On a winter evening outside Rome, a cloud of roughly 3,000 starlings folded in on itself, split around a diving silhouette, and snapped back together in under two seconds. The silhouette was not a real peregrine falcon. It was RobotFalcon, a remotely piloted raptor replica steered by researchers from the University of Groningen and colleagues in Italy, and every wingbeat of the flock’s escape was captured by synchronized high-speed cameras on the ground.
The resulting study, published in Communications Biology in May 2026, offers the most detailed picture yet of the rule that keeps thousands of starlings from crashing into each other at speed: each bird tracks roughly six or seven of its nearest neighbors, no matter how tightly or loosely the flock is packed. That single principle, fed into a 3D computer model calibrated against the field footage, reproduced the splits, compressions, and fluid reshaping that make murmurations so mesmerizing to watch.
A robotic falcon and a wall of cameras
Previous studies of starling flocking relied on filming undisturbed murmurations and reconstructing bird positions after the fact. The new work added a controlled threat. By flying RobotFalcon on repeatable attack trajectories, the team could trigger genuine escape maneuvers on demand and compare the flock’s real-time shape changes against predictions from their agent-based model.
RobotFalcon itself was validated in earlier peer-reviewed experiments published in the Journal of the Royal Society Interface, which showed that wild flocks react to the robot with collective escape responses comparable to those triggered by live raptors. Those responses held across multiple bird species, suggesting the coordination mechanism is not a starling quirk but a broadly shared survival strategy shaped by millions of years of predation pressure.
The camera array, building on protocols developed by the STARFLAG consortium, captured each bird’s three-dimensional position and velocity at high frame rates. That positional data became the raw material for the computer model: if the simulated flock, governed by a candidate set of rules, produced shapes that matched the real footage, the rules earned support. If it did not, they were discarded.
Why six or seven neighbors, not a fixed distance
The winning rule is called topological interaction. Instead of reacting to every bird within a set radius, each starling monitors a fixed handful of companions. The number, roughly six or seven, was first identified through 3D field reconstructions of starling positions and has since been confirmed across independent datasets. A network analysis published in PLOS Computational Biology found that this neighbor count represents a sweet spot: enough connections to hold the group together, few enough to keep the cognitive load manageable for a bird-sized brain.
The practical payoff is resilience. When a flock compresses during a predator dive, a distance-based rule would suddenly flood each bird with information from dozens of new neighbors. A topological rule keeps the workload constant. When the flock stretches out, the same rule prevents birds on the edges from losing contact, because they are still tracking their nearest six or seven companions even as physical gaps widen.
How a turn crosses a flock of thousands
One of the most striking findings from earlier STARFLAG research, reinforced by the new model, is the way directional changes propagate. When a cluster of birds on one flank banks left, the signal does not radiate outward from a leader like a broadcast. Instead it passes neighbor to neighbor in a wave, much like a ripple crossing a pond. Researchers reconstructing three-dimensional flock trajectories using methods described in a STARFLAG technical paper on measurement protocols measured the speed of that wave and found it fast enough to keep a flock of several thousand individuals coherent through sharp collective turns.
Underlying this speed is a property physicists call scale-free correlation: the degree of directional alignment among birds does not fade proportionally with distance. Work by Cavagna and colleagues, drawing on the same STARFLAG positional datasets, showed that velocity fluctuations in starling flocks are correlated over lengths proportional to the flock’s overall size rather than decaying over a fixed distance. In plain terms, a turning cue that starts with a handful of starlings on the left edge can reach the right edge without losing strength. That is why a murmuration containing 3,000 birds can execute a hairpin turn without the trailing half lagging behind or splintering off.
What the study does not settle
For all its sophistication, the work leaves several questions open. No study in the current literature has published direct collision counts from undisturbed wild flocks flying without any robotic stimulus. The RobotFalcon experiments reveal what happens under threat, but the baseline collision rate during calm, unprovoked murmurations has not been independently measured. Without that comparison, the precise safety margin provided by topological interaction is hard to pin down.
Long-term fitness data are also missing. The six-to-seven-neighbor threshold looks efficient in network models, but no published dataset tracks whether birds that deviate from it, attending to fewer or more neighbors, pay a measurable cost in survival or energy expenditure over a breeding season.
The raw 3D trajectory files from the RobotFalcon trials have not yet been released for independent re-analysis. Earlier STARFLAG technical reports described detailed protocols for recording and reconstructing large flocks, including error-handling procedures outlined in a companion paper on experimental reconstruction techniques. Applying those verification steps to the newest data would require open access to the underlying position and velocity files. Until they are available, outside teams cannot test whether alternative neighbor-count thresholds might fit the observed flock shapes just as well.
From murmurations to machines
The findings have drawn immediate interest from engineers working on drone swarms and autonomous crowd-flow systems. A coordination rule that scales without centralized control and stays stable under sudden compression is, on paper, exactly what a fleet of delivery drones or search-and-rescue robots needs. But translating biology into hardware is not straightforward. Starlings operate with flexible wings, instantaneous sensory feedback, and millions of years of evolutionary tuning. Fixed-wing or rotary drones face different aerodynamic constraints, communication latencies, and failure modes. The biological evidence shows that topological rules produce coherent group motion in birds; whether they do the same for machines is an engineering hypothesis, not a demonstrated result.
Popular accounts sometimes go further, implying that scale-free correlations guarantee collision-proof flight. The evidence is more cautious. Topological interaction and rapid information transfer make collisions rare in the contexts studied, especially during predator evasion, but they do not eliminate them. Wind gusts, physical obstacles, and asymmetric predator attacks could still create local density spikes where individuals misjudge trajectories.
Three levels of confidence in the flocking framework
Strip away the spectacle and three tiers of certainty emerge. The most secure claims rest on direct measurements: 3D reconstructions of positions and velocities, and the statistical regularities pulled from them. At an intermediate level sit inferences about internal rules, such as the six-to-seven-neighbor interaction, which are strongly supported but still depend on the model used to extract them. At the most speculative level are extrapolations to other species, technologies, or environments, which remain hypotheses awaiting targeted tests.
What the converging lines of evidence do support is a simple, striking conclusion: starlings manage complex, high-speed group maneuvers by following a small set of local rules that prioritize a fixed number of neighbors, letting information sweep through the flock without any bird in charge. Opening the underlying datasets, comparing alternative interaction models head to head, and linking flocking behavior to long-term survival will determine how far this framework extends beyond the spectacle of a murmuration folding across a winter sky.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
