MIT Built a Robot That Flies Like a Real Insect
MIT engineers have built an aerial microrobot that weighs less than a paperclip and flies with the speed and agility of a real insect. It is the fastest and most agile robot of its kind ever demonstrated.
Photo source:
Mit.edu
MIT Aerial Microrobot: Closing the Gap With Real Insects
MIT
researchers have built an aerial microrobot that can fly with the speed and
agility of a real insect, something no robot of its scale has achieved before.
Tiny flying robots have existed in research labs for years. The concept has
always been compelling. Machines small enough to fly through spaces no
conventional drone could reach, navigating tight environments the way real
insects do naturally. The reality has always fallen short. Until now, aerial
microrobots could only fly slowly along smooth, predictable paths, nowhere near
the speed and agility of the biological insects they were modelled after. A new
AI-driven control system developed at MIT changes that entirely.
MIT Aerial Microrobot: Insect-Scale, Insect-Speed
The
MIT aerial microrobot is a microcassette-sized device that weighs less than a
paperclip. Insect-like wings powered by soft artificial muscles flap at an
extremely fast rate, producing the agile movement needed for complex
manoeuvres. Previous versions were controlled by a hand-tuned system that
limited how fast and aggressively the robot could fly. The new AI-based
controller changes that entirely. Speed increased by approximately 450 percent.
Acceleration increased by approximately 250 percent compared to the
researchers' best previous demonstrations. The robot completed 10 consecutive
somersaults in 11 seconds, even when wind gusts of more than one metre per
second threatened to push it off course. It never strayed more than four or
five centimetres from its planned flight path throughout.
MIT Microrobot AI Control System Explained
The
breakthrough behind the performance improvement is a two-part control system
developed jointly by MIT's Soft and Micro Robotics Laboratory and the Aerospace
Controls Laboratory. The first part is a model-predictive controller, a
powerful planning system that uses a mathematical model to predict the robot's
behaviour and calculate the best sequence of actions for complex manoeuvres
like aerial flips and rapid turns. This controller accounts for the precise
conditions the robot needs to perform repeated flips without accumulating
errors that would cause it to crash. The second part uses imitation learning to
compress that powerful planning system into a lightweight AI model that runs in
real time. The result is a controller that delivers both the performance of a
complex planning system and the computational efficiency needed for live
flight.
Search and Rescue Applications for MIT Microrobot
The
researchers describe search and rescue as a key future application for the MIT
aerial microrobot. A robot this small could fly through rubble and tight spaces
after a disaster, reaching survivors in locations larger drones cannot access.
The team is also working toward adding onboard cameras and sensors so the robot
can fly outdoors without being connected to an external motion capture system.
Coordination between multiple robots is another area of future research.
You have exceeded your free limits for viewing our premium content
Please subscribe to have unlimited access to our innovations.
Get The Latest In Innovation
Join Our Newsletter
Where bold ideas find bold readers.
