The Flight Trick Insects Already Know — and Robots Still Don't
Insects fly without thinking about it. That turns out to be the whole point.
A new study from Cornell shows that most flying insects don't rely on constant neural feedback to stay aloft. Their bodies are physically tuned — wing shape, mass distribution, flap frequency — so that stability emerges from the mechanics itself. The flight control happens in the geometry, not the brain. This finding, published May 1 in the Proceedings of the National Academy of Sciences, offers a fresh set of design rules for building flapping-wing microbots — a class of flying robots that has proved stubbornly hard to make work at small scales.
The research was led by Z. Jane Wang, a professor whose 2014 work showed that fruit flies sense their body orientation every wing beat — roughly every 4 milliseconds — and make mid-flight corrections using neural circuits. That earlier result emphasized active control: the insect as a tiny autopilot. The new paper inverts the picture. Wing to body mass ratio, wing loading, hinge position, flap frequency, and wing motion amplitude — taken together, Wang calls this a five-dimensional morphological and kinematic space. The analysis produced two explicit stability formulas, one of which describes an anti-resonance state where the coupling between wing inertia and body mass produces natural stability. No neural intervention required.
"The surprise was that many insects don't need active neural control to maintain stable flight," Wang told the Cornell Chronicle. "We found passive stability in forms we thought required active stabilization."
The contrast with current engineering practice is sharp — Robohub noted in its coverage of the Cornell work. A recent paper from MIT, published in Science Advances, describes building a working insect-scale flapping-wing robot weighing 750 milligrams — about 0.75 grams — that requires a feedback controller running at 480 hertz to stay stable during aggressive maneuvers. The researchers achieved that speed using deep learning to compress the control model down to a two-layer neural network running on a constrained microcontroller. The robot demonstrated saccade maneuvers at 197 centimeters per second and performed 10 consecutive body flips in 11 seconds — the paper calls it a milestone. What it also shows is the engineering tax that comes with treating stability as an active problem to be solved continuously rather than a passive property designed into the system from the start: every gram of compute, every sensor polling at that rate, adds weight that the flapping mechanism has to lift.
"In principle, this offers a completely new route for designing a robotic flapping-winged machine," Wang said, per the Cornell Chronicle. "Instead of relying on extensive feedback control, which is only partially successful, our results suggest that we can tune the shape and the frequency of the flapping devices such that the flyers are passively stable already."
Wang's formulas give engineers a target to hit. But the anti-resonance condition requires getting several physical parameters right simultaneously — wing inertia, hinge placement, mass ratios, flap frequency. At centimeter scale, manufacturing a wing that meets those specifications with sufficient precision is a materials and assembly problem the paper doesn't solve. What the research offers is a design map, not a route to a working robot. Potential uses — pollinator-scale crop monitoring, confined-space inspection, short-duration environmental sensing — remain speculative until someone does the manufacturing work the formulas describe.
The Cornell Chronicle published the study May 1, 2026. The paper is co-authored by Wang and Owen Wetherbee, also of Cornell, and was funded by the National Science Foundation. It is available open access at PNAS.