An MIT team guided a flying microrobot around several plants and onto a sunflower without the external cameras its earlier robots needed to track their movement. An operator used a joystick to specify where the robot should go, but the machine estimated its own position and controlled its flight. The landing demonstrated neither independent flower recognition nor pollination, and the robot drew power through a tether.
In other tests, the robot hovered at several heights and flew between designated points, with position errors of just a few centimeters. Kevin Chen, who heads MIT’s Soft and Micro Robotics Laboratory, described the work as the first demonstration of an aerial microrobot able to handle both sensing and computation itself. The research was published in Science Robotics.
Finding enough lift for a 244-milligram payload
The team started with a machine roughly the shape and size of a microcassette but lighter than a paperclip. Soft actuators serve as artificial muscles for its four wings. Earlier versions could maneuver quickly but lacked the carrying capacity for navigation electronics, leaving them dependent on a laboratory motion-capture system.
Bigger muscles would themselves add weight. As Chen explained, enlarging an actuator can also reduce its power output, so an increase in size need not translate into more lift. By adjusting both the radius and the length of the actuators, the team increased their size by 50 percent and gained 46 percent in net lift. The resulting lift-to-weight ratio of 1.9 was sufficient for a 244-milligram package containing the sensors and a microcontroller. A fiberglass base kept the package secure against vibration in flight.
Fast position estimates limit drift
Three instruments divide the tracking work. An inertial measurement unit estimates the robot’s orientation, an optical flow sensor tracks sideways speed and position, and a time-of-flight sensor measures height. Without the external cameras, however, the robot lacks an absolute position measurement, and its estimate of where it is can drift.
The team simplified the calculations for position, orientation and rate of rotation so the small computer could update its estimates more often. For flight performance, frequent position updates proved more valuable than a slower calculation that was more accurate at each step. Co-lead author Yi-Hsuan (Nemo) Hsiao said the resulting positional accuracy was comparable to what the robot achieved with external cameras.
The electronics accounted for about 6 percent of total power use; the muscles and wings consumed the rest. The researchers identified possible savings in the electronics that could leave room for a camera. They are working toward onboard recognition of objects and avoidance of obstacles, as well as batteries and power electronics that would let the robot fly outdoors without a tether.
The researchers envision small robots moving among almond trees or blueberry bushes to pollinate blossoms in places where habitat loss has depleted natural pollinators. Another possible use is hovering near plants to check crop health or detect pests without damaging the plants. Those remain proposed agricultural tasks; the sunflower test showed the robot could reach a flower when given a destination.




Discussion