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

Aerial insects exhibit highly agile maneuvers such as sharp braking, saccades, and body flips under disturbance. In contrast, insect-scale aerial robots are limited to tracking non-aggressive trajectories with small body acceleration. This performance gap is contributed by a combination of low robot inertia, fast dynamics, uncertainty in flapping-wing aerodynamics, and high susceptibility to environmental disturbance. Executing highly dynamic maneuvers requires the generation of aggressive flight trajectories that push against the hardware limit and a high-rate feedback controller that accounts for model and environmental uncertainty. Here, through designing a deep-learned robust tube model predictive controller, we showcase insect-like flight agility and robustness in a 750-milligram flapping-wing robot. Our model predictive controller can track aggressive flight trajectories under disturbance. To achieve a high feedback rate in a compute-constrained real-time system, we design imitation learning methods to train a two-layer, fully connected neural network, which resembles an insect flight control architecture consisting of a central nervous system and motor neurons. Our robot demonstrates insect-like saccade movements with lateral speed and acceleration of 197 centimeters per second and 11.7 meters per second square, representing 447% and 255% improvement over prior results. The robot can also perform saccade maneuvers under 160 centimeters per second wind disturbance and large command-to-force mapping errors. Furthermore, it performs 10 consecutive body flips in 11 seconds - the most challenging maneuver among sub-gram flyers. These results represent a milestone in achieving insect-scale flight agility and inspire future investigations on sensing and compute autonomy.

Methods
The dataset comprises raw sensing data and commanded voltages, including position and Euler angles (using the XYZ convention), collected from a motion-capturing system (Vicon Vantage V5 and Vicon Tracker 3.9) and the voltages computed by the controller. The sensing data was retrieved from Vicon Tracker 3.9 and transmitted in real-time to a target computer (Speedgoat) via asynchronous UDP. All sensing data was saved at 10 kHz on the target computer, with no post-processing applied. The four voltages commanded by the controller were saved at 1 kHz.