Skip to content

Use cases

The getting-started tutorial teaches the concept ladder in game terms. This page walks the same shipped machinery through three other adopter frames, each backed by an existing self-checking example.

Robotics prototyping: occupancy grids and replanning

A tess world with a passability field is an occupancy grid, and the dirty-driven loop is a replanner. examples/stairs_3d.cc demonstrates the full cycle in under two hundred lines: build a two-level world joined by a stair transition, verify reachability with the topology precheck, then demolish the stair — a queued edit marks the region dirty, tess::update_region_graph refreshes only the affected chunks, and the next query correctly reports the goal unreachable.

Mapped to robotics vocabulary:

  • Occupancy update — a queued field edit with a dirty mask, not a full-map rewrite.
  • Replan trigger — Cadence::on_dirty(mask) runs the planner exactly when the map changed.
  • Feasibility gate — the precheck rejects definitively unreachable goals without expanding the grid; only Unreachable is trusted, so it never wrongly fails a solvable query.
  • Reproducibility — ticks are fixed-step and deterministic: the same edits in the same order produce the same plans, which makes experiment runs repeatable and regressions bisectable.

The library is single-process and grid-based: it complements, rather than replaces, continuous-space planners and ROS-style middleware.

Agent-based modeling: many agents, shared fields

examples/ant_farm_vertical.cc is an agent-based model wearing game clothes: a vertical cross-section world where a colony of ants shares one multi-goal distance field through the byte-budgeted FieldProductCache instead of searching independently. The same shape serves evacuation, foraging, and diffusion-style studies:

  • Arbitrary per-tile state — a field schema holds whatever the model needs (pheromone, hazard, capacity), not just passability.
  • Population-scale routing — agents sharing a goal set amortize one field build; the pathfinding note maps each workload shape to its API.
  • Determinism — identical seeds and schedules reproduce identical runs, so results are citable and diffable.

Headless simulation: servers and batch runs

Nothing in the core loop needs a window. A server or batch experiment runs the schedule under its own fixed-step clock and simply omits the render bridge, which is optional. When observers do exist (a network mirror, a monitoring UI), DeltaFrame versioning gives them gap detection and explicit resynchronization: frames record what changed (individual tiles or box-granular dirty bounds), and the consumer re-reads those tiles from the authoritative world — or ships the values over its own channel — instead of rescanning the map. examples/render_delta_consumer.cc shows a consumer maintaining shadow state this way.