research/650-stygmergy-controller

[jharwell]

There are a number of different dimensions to this, but the main idea is that
the controller does things based on what it finds in the environment, instead of
having any memory (i.e. it is purel reactive, and thus easy to model
mathematically).

• Robot allocates a D1/D2 task based on what it encounters in the environment,
  and if it doesn't encounter anything interesting for a set period of time, it
  chooses a random task instead. No task length estimates!

• Robots bring blocks they pick up partway back to the nest and then MAYBE drop
  them near a pile of other blocks (i.e. doing caches without the cache data
  structure), where the chance a robot drops vs. not is somehow based on the
  size of the cache/pile of blocks it encounters.

From Lerman2003, if the controller is purely reactive to the environment, they
it SHOULD be possible to predict their rate of ecountering various stimuli
sufficiently well that the overall task allocation distribute can be modeled by
the rate equation idea from the paper.