Robotics & Physical AI
MIT Microrobot Flies Without Motion Capture Using Onboard Sensors

Key PointsOnboard sensing and computation let the microrobot navigate without external motion…
View more
- Onboard sensing and computation let the microrobot navigate without external motion capture.
- Enlarged artificial muscles lift the electronics package while preserving flight control.
- Human operators still supply high-level commands so autonomy remains guided rather than independent.
MIT researchers announced on November 18, 2026, a 1.29-gram flapping-wing robot that can hover, follow trajectories and land using onboard sensing and computation, without relying on an external motion-capture system.
The Science Robotics paper reports centimeter-scale positional accuracy from a combination of sensors, estimators and a low-level controller. In a 30-second experiment outside a motion-capture system, the robot avoided obstacles and landed on a sunflower. It still receives power from an offboard source.
Kevin Chen, an MIT associate professor and co-corresponding author, described it as “the first demonstration of an aerial microrobot that is sensory and compute autonomous.” The claim concerns onboard perception and control; the robot does not yet supply its own power or set its own high-level destinations.
Larger Actuators Carry the Electronics
The team’s earlier aerial microrobots could not carry the electronics required for navigation. To increase payload capacity, the researchers tuned the radius and length of the soft actuators, or artificial muscles, that flap the robot’s four wings. They report increasing actuator size by 50% while boosting net lift force by 46%, producing a lift-to-weight ratio of 1.9.
That additional lift supports a 244-milligram electronics package containing the sensors and microcontroller. The sensor suite combines an inertial measurement unit for attitude, an optical-flow sensor for lateral speed and position, and a time-of-flight sensor for height. A fiberglass substrate holds the package in place against the robot’s vibrations.
Removing motion capture also removes the system’s absolute-position reference, so the robot can drift off course. The team simplified its estimator to calculate orientation, angular velocity and position quickly, reducing computational cost. The researchers found that continuous estimates at a faster rate performed better than slower, more accurate estimates for this flight system.
Human Commands Guide the Demonstration
Flight tests included hovering at different heights and following point-to-point trajectories. For the plant demonstration, the team used a joystick to send high-level position setpoints, guiding the robot around several plants before it landed on the sunflower.
The paper describes this arrangement as a hierarchical controller: the operator supplies high-level commands, while the robot’s onboard systems handle estimation and low-level flight control. The sunflower landing therefore demonstrates human-guided autonomy, not independent mission planning or camera-based object recognition.
The Earlier Controller Ran Offboard
MIT’s December 3, 2025 report describes an earlier robot whose AI-based controller enabled 10 somersaults in 11 seconds. That controller ran on an external computer, and onboard sensing remained future work. The latest system addresses that limitation by integrating navigation sensing and computation on the aircraft.
Vision and Batteries Are the Next Steps
The researchers identify precision pollination, crop-health monitoring, pest detection and industrial surveillance as potential applications. Such robots could hover near delicate plants without damaging them or operate in settings where larger machines cannot be used.
Power remains a separate engineering constraint. The team’s analysis attributes about 6% of total power consumption to the electronics, with the rest supplying the artificial muscles and wings. Reducing electronics consumption could provide capacity for additional components such as a camera.
The researchers say they are now working to add a camera for autonomous object recognition and obstacle avoidance, and to integrate power electronics and batteries for untethered outdoor flight. The next phase tests whether vision and batteries can be integrated while preserving the lift margin and precise control that made this demonstration possible.












