The key insight: "The dogs could simply wait until the sheep were facing in the direction they wanted before closing in".

This is standard herding procedure. Head them up in the desired direction, then move them out.[1]

Sheep are so herd-bound that once the front of the herd is going in the desired direction, the rest will bunch up and follow. Even through a narrow gate. There's been millennia of selective breeding for that. Cattle are not as herd-bound. Horses even less so.

[1] https://smallfarmersjournal.com/cattle-handling-part-1-basic...

Cute dogs, interesting dynamics with chaotic elements, physics, and did I mention the cute dogs?

> Raissa D’Souza, a professor of computer science and engineering at the University of California, Davis, praised the scientists’ clever use of real world data and agreed that the findings could be useful in the context of robot swarms, particularly for search and rescue. But she pointed out that the results only apply to a very specific type of physics problem—small noisy systems that feature random switching between two states—and don’t generalize over all examples of small, noisy collectives.

So how would this apply to drones tracking small groups of soldiers, or tracking AI-driven drone "flocks"?