On winter evenings across Rome, thousands of European starlings pour into the sky above Termini station and perform one of nature’s most baffling stunts: a murmuration that wheels, contracts, and reverses direction in a fraction of a second, all without a single collision. For years, physicists and biologists could describe pieces of the trick. Now, a team led by Andrea Cavagna and colleagues at Italy’s Institute for Complex Systems has stitched those pieces together. Their study, published in Communications Biology in May 2026, presents a unified mechanistic model of how starling flocks execute collective escape maneuvers, built in part on data gathered by flying a robotic hawk directly into wild murmurations.
A turning signal that refuses to fade
The foundation of the new model rests on more than a decade of field work. Using synchronized high-speed cameras, researchers reconstructed the three-dimensional trajectories of individual starlings mid-flight and measured how directional cues pass from bird to bird. A landmark 2014 study in Nature Physics showed that turning information propagates linearly across a flock with almost no loss of intensity. A cue launched by one bird on the left edge arrives at the right edge nearly as strong as it started. That is why a cloud of 400 starlings can reverse course almost as a single body.
The mechanism behind that stability is what the researchers called “behavioral inertia.” Each bird resists changing direction for a brief moment after picking up a neighbor’s cue. That tiny delay acts as a buffer, preventing overcorrection and keeping the flock from splintering. Because every individual applies roughly the same pause, the turning wave rolls through the group in an orderly, predictable sweep rather than triggering chaotic ripples.
Fixed neighbors, not fixed distances
A separate puzzle was how birds maintain spacing while pulling sharp turns. Research published in Proceedings of the Royal Society B showed that starlings do not track every bird within a set radius. Instead, each individual pays attention to a fixed number of nearest neighbors, roughly six or seven, regardless of how compressed or stretched the flock becomes. When the group tightens during a dive, the interaction range shrinks automatically. When it spreads, the range expands. The result is a self-adjusting safety margin that keeps collision risk low even at high speed.
That neighbor rule also helps explain why murmurations scale so well. Data gathered under the European STARFLAG project revealed that starling flocks exhibit scale-free correlations in both position and velocity. A heading change by one bird correlates with changes across the entire flock, no matter how large the group. The same local rules that keep a cluster of 50 birds coherent also work for a swarm stretching across hundreds of meters.
What the robotic hawk revealed
To study escape behavior under controlled conditions, the team deployed a robotic raptor designed to mimic the approach profile of a peregrine falcon, the starling’s most dangerous aerial predator. By varying the robot’s angle and speed of attack, researchers could trigger different escape responses and film the results with calibrated camera arrays on the ground.
Under real and simulated predator attacks, the flocks produced visible “agitation waves” that swept through the group faster than the birds themselves were flying. Work by Procaccini et al., published in Animal Behaviour in 2011, measured those waves and found that their speed often exceeds the flight speed of the flock itself. That speed advantage gives individuals downstream extra reaction time before the threat arrives. Field observations cataloged multiple escape patterns, including flash expansions (the flock balloons outward), lateral wave events, and full splits, and found that the geometry of the predator’s approach largely determines which pattern the flock deploys.
The Communications Biology paper pulls these threads into a single framework. It links each bird’s position relative to the predator with the type and speed of its response, then shows that the previously measured propagation rules, neighbor counts, and wave dynamics are sufficient to reproduce several distinct escape patterns observed in real flocks. The model also predicts how those patterns should shift as the predator’s trajectory changes, offering testable hypotheses for future fieldwork.
Where the gaps remain
The new model is consistent with existing data, but it has not yet been independently reproduced. Several open questions remain. The exact control algorithms and flight-path telemetry for the robotic hawk have not been published in full technical detail, which means other teams cannot yet replicate the predator-stimulus conditions precisely. Whether repeated robotic passes alter flock density or individual stress levels over days or weeks has not been tested in a controlled longitudinal study, leaving open the possibility that habituation could skew results gathered late in a field campaign.
There is also the question of generality. Most high-resolution datasets come from European starlings gathering at predictable roosts in open airspace. Whether birds in cluttered habitats, or species with different social structures, follow the same fixed-neighbor rules remains untested. Weather, light levels, and background noise could all affect how reliably cues propagate, but systematic comparisons across those conditions are still lacking.
What drone engineers are watching
For researchers designing autonomous drone swarms, the starling findings offer a concrete design principle: fixed-neighbor interaction rules paired with weak-attenuation signal passing can maintain group coherence without any central controller, even when the group must react quickly to a moving obstacle. Programming each drone to track a limited set of nearest peers, relay course corrections with minimal signal loss, and apply a brief standardized response delay could make artificial swarms more robust in crowded airspace.
That translation from biology to engineering is still early. No commercial drone system yet operates on starling-derived algorithms at scale. But as the field measurements grow more precise and the models more predictive, the murmuration above Rome’s train station may end up shaping the traffic rules for the skies above cities worldwide.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
