Aerobatic maneuvers in insect-scale flapping-wing aerial robots via deep-learned robust tube model predictive control.

Abstract

Aerial insects exhibit agile maneuvers such as sharp braking, saccades, and body flips under disturbances; in contrast, insect-scale aerial robots are limited to tracking smooth trajectories with small acceleration. To achieve similar flight capabilities, insect-scale robots require a robust and computationally efficient controller. Here, through designing a deep-learned robust tube model predictive controller, we showcase exceptional flight agility in a 750-milligram flapping-wing robot. Our neural network controller can track aggressive trajectories and run at a high rate on a compute-constrained system. The robot demonstrates saccades with a lateral speed and acceleration of 197 centimeters per second and 11.7 meters per square second, respectively, representing improvements of 447 and 255% over prior results. The robot also performs saccades under 160-centimeters per second wind disturbance and completes 10 consecutive somersaults in 11 seconds. These results represent a milestone in achieving insect-scale flight agility and inspire future investigations on sensory and compute autonomy.