The Magic of Herding – Nautilus
Explore
WHen Doyle Ivie, a 77-year-old farmer and coach of Chouton-Berger in northern Georgia, received an email from Saad Bhamla and Tuhin Chakrabortty, two biophysicians of the Georgia Institute of Technology, he was intrigued. The researchers wanted to know if he would let them record the to-fro-ing of his shepherd songs.
The farmer could not understand what scientists could learn about the abstractions of physics by studying his muddy canines, but he accepted. Shortly after, Bhamla and Chakrabortty, who study collective behavior, stopped at the IVie farm to watch the dogs in action. On that day, the farm organized a chouton-berger test, a secular competition in which shepherd dogs show their breeding skills by directing small herds of sheep through a field, as well as dividing the group, known as “parade”, among other tasks.
Physicists have studied the farming of the huching for decades, but they have mainly modeled the way dogs bring together large herds, says Bhamla. In these large herds, sheep display what is called selfish behavior, gravitating the center and putting the others between them and the danger. But in small groups, as in singing tests, their decisions become more erratic, because they vacillate between the follow -up of the herd or go alone.
How the songs of shepherd manage to bring the sheep to go where they want in these chaotic conditions suggest a solution to a difficult problem in physics – controlling noisy and unpredictable collectives, which is relevant not only for sheep, but also for drones and robotic swarms and pedestrian movements, among others. These emerging collectives are motivated by the two models at the group level and by the interactions between individual members of the group.
ADVERTISEMENT
Nautilus members benefit from experience without advertising. Connect or join now.
Instead of deleting randomness, they found that shepherd’s dogs seemed to kiss him.
“Dogs do incredible things, so naturally we were interested,” says Bhamla.
After their visit to the farm, Bhamla and Chakrabortty gathered a series of sheepfold test recordings from YouTube videos. While they were looking for models in the recordings and spoke with shepherd’s handlers, they noticed that breeding and loss were summed up in a two -step process. In the first step, the dog slowly pushes the sheep at a distance without inducing panic, waiting for them to face the desired direction. In the second step, the dog approaches, increasing the threat of moving the sheep; Instinctively, they flee the predator perceived in the direction they face.
The next challenge for physicists was to translate what they had learned into a mathematical model. Their model included five sheep that changed their orientation either in response to a social influence (copying their neighbor), or in response to external force, such as a dog or a manager. To which the influence they have chosen to respond were determined at random, although individual sheep have been programmed to be more or less sensitive to dogs and managers. The dog dogs also followed the same process in two stages as in real life, the boost, then the trip.
ADVERTISEMENT
Nautilus members benefit from experience without advertising. Connect or join now.
Scientists have found that although random behavior is generally considered an obstacle to the control of a small collective, the indecision of sheep was in fact an advantage for dogs in the “division” task. Dogs could simply wait for the sheep to face the direction they wanted before closing, which limits the required efforts. Instead of deleting randomness, they found that shepherd’s dogs seemed to kiss him.
Based on these results, scientists have built an algorithm that could be used to predict behavior in other small indecisive behavior switching collectives. The results, which have not yet been evaluated by peers, were published in June on the Preprint server arxiv.
Ted Pavlic, Biologist and IT teacher at Arizona State University said that the results may apply to any situation where a person needs to guide a group of people with whom he could not communicate directly. “When I develop a strategy for a group of individuals I want to direct, I would suppose a coherent behavior of all,” explains Pavlic. “It shows that it may not be the best strategy.”
But Pavlic notes the limitations of the model, especially in the way it represents real sheep. For example, the model’s sheep do not avoid collisions and can be transmitted to each other. They don’t remember where they were. “It would be interesting to see what they find if they add these things to the model,” he said.
ADVERTISEMENT
Nautilus members benefit from experience without advertising. Connect or join now.
Raissa d’ouza, professor of computer science and engineering at the University of California in Davis, congratulated the intelligent use of real data by scientists and agreed that the results could be useful in the context of robot swarms, in particular for research and rescue. But she pointed out that the results only apply to a very specific type of physics problem – noisy systems that have a random switching between two states – and do not generalize on all the examples of small noisy collectives.
Bhamla and Chakrabortty agree that there is a lot to learn. In the future, they hope to explore more how the individual behavior of sheep influences the results. “The animals are so complicated. We are always gushing the surface, ”explains Bhamla. “What I find incredible is that we have taken such a simple competition and apparently and made it a whole story,” said Bhamla. “How beautiful is it?”
Main photo: Alexandra Morrison Photo / Shutterstock
