Sheep dogs are amazing creatures, and easily the best thing for rounding up sheep. But how do they do it?
A study published in the Journal Of The Royal Society ("Solving the shepherding problem: heuristics for herding autonomous, interacting agents", because science papers rarely have blockbuster titles) observed sheepdogs via GPS to gain insight into how it is that they're so effective at herding sheep. Using the GPS data they were able to build a mathematical model of how the dogs shifted sheep around, and apply that to what they knew about the dog's senses.
Essentially the dogs are just using two simple rules, collecting sheep together if they observe gaps in the herd, and shifting them forwards if they can see a single mass of sheep ahead of them. If gaps emerge, the dogs shift from driving the dogs forward to collecting them back into a mass again. Using this data, they were then able to build a computational model following the same rules, and observe the same results at a computer level that the dogs already achieved.
So what's the point of all this research, apart from to prove that kelpies are really quite smart critters? It's got applications in fields where multiple robots work together, such as in dangerous environments, because rather than applying complex logic to each robot for returning to base or coordinating together, it's feasible that they could be driven by a single simple robot designed like a cattle dog to herd other robots together.