Tiny flying robots may soon help rescue people trapped after earthquakes or explore dangerous, hard-to-reach places.
But until now, these microrobots were slow, fragile, and unable to match the quick and acrobatic flight of real insects.
That has changed with a breakthrough from MIT researchers, who have created an aerial microrobot capable of flying with the speed, agility, and precision of a bumblebee.
Their new robot, described in Science Advances, can perform rapid turns, aggressive accelerations, and even gymnastic flips—completing 10 somersaults in just 11 seconds.
This represents an enormous leap in performance: speed increased by nearly 450% and acceleration by 250% compared with previous versions.
The microrobot is tiny—about the size of a microcassette and lighter than a paperclip. It flies using four rapidly flapping wings powered by soft artificial muscles.
These muscles contract and expand so quickly that the wings beat hundreds of times per second, just like those of real insects. While the hardware has improved in recent years, the missing piece was the robot’s “brain”—its flight controller.
Earlier versions relied on hand-tuned controllers written by engineers, which limited the robot’s ability to handle complicated, fast maneuvers. To overcome this, MIT teams led by Professor Kevin Chen and Professor Jonathan How developed a new AI-based, two-step control system.
The first step uses a model-predictive controller, a powerful algorithm that simulates the robot’s motions in real time and computes the best possible actions to follow a planned flight path.
This type of controller can plan flips, sharp turns, and extreme tilts, but it requires too much computing power to run directly on a small robot.
To solve this, the researchers used imitation learning—a form of artificial intelligence—to “teach” a simpler controller to mimic the expert planner.
They fed the AI model thousands of examples generated by the high-level controller, training it to predict the best movement commands almost instantly. The result is a lightning-fast decision-making system that can run in real time while the robot is flying.
With this new controller, the microrobot showed insect-like reflexes. It maintained stability even when struck by gusts of wind over one meter per second. It repeated precise flips without drifting off course and successfully performed saccade movements, a behavior used by insects to quickly shift direction and stabilize their vision.
These abilities open the door to future microrobots that can navigate cluttered environments, search through rubble, or fly outdoors with onboard cameras and sensors. Adding these features will be a focus of the researchers’ next steps. They also hope to develop systems that allow multiple microrobots to coordinate with each other and avoid collisions during flight.
Outside experts have praised the work as a major milestone. Sarah Bergbreiter of Carnegie Mellon University notes that the robot can still fly precisely despite challenges such as wind, imperfect manufacturing, and its own power cable tangling during flips. She says the approach points toward truly autonomous insect-scale robots with natural agility.
For now, the controller runs on an external computer, but the team believes simplified versions could one day operate directly onboard. If successful, these microrobots could become powerful tools for disaster response, environmental monitoring, or even pollination support.
This breakthrough shows that the dream of insect-level robotic flight is no longer science fiction—it is becoming reality.
