FlowField

A precomputed distance-from-sources field over a GridGraph, exposed at point-level granularity for continuous-space agents.

Wraps three things together that previously had to be carried in parallel by every continuous-space agent that wanted to follow a gradient: the graph, a set of goal cells (sources), and the distanceField map that quantifies "how far from any goal." Adds point-level conversions (cellOf / centerOf) so agents holding continuous positions can query the field without rewriting the same point-to-cell arithmetic at every call site.

Construction runs one multi-source Dijkstra (the GridGraph.distanceField call). Queries are O(1) for distanceAt / arrivedAt and O(|passable neighbors|) for directionAt. Reconstruct (with the same or different sources) to refresh — there is no in-place recompute. Reconstruction is cheap, so building a fresh FlowField in initialize() of each replication is the standard pattern when the graph or sources may change between replications.

Typical use (evacuation):

class MyModel(parent: ModelElement) : AgentModel(parent, "evac") {
    val graph: GridGraph = ...
    val exits: Set<Cell> = setOf(Cell(0, 0), Cell(29, 29))
    lateinit var field: FlowField
    override fun initialize() {
        super.initialize()
        field = FlowField(graph, exits)
    }
}
// Per-agent: while (!field.arrivedAt(myPos)) { step in field.directionAt(myPos) }

Geometry model: cells form a uniform grid of side cellSize anchored at origin. Cell (c, r) covers the continuous-space half-open rectangle [origin.x + c*cellSize, origin.x + (c+1)*cellSize) in x and similarly in y, with center at centerOf(Cell(c, r)). For non-uniform or transformed layouts, subclass or compose; the simple uniform case covers every shipped example.

Parameters