Interesting tidbit here. It's common to make models using attraction and repulsion forces between agents (or particles), but this model uses projected time for a collision instead, which is quite a bit different and assumes agents with perception and planning. Projected time to collision turns out to yield more accurate models of pedestrian behavior in a bunch of situations.

It does make sense that two agents wouldn't bother changing their path if their current path wasn't expected to yield a collision, but the repulsion models would change the path anyway in many of those cases.

The source paper is here:
Ars Technica article on the paper below.
Shared publiclyView activity