Research

MIT's 1.29-Gram Flying Microbot Navigates Without Motion Capture

MIT's 1.29-gram flapping-wing robot hovered, tracked trajectories and landed on a sunflower using only onboard sensors and computation — no motion-capture rig. Power still comes from a tether.

MIT Microrobot Flies Without Motion Capture Using Onboard Sensors
MIT Microrobot Flies Without Motion Capture Using Onboard Sensorsstrikers / pixabay
By Sophie Lindqvist5 min read

Why it matters

  • MIT announced on November 18, 2026 a 1.29-gram flapping-wing robot that navigates using onboard sensors and computation, without external motion capture.
  • A 50% larger actuator size boosted net lift by 46%, yielding a lift-to-weight ratio of 1.9 and carrying a 244-milligram electronics package.
  • In a 30-second test outside motion capture, the robot avoided obstacles and landed on a sunflower with centimeter-scale positional accuracy.
  • Kevin Chen called it "the first demonstration of an aerial microrobot that is sensory and compute autonomous"; power remains offboard.
  • About 6% of total power goes to electronics; the team is adding a camera and batteries for untethered autonomous flight.

MIT researchers have flown a 1.29-gram flapping-wing robot that hovers, follows trajectories and lands using only its own sensors and computation — with no external motion-capture system telling it where it is. The team announced the result on November 18, 2026, in a paper published in Science Robotics. In a 30-second demonstration outside a motion-capture arena, the robot avoided obstacles and landed on a sunflower.

Kevin Chen, an MIT associate professor and co-corresponding author, called it "the first demonstration of an aerial microrobot that is sensory and compute autonomous." That claim is precise: the robot perceives and controls its flight onboard, but it still draws power from an offboard source and still relies on a human operator for high-level commands. It doesn't choose its own destinations.

The distinction matters because motion capture has been the crutch of insect-scale robotics for years. Tiny flying robots could hover and even do somersaults in the lab, but only while an external camera-and-computer rig tracked their position in real time. Take that rig away and the machines were effectively blind. This work moves perception and control onto the aircraft itself — the step that has to happen before any real-world use.

Why couldn't earlier microrobots do this?

The short answer: weight. MIT's earlier aerial microrobots simply couldn't lift the electronics needed for navigation. Every sensor and microcontroller adds mass, and at insect scale the lift budget is brutally tight.

The fix came from the actuators — the soft artificial muscles that flap the robot's four wings. The researchers tuned the radius and length of these actuators, increasing their size by 50%. That change boosted net lift force by 46% and produced a lift-to-weight ratio of 1.9. In plain terms, the robot generates almost twice the thrust it needs to carry itself, and the surplus goes to payload.

The payload in question is a 244-milligram electronics package containing the sensors and a microcontroller, mounted on a fiberglass substrate that holds it steady against the robot's vibrations.

The sensor suite splits the navigation problem into three parts:

  • An inertial measurement unit tracks the robot's attitude.
  • An optical-flow sensor measures lateral speed and position.
  • A time-of-flight sensor measures height.

The researchers report centimeter-scale positional accuracy from this combination of sensors, estimators and a low-level controller.

What did the robot lose when it lost motion capture?

Motion capture doesn't just track a robot — it gives the robot an absolute reference for where it is in space. Remove it, and the robot's position estimates can drift off course over time.

The team's answer was to simplify the estimator so it could compute orientation, angular velocity and position quickly. That trade-off — speed over precision — turned out to be the right one. The researchers found that continuous estimates produced at a faster rate performed better than slower, more accurate estimates for this flight system. A fresh, slightly rough estimate of where you are beats a perfect one that arrives too late to act on.

The flight tests covered hovering at different heights and following point-to-point trajectories. In the plant demonstration, an operator used a joystick to send high-level position setpoints, steering the robot around several plants before it landed on the sunflower.

The paper describes this setup as a hierarchical controller: the human supplies the goals, and the robot's onboard systems handle estimation and low-level flight control. The sunflower landing demonstrates human-guided autonomy — not independent mission planning, and not camera-based object recognition. The robot didn't see the flower; it flew to where it was told.

How does this compare to MIT's earlier work?

MIT reported on December 3, 2025 an earlier version of the robot whose AI-based controller could perform 10 somersaults in 11 seconds. Impressive acrobatics, but that controller ran on an external computer, and onboard sensing was listed as future work.

The new system closes that gap. Navigation sensing and computation now live on the aircraft itself, which is what allowed the robot to fly outside a motion-capture volume in the first place. The somersault robot needed the lab's infrastructure; this one needs an operator and a power tether.

What comes next: cameras and batteries

The researchers point to precision pollination, crop-health monitoring, pest detection and industrial surveillance as potential applications. A robot this size could hover near delicate plants without damaging them, or work in spaces where larger machines don't fit.

Power is the next wall. The team's analysis attributes only about 6% of total power consumption to the electronics — the rest goes to driving the artificial muscles and wings. Cutting the electronics' share of that budget would free capacity for additional components, most importantly a camera.

The researchers say they are now working on exactly that: adding a camera for autonomous object recognition and obstacle avoidance, plus integrating power electronics and batteries for untethered outdoor flight. That would remove both remaining dependencies — the human operator's setpoints and the offboard power supply.

The next phase tests whether vision and batteries can be added while preserving the lift margin and the precise control that made this demonstration possible. Every gram of sensor added eats into the 1.9 lift-to-weight ratio that carries it.

For now, the result stands as a proof point: an insect-scale aircraft that senses and computes its own flight. The pollination and surveillance scenarios remain distant, but the enabling step — getting navigation off the lab bench and onto the robot — is done.

Source: unite.ai

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.

823 articles

Related articles

  1. NXP’s modular robotics platform targets faster drone development
  2. Google Launches Gemini Robotics ER 2 With Multi-Robot Collaboration
  3. Ring Launches Robotics Subbrand With a $249 Indoor Security Drone and $999 Patrol Robot
  4. FLEX scores 0.99 as robots learn broader human preferences
  5. Samsung and LG split on humanoid robots ahead of CES 2027

« Previous articleNext article »