MIT engineers have given their paperclip-weight flying microrobot a new two-part AI control system that lets it move with genuine insect-like agility. The upgrade boosted the robot's top speed by about 447% and its acceleration by 255%, letting the roughly 4-centimeter device complete 10 consecutive body flips in just 11 seconds while staying within 4-5 centimeters of its planned flight path, even as researchers pushed it with wind disturbances.
The controller pairs a model-predictive planner for complex maneuvers with a deep-learning policy trained through imitation learning for split-second flight decisions, balancing raw performance against the tiny computational budget the robot can carry. "The flight performance of our robot is comparable to insects in terms of speed, acceleration, and pitching angle," said Kevin Chen, the MIT associate professor who leads the project's Soft and Micro Robotics Laboratory.
The team, which also includes aeronautics professor Jonathan P. How, sees the breakthrough as a step toward deploying swarms of insect-scale robots in places conventional drones can't go, such as searching collapsed buildings after earthquakes. The work was published in the journal Science Advances.