Skip to content
Autonomous Systems Review
Feature / Robotics

MIT's Robotic Insect Lands on a Sunflower Under Its Own Navigation

A machine weighing less than a paperclip's worth of electronics has demonstrated autonomous flight and landing on a living plant — suggesting that insect-scale robots may be closer to practical use than most people assumed.

Rob Polli — Editor

22 Nov 20266 min read

Post on X LinkedIn
A robotic insect with a safety orange light sits on a stylised sunflower against a plain grey background.
Illustration: Autonomous Systems Review

The Weight of a Possibility

Picture a sunflower standing in a field, petals open, stem swaying in a light breeze. Now picture a robotic insect — no larger than the real thing — descending toward it, correcting its trajectory, and settling on the flower's face with the kind of quiet precision that takes a human pilot years of practice to develop. The robot was not remote-controlled in that moment. A pilot gave it a target position using a joystick, and the machine handled everything else: working out where it was in space, adjusting its wings, and placing itself exactly where it was told to go.

That entire autonomous capability — the sensing, the state estimation, the flight control — ran on just 244 milligrams of onboard sensors and computing hardware. To be clear about what that number means: it is a fraction of the mass of a single raindrop. And yet it was enough.

What the Machine Actually Did

The robot was developed at MIT. The experimental setup was deliberately simple in one sense and demanding in another. A human operator with a joystick specified target positions — essentially telling the machine where to go — but offered no guidance on how to get there. The robot computed its own position, managed its own attitude and executed its own control loops in real time, all without offloading computation to an external system.

Landing on a sunflower sounds poetic, but it is also a genuinely hard engineering problem. A flower is a soft, irregular, moving surface. It is not a flat pad marked with fiducials. The robot had to interpret its environment and commit to a landing without the kind of rigid, predictable target that makes most robotic landings tractable. The fact that it succeeded — and that it did so at insect scale, with 244 milligrams of electronics doing the heavy lifting — is the detail that makes researchers stop and reconsider what is actually possible at this end of the size spectrum.

Why Scale Is the Hard Part

Insect-scale robotics has long occupied a peculiar position in the field: endlessly cited as a long-term ambition, rarely demonstrated doing anything autonomous in an unstructured environment. The reason is straightforward. As you shrink a flying robot, almost everything becomes harder simultaneously. Batteries hold less energy. Sensors with useful range and resolution tend to be heavy. Processors capable of real-time flight control draw power that a small battery cannot supply for long. And the aerodynamics of very small flapping or spinning wings are less stable and less well-understood than those of larger aircraft.

The achievement at MIT does not dissolve all of those problems at once. What it does is demonstrate that the specific problem of autonomous position estimation and flight control — historically one of the most computationally and sensor-intensive challenges in small robotics — can be squeezed into a 244-milligram package without rendering the robot incapable of real tasks. That is a meaningful boundary crossed.

Autonomous navigation without offboard computation is what separates a genuinely deployable small robot from a tethered demonstration. When the sensing and control live on the robot itself, the machine can go places where a wireless link is unreliable, where GPS is unavailable, and where the environment is too dynamic for a remote operator to react in time.

From Sunflower to Application

The sunflower landing is an experiment, not a product. It would be a mistake to read it as a signal that insect-scale autonomous drones are ready to be ordered from a catalogue. But it is equally mistaken to treat it as a curiosity with no practical direction, because the applications that would benefit from exactly this kind of machine are not difficult to imagine — and several of them are urgent.

Crop monitoring is the most obvious. Modern precision agriculture already uses drones, but the drones available today are orders of magnitude larger than insects. They survey from altitude; they do not land on individual plants and examine them at close range. A swarm of insect-scale robots that could perch on flowers, leaves and stems — gathering data on moisture, disease, pollination status or pest damage at the level of the individual plant — would generate a quality of information that no current aerial system can match. The fact that such a robot can now navigate and land autonomously, without depending on external infrastructure, is a step toward that possibility rather than away from it.

Search and rescue in confined spaces is another territory where small autonomous fliers could matter. So is infrastructure inspection in locations inaccessible to larger platforms — inside turbine housings, within dense pipe networks, through the narrow voids of a collapsed structure. In each case the constraint is the same: the robot must be small enough to fit, and it must be capable enough to navigate without someone guiding it continuously from outside.

The Sensor and Compute Problem, Miniaturised

The 244-milligram figure deserves attention not just as a headline number but as an engineering statement. Packing position estimation and flight control into that mass implies that the sensors involved — almost certainly including some form of inertial measurement, possibly optical flow or other visual input — have been selected and integrated with extreme care for weight and power consumption. The same applies to whatever processor is running the control algorithms.

This is where insect-scale robotics connects to a broader trend in embedded AI and low-power computing. The past several years have seen significant progress in running neural network inference and classical control algorithms on microcontrollers and ultra-low-power chips that would have been considered hopelessly underpowered for the task not long ago. What MIT has demonstrated with this robot is partly a mechanical and aerodynamical achievement, and partly a testimony to how much intelligence can now be packed into very small, very light electronics.

The combination — lighter sensors, smarter algorithms, more capable low-power processors — is what makes the 244-milligram autonomous control package possible. Each of those trends is continuing. That implies the capability demonstrated here is not a ceiling. It is more likely a floor from which the next iteration builds.

Where This Points

A robot that can be told a target position, navigate to it autonomously and land on a living, moving plant, all using onboard hardware that weighs less than a small insect, is not a trivial accomplishment. It resolves, at least in prototype form, a question that has been open for years: can insect-scale flying robots carry enough intelligence to operate without continuous human guidance in environments that are not engineered to assist them?

The answer emerging from MIT's sunflower experiment is yes, under the right conditions, with the right engineering. That does not mean the work is finished — endurance, reliability at scale, and operation in complex wind environments all remain genuine open questions. But the particular barrier that this experiment addresses — autonomous navigation and landing at 244 milligrams — has now been cleared.

Somewhere between that sunflower and the applications that actually need this technology, there is a lot of engineering still to be done. The distance, however, looks shorter than it did before.

Cite this article

Rob Polli. "MIT's Robotic Insect Lands on a Sunflower Under Its Own Navigation." Autonomous Systems Review, 22 Nov 2026. https://autonomoussystemsreview.com/articles/mits-robotic-insect-lands-on-a-sunflower-under-its-own-navigation.

Related coverage

Explainer / Robotics

Explainer: Why a $500m Robotics Bet Hinges on Disposable Hands

Genesis AI has raised $500m and built a robot workforce taught by humans wearing sensor gloves. The commercial logic depends on whether mechanical hands that last only weeks represent an unavoidable trade-off or a design problem the industry has yet to solve.

Rob Polli22 Nov 20265 min

The Autonomy Briefing

The week in autonomous systems, in one email.

Launching soon. Sign up and you’ll get the first issue.