The maths behind a mid-air dogfight: what dragonfly territorial fights reveal about contested airspace

A dragonfly closing on a rival does not improvise. It runs through a recognisable sequence: a rapid climb, a tight turning engagement, and a sharp climb-and-turn at the end if it loses contact with its opponent. That ordering, documented across hundreds of filmed territorial contests and reported on 19 July 2026, is the central finding from a research team at Imperial College London that has spent the past several field seasons turning what looks like aerial chaos into a measurable, repeatable pattern book.

The study, summarised in a 19 July 2026 Phys.org dispatch, treats the dragonfly dogfight the way an aerospace engineer treats an engagement: as a sequence of stabilised flight modes that recur across individuals, sites, and days. Male dragonflies contest small patches of pond and stream edge to defend mating territory. The Imperial group argues that the contests resolve not by which insect flies faster, but by which one executes a small menu of recognised manoeuvres first, and in the right order. The implication is taxonomic humility: a brain the size of a sesame seed is running a structured engagement protocol, and most of the apparent variability in the fights is being generated by a handful of underlying rules.

A behaviour, then a model

The team built its case in two layers. First, observational: high-speed video of naturally occurring male–male contests at field sites in the United Kingdom, classified by an ethogram that broke each interaction into discrete moves. Second, computational: a simplified model in which each contestant follows a small set of rules and the emergent encounter is the one the camera recorded. The Phys.org write-up reports that the simulated encounters reproduced the observed sequence statistics closely enough that the model is being treated as a working description, not a metaphor.

What the model discards matters as much as what it keeps. It does not require detailed knowledge of the rival's identity, history, or motivation. It does not assume the insects are solving a multi-step game-theoretic puzzle on the wing. It assumes they are responding to local cues: a rival's angular position, rate of closure, and a clock that determines when to escalate, hold, or break off. That minimalism is the point. When a small set of local rules reproduces a large set of observed encounters, the burden of explanation shifts away from individual intelligence and toward the dynamics of the interaction itself.

The recurring playbook

Three moves do most of the work, according to the summary. A fast climb to gain altitude relative to a rival, used to set up the next exchange from above. A turning engagement, a near-symmetric spiral in which both insects trade position and rate of closure, which functions as the contest proper. And an exit: a climb-and-turn when the chaser loses visual contact, ending the encounter without a decisive close pass. The frequency and ordering of these moves, rather than their absolute geometry, is what separates winning trajectories from losing ones.

That finding is consistent with earlier work on territorial flight in other insect groups and with the broader intuition of an aerodynamic laboratory that studies how insects use unsteady lift, body rotations, and wing–wake interactions to produce forces a fixed-wing aircraft cannot. Dragonflies are an extreme case because they have two pairs of independently controlled wings and can hover, fly backwards, and pivot in place; their control authority is closer to a quadcopter's than to a bird's. What the Imperial group has shown is that the extra authority does not produce extra behavioural complexity in this context. It compresses the contest.

What it is not

A reading the research does not support: that dragonflies are running anything like the strategic games associated with fighter pilot training, or that the territorial contests have a stable winner-take-all payoff structure that a simple model can recover. The Phys.org summary is careful on this point. The model fits the data, but it is a description of recurrent geometry, not a theory of motivation. Mating territory is the proximate context; whether the contests are honest signals, costly rituals, or something closer to error-corrected conflict resolution is not settled by this paper.

A second caution. The filmed contests are pond- and stream-edge encounters in the United Kingdom, recorded in daylight, against a backdrop of abundant prey and well-defined perches. Whether the same rule set holds for migratory dragonfly species that form massive feeding swarms, for tropical species with year-round breeding, or for high-latitude populations facing short breeding seasons is an open question the Imperial group has not, on the available evidence, addressed. The rule book the paper extracts is a temperate, territorial one.

Why the structure matters

There is a temptation to treat any rule-like behaviour in a small animal as either trivial or over-interpreted. The Imperial work resists both. The contest it describes has enough variation that a purely reflexive account does not fit, and enough regularity that a high-dimensional cognitive account is unnecessary. That middle position is the most interesting one, because it is the position that much of contemporary research on insect flight is converging toward: not reflex, not deliberation, but a structured interaction in which local rules produce global order.

The structural read is straightforward. If territorial combat in a four-winged insect with a brain measured in milligrams resolves through a recognisable sequence of stabilised modes, then the apparent variety of in-flight contests is mostly the noise of execution rather than the signal of intent. For biologists, that sharpens the next round of questions: which cues gate the transitions between moves, how the rules develop across the adult lifespan, and whether the same rule set reappears in non-territorial flight, such as prey capture or mating display. For engineers working on small autonomous aerial systems, the work is one more data point in a quietly accumulating literature suggesting that useful flight behaviour does not require large neural budgets, only well-chosen local rules and a body that can execute them.

The research also slots into a wider pattern. Across the past decade, bio-inspired aerodynamics labs have documented similar compression in flocking birds, schooling fish, and swarming midges: a small rule set, a constrained body, and a recognisable behavioural signature. The Imperial dragonfly result is the territorial version of that finding, and it sits comfortably next to work on how mosquitoes evade raindrops, how bees read optic flow on approach, and how hawkmoths stabilise hover in unsteady flow. The deeper the field drills into insect flight, the less mystery remains about the moves and the more mystery attaches to the sensors and the wiring that decide between them.

The next test will be whether the same rule set survives contact with a different body of data. If it does, the field will treat the Imperial model as a working default, much as flocking models are now used as defaults for bird movement. If it does not, the most likely culprit is the simplification of the visual-field model: real dragonflies track with nearly full spherical coverage, and a model that treats the rival as a point in the frontal field may be hiding subtler structure in the periphery. Either result is informative.

Desk note: Monexus has framed this as a story about behavioural structure rather than insect intelligence. The wire summary foregrounded the "simple rules" angle; we have added the comparative biology context and the bio-inspired-aerodynamics frame to clarify why a small rule set in a small animal is treated as a serious result.