Navigation in Cities

This paper has just been published in Nature Computational Science, and it occurs to me that it’s crying out for a more parsimonious PCT-based interpretation: Vector-based pedestrian navigation in cities | Nature Computational Science

Good to hear from you, Roger!

“We posit that direction to goal is a main driver of path planning and develop a vector-based navigation model; the resulting trajectories, which we have termed pointiest paths, are a statistically better predictor of human paths than a model based on minimizing distance with stochastic effects.”

Urban walking is usually constrained to path segments that do not fall on a straight line between origin and destination. A straight path from A to B is the exception.

Pointiest paths would optimize a zigzag approximating a diagonal across a grid. Each zig and zag departs from the diagonal from origin to destination, so minimizing such lateral departures results in more turns from zig to zag and from zag to zig.

“(1) people increasingly deviate from the shortest path when the distance between origin and destination increases and (2) chosen paths are statistically different when origin and destination are swapped. We posit that direction to goal is a main driver of path planning and develop a vector-based navigation model”.

For (1), there are more options as to which turnings and segments to take. A path segment may have advantages meriting continuing on it longer (easier walking, wider sidewalk) or disadvantages (construction clutter, crowd at an entrance) increasing the perceived cost of making that turn – unless you like watching construction or want to see what the crowd is about. Each turn across vehicle traffic incurs a cost that you avoid by continuing straight, and that cost may be less at that farther intersection. Longer path, more occasions for preferring one way or the other, more variability, but all but one of the paths “deviate from the shortest path”.

For (2), I would guess that when one is facing in the opposite direction various environmental factors are different, e.g. oncoming pedestrian traffic, visual appearance of landmarks, sun in the eyes.

For a PCT simulation you would want to look at individual data. However, some results might be as simple as the CROWD simulation, if avoidance of oncoming crowds is a variable controlled by at least some individuals.

Consider directions like “four blocks east and nine blocks south”. Taken literally, you go east four blocks, turn right, and go south nine blocks. Or you go south until you see an easy place to cross toward the east, always counting blocks.

I wonder if they included data from European cities or from places like parts of Boston where it’s said that the roads were originally laid out by the cows. There, directions may be more explicit, like driving directions, with named streets, landmarks, and places to turn.