In the fractal turtle example , we let 200 turtles loose to roam across a complex plain, computing iterations of a mandlebrot set calculation for each point as they passed over it.
The above program worked very nicely except that very late in the calculation the "dumb" turtles were content to roam about the entire plain, computing points that long since had hit a predefined threshold.
Our solution was to tell the turtles to "climb upward" as they computed each point, in essence climbing up on top of the points that they were computing ever "higher". To avoid having turtles get stuck on local maxima , we had them leap forward 5 paces at every step rather than simply step forward 1 pace. We hoped that the effect would be to have them constantly leaping off of local maxima.
A sequence of snapshots was taken during the "emergence" of the mandlebrot set visualization.