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

Area of Science:

• Robotics
• Bio-inspired Engineering
• Control Systems

Background:

• Insect-scale aerial robots currently lack the agility of natural insects, being limited to smooth trajectories.
• Achieving insect-like flight agility requires robust and computationally efficient control systems.

Purpose of the Study:

• To develop a deep-learned robust tube model predictive controller for insect-scale flapping-wing robots.
• To enhance the flight agility and maneuverability of aerial robots.

Main Methods:

• Designed a deep-learned robust tube model predictive controller (MPC).
• Implemented the controller on a 750-milligram flapping-wing robot.
• Tested the robot's ability to track aggressive trajectories and perform agile maneuvers.

Main Results:

• The robot demonstrated exceptional flight agility, including saccades with significantly improved lateral speed and acceleration.
• The neural network controller successfully tracked aggressive trajectories on a compute-constrained system.
• The robot performed saccades under wind disturbance and completed 10 consecutive somersaults.

Conclusions:

• The developed controller represents a milestone in achieving insect-scale flight agility.
• The findings inspire future research in sensory and computational autonomy for aerial robots.
• This work bridges the gap between insect flight capabilities and robotic systems.