Limitations
Crowd Toolkit aims to be honest about its current scope. These are the known constraints in
1.0 - most are on the
Current constraints
The simulation reasons on one world X/Y plane. Height is tracked per agent UE-side (see Ground Snapping), which handles slopes and uneven terrain, but multi-floor or overlapping geometry (two walkable surfaces above each other) is not supported - agents can't be on a bridge and under it at once.
All agents share one radius (Agent → Radius). Per-agent / per-type radius (and the
cross-type avoidance it enables) is on the roadmap. Per-agent speed is already supported.
The crowd runs locally with no network replication built in. A server-authoritative pattern (simulate on the server, replicate transforms) works today and is the documented approach; lockstep determinism is being investigated as a longer-term feature.
The simulation advances on the game thread. It's built to handle large agent counts there, but a multithreaded tick is the headline v2.0 item for pushing the ceiling higher.
Wall tiles can be added (and the whole grid reset), but individual tiles can't be removed yet. Runtime obstacle removal and an actor-driven obstacle component land in v1.x. See Wall Generation.
On the roadmap
Briefly, in roughly the planned order:
- Runtime obstacle removal + an obstacle component (place a building → crowd reroutes)
- Agent following (move-to-moving-target), single-unit micro, ray/segment queries
- Collision groups / team flags filtering avoidance
- Fixed tick + interpolation (smoothness decoupled from frame rate)
- Per-agent radius with cross-type avoidance
- Multithreaded simulation and time-sliced flow-field builds (v2.0)
TIP
Hitting one of these in your project? It helps to know which matter most - feature priority is driven by what users actually ask for.