on success, failure, abort, death, possession change, timeout, and task/state exit.
• Handle OnSlotInvalidatedDelegate; the object may disable/unload/destroy while claimed or occupied.
• Keep durable gameplay outcome outside the transient claim.

A Persistent Collection keeps included Smart Object runtime instances in the simulation even if source actors stream out. That is a lifetime decision, not permission to execute interaction against unloaded visuals/collision.

RVO versus Detour Crowd

System	Use	Constraint
RVO	CharacterMovement-local velocity avoidance, including without NavMesh	Character only, less configurable, can push agents outside NavMesh
Detour Crowd	path-corridor-aware crowd avoidance for Pawns	fixed Max Agents/Max Agent Radius and Crowd Manager tuning

Epic states the two systems work independently and should be used exclusively. Do not enable both for one agent.

Crowd tuning

Start without avoidance and prove paths. Then add one system and test:

• maximum simultaneous agents and radius support;
• consideration radius/neighbors and wall samples;
• corridor width, opposing flows, doors, corners, merges, and destinations;
• separation/collision resolution and movement-component interaction;
• agents leaving NavMesh, deadlocks, oscillation, and arrival congestion;
• CPU cost at worst population.

Crowd avoidance cannot solve a level whose doorway is narrower than the throughput the encounter requires. Fix geometry, lane design, destination distribution, or scheduling when appropriate.

Smart Object + crowd pattern

Reserve before entering a congested approach, expose one Slot per true simultaneous user, move to a reachable slot transform, queue outside the interaction corridor, and release immediately on cancellation. Do not let every waiting agent continuously re-query every object.