The things depend a little bit what agent type you have (AGENT_TYPE). And what I_HERDING_TYPE. If I_HERDING_TYPE is
"large" (2 or more?), then the agents do not move at all, if they have not have found any door that they could use (visible or
known and not too much smoke). If I_HERDING_TYPE is "small" (less than 2?) they will follow the first guiding flow field
that is in the array inside the Fortran source code. The DOORs and EXITs are in the same array in the door selection algorithm.
But I do not remember if EXITs are before DOORs or vice versa, so the first one might be door or exit. But you could test this
quite easily, put one exit and one door and both not visible and not known and see, where the agent goes.
And the "lost agents", at least for AGENT_TYPE=1 and 2(default), are checking around to see, if they see other agents
that know some door. If they find these, they will follow. I.e., they are using the "herding algorithm", if they do not have
anywhere to go.
Well, in the door selection visible is 360 deg angle, so the agents looks everywhere. First the closest neighbors, then
all visible agents. Well, for "follower agents" things are a little bit different. They see the agents in front of them ("in front
of them" is some angle, I do not remember too well). So, if someone opens an emergency exit in front of the follower
agent, the agent will follow the agent that opened and went trough the emergency exit that is not known to this
follower agent before.
TimoK