Basically putting a gradient field on a discretized map for rogue-like games and then doing motion planning accordingly. For example, having enemies gradient ascent/descent to attack players, gradient descent/ascent to flee from players.

Since many rogue-likes are played on ASCII grids, using an array to store the gradient field is not that memory intensive. I suspect one can do this in a "continuous" setting by overlaying a gradient field by consider each point of interest (enemies, players, gold) to deform the gradient in some way, say with a localized Gaussian falloff or something similar.

EDIT: Sorry, this specific algorithm is linear falloff in all directions. So it's more akin to calculating Voronoi regions and saving the distance to the point of interest.