A two-stage controller, designed to fit within a tiny computational budget, takes the flapping-wing robot from stabilized flight to aerobatic maneuvers.
MIT engineers have equipped their flapping-wing microrobot with a learning-based controller capable of making it perform demanding aerial maneuvers, including full-body flips. The architecture is divided into two parts in order to reconcile flight performance and on-board computing budget, which is very constrained on this scale.
Compared to the team's previous best results, the speed has improved by about 450% and the acceleration by about 250%. In the tests, the machine did ten consecutive somersaults in eleven seconds, including under gusts of wind trying to deviate it from its trajectory. The researchers also managed to reproduce the so-called jerky movement, by which some insects suddenly topple over, fly towards a position and then reverse their inclination to stop.
The interest of these performances goes beyond demonstration: in the long term, robots of this size could inspect rubble after an earthquake or circulate in spaces that are too cramped or too dangerous for a conventional drone.
Read the full article on the original source
Go to the source →