By watching “sheepdog YouTube” trials where dogs shepherd only four or five unpredictable sheep, researchers describe a two-step approach that turns the animals’ indecision into a steering tool. The same idea has been translated into an “Indecisive Swarm Algor
For hours. the videos look almost impossible: trained sheepdogs trying to herd a handful of sheep that refuse to commit to a direction. The scene is familiar to anyone who has watched sheepdog trials—except for one detail that researchers say changes everything. In the century-old tradition of sheepdog competitions, the dogs don’t work with a calm, crowd-like herd. They control only four or five sheep.
That small number. the researchers write in Science Advances. makes the flock “noisier”—less predictable and harder to manage than a larger group of 50 or more. Without the safety of numbers. individual personalities can dominate the motion. driving the sheep to rapidly flip between two instincts: “Flee the dog!” and “Keep calm and follow the others.”.
The question that kept returning, even as the dogs seemed to succeed anyway, was simple: how do they steer a small group of erratic animals into any kind of forward movement at all?
The answer begins with what the sheep are doing when no one is “in control.” The researchers analyzed sheepdog trials on YouTube and also talked with farmers in Georgia. What they found wasn’t a forceful, constant chase. It was a two-step strategy.
First, the dog waits. Specifically, it waits for all the sheep to face the right direction as they flip around at random. Then comes the second step: immediately chase them.
Once the aligned sheep start moving, the group does not stay obedient. The sheep quickly switch between “follow” and “flee,” and their formation breaks. At that point, the dog pauses again, waiting for the animals to orient once more. The result. the study team describes. is a slow. “delicate dance.” Study co-author Saad Bhamla. a biomolecular engineer at the Georgia Institute of Technology. compares the timing to sailing a ship on a windy day—raising sails only when the wind is headed your way.
The researchers say they weren’t just impressed by the choreography. They treated the sheep’s randomness as something you could exploit.
From there, they built a robotic method designed around indecision, not suppression. Inspired by the chaotic sheep. the researchers devised an “Indecisive Swarm Algorithm” for robots that keep switching who they follow—either the researchers’ controller or a neighboring bot. In tests described by the researchers. the tactic produced robots that were easier to control than approaches where the bots either followed only the controller or averaged the movement of all their neighbors. which eventually dilutes the controller’s signal.
Bhamla and colleagues frame the core idea as leverage: once the group is already switching behavior, a controller can time its influence so the swarm stays maneuverable rather than locking up.
Ted Pavlic. a computer scientist at Arizona State University who was not involved in the study. puts it in practical terms: “Indecisiveness prevents the group from binding up and makes it more pliable.” He also points to deadlocks—he mentions groups of autonomous cars—as the kind of failure mode that noise can help avoid.
Raphaël Sarfati. a physicist at Goodfire AI and also not involved in the work. challenges a default instinct to eliminate disorder. “We tend to think of noise as a problem that makes systems less predictable, less optimized,” Sarfati says. “[But] we see everywhere that noise. a little bit of noise at least. is really. really good for driving systems toward a better optimum.”.
If the sheepdog work looks like an odd corner of science—farm practice turned into algorithms—that’s exactly the point. Watching how a dog handles a handful of unpredictable sheep has become a blueprint for managing other collectives that behave badly: the researchers say the “Indecisive Swarm Algorithm” could help organize other noisy groups in the future. such as drones. self-driving cars. or even AI agents working together. The programming idea is straightforward: swap between following other individuals’ leads and following the overarching controller.
In a world where engineers often fight randomness, these findings offer a different lesson—one taught by a dog that doesn’t rush, waits for alignment, and then moves. It’s not the absence of chaos that makes control possible. It’s the timing of it.
sheepdog trials Science Advances indecisive swarm algorithm robotics swarms biomolecular engineering Georgia Institute of Technology autonomous cars drones self-driving cars AI agents noise in systems