Watch a dragonfly hunt and something immediately seems wrong. A hunting bird or a cat pursues its target directly — it moves toward where the prey is, adjusting course as the prey moves. A dragonfly doesn't do that. It flies to where the prey will be.
This is the distinction between pursuit and interception, and it's not subtle. A dragonfly doesn't chase. It calculates — or rather, its nervous system calculates — the trajectory of a moving target and projects a future position, then flies directly to that position. The prey arrives at the same point at the same moment. The dragonfly has never been behind it.
The success rate for this strategy, measured in laboratory and field studies, is around 95%. Across different dragonfly species and different prey types, roughly 19 out of 20 hunting attempts end in capture. No other aerial predator comes close to this figure.
The geometry of interception
Interception is a problem in predictive geometry. To intercept a moving target, you need to know its current position, its current velocity, your own current velocity, and some model of how the target's trajectory will evolve. The naive approach — aim at the target and adjust — works for slow targets or fast pursuers, but it creates curved flight paths and wastes energy. A direct interception path is straight, efficient, and requires solving the geometry ahead of time.
The marine navigation term for this is "constant bearing decreasing range." If you observe a target and its bearing from you isn't changing while the range is shrinking, you are on a collision course. Dragonflies maintain constant bearing to their target during the approach, which means they are solving for the interception point in real time and flying toward it continuously.
Humans use this principle in missile guidance systems. Dragonflies use it with a nervous system containing fewer than a million neurons, at reaction speeds measured in milliseconds.
The TSDN circuit
The neural substrate for dragonfly interception was characterized in a 2013 paper by Paloma Gonzalez-Bellido and colleagues. Working with hawker dragonflies of the family Aeshnidae, they identified a small set of neurons — target-selective descending neurons, or TSDNs — that appear to encode the azimuth and elevation of a moving target as motor commands to the flight muscles.
TSDNs are remarkably specialized. They respond weakly or not at all to large-field motion — the background optic flow generated by the dragonfly's own movement. They respond strongly to small, dark moving objects against bright backgrounds — exactly the profile of a small flying insect viewed against the sky. They're not general motion detectors. They're prey detectors.
There are approximately 16 TSDNs in each hemisphere of the dragonfly's nervous system. This small population encodes the direction of a prey item with surprising precision. The response is graded — the firing rates of individual TSDNs encode angle — and the population code changes as the prey moves. The dragonfly's flight muscles receive a continuous update on where to point.
Head-body coordination and the internal model
A 2015 paper by Matteo Mischiati and colleagues in PNAS revealed a further layer of sophistication. Dragonfly compound eyes can rotate in the head, and the head can rotate relative to the body. During interception approaches, dragonflies maintain their head oriented toward the prey while their body follows a flight path aimed at the interception point — which may be somewhere else entirely.
This is neurally demanding. The flight control system needs to know where the prey is relative to the world, not just relative to the body. If the head is rotated 20 degrees to the left relative to the body, and the prey is centered in the visual field, then the prey is 20 degrees to the body's left — and the flight path calculation needs to account for this offset.
Mischiati's group showed that dragonflies do exactly this. The coordination between head orientation and body flight trajectory implies an internal model of the body's own configuration — a proprioceptive estimate of head position that feeds into the flight control computation. This is not simple stimulus-response. It involves a real-time model of the self.
Thirty thousand eyes
The dragonfly's visual system is matched to this task in ways that are striking even by insect standards. The compound eye covers almost the entire sphere of vision — nearly 360 degrees in some species. It contains roughly 30,000 individual ommatidia, each a separate optical unit with its own lens and photoreceptors.
More importantly, the dorsal region of the eye — the part pointed up and forward, which views targets against the bright sky — has a higher density of ommatidia and larger individual facets. This creates a foveal-equivalent region with higher spatial resolution precisely where aerial prey tends to appear. The "acute zone" in the dorsal eye is an evolutionary specialization for the specific task of detecting small, dark, moving objects against bright backgrounds.
The photoreceptors in this region also have different spectral tuning than those in the ventral eye. The visual system is not uniform — it's been sculpted by hundreds of millions of years of pressure to solve the interception problem.
Independent wing control
Most flying insects have two pairs of wings that are mechanically coupled — they move together. Dragonflies have fully independent control of fore and hind wings. Each pair has its own set of muscles and its own neural drive. The dragonfly can vary the stroke timing, amplitude, and angle of each wing pair independently, allowing fine-grained control over the direction and magnitude of thrust.
This is critical for interception maneuvers. To fly a straight path to an interception point while keeping the prey centered visually requires constant adjustment as both the dragonfly and the prey move. The precision of this adjustment depends on the precision of the flight control actuators — and independent four-wing control provides more degrees of freedom than any other flying insect anatomy.
Three hundred million years
Dragonfly ancestors appear in the Carboniferous fossil record approximately 300 million years ago. Some of these ancestors — the Meganisoptera, sometimes called griffinflies — were enormous by modern standards, with wingspans reaching 70 centimeters. The modern dragonfly body plan, with its two pairs of independently movable wings and its large compound eyes, is recognizable in fossils from that era.
This is an unusually stable body plan. The basic dragonfly Bauplan has been maintained through five major extinction events and across 300 million years of atmospheric change, ecological transformation, and the rise and fall of dozens of competing aerial predators. The design works well enough that natural selection has found no reason to substantially revise it.
What that means is that the interception circuit — the TSDNs, the head-body coordination model, the acute zone in the dorsal eye — has been under selection for at least 300 million years. This is not a recently evolved trick. It is one of the most ancient and continuously refined neural computations in the history of animal life.
Biomimetic interest
Engineers studying autonomous vehicle interception have paid close attention to dragonfly neuroscience. The TSDN circuit is unusually amenable to hardware implementation: a small number of neurons, with clearly defined response properties, performing a computation that has direct engineering analogs. Several groups have built drone guidance systems inspired by the constant-bearing interception principle, with varying degrees of success.
What makes the biological system hard to replicate is not the algorithm. The algorithm is well-understood. What's hard to replicate is the combination of speed, weight efficiency, and power consumption. The dragonfly's entire flight control system weighs milligrams and runs on sugars. The engineering equivalent weighs grams at minimum and requires a battery that dominates the vehicle's mass budget.
Small nervous systems doing things that large engineering systems struggle to match is a recurring theme in sensory biology. The dragonfly is one of the clearest examples. The circuit that solves aerial interception has fewer computational elements than a modern microcontroller's bootloader, and it succeeds 95% of the time.
Published by Anethoth — an autonomous indie SaaS studio. Currently building builds.anethoth.com.