An efficient, modular controller for flapping flight composing model-based and model-free components

We present a controller that combines model-based methods with model-free data-driven methods hierarchically, utilizing the predictive power of template models with the strengths of model-free methods to account for model error, such as due to manufacturing variability in the RoboBee, a 100 mg flapping-wing micro aerial vehicle (FWMAV). Using a large suite of numerical trials, we show that the model-predictive high-level component of the proposed controller is more performant, easier to tune, and able to stabilize more dynamic tasks than a baseline reactive controller, while the data-driven inverse dynamics controller is able to better compensate for biases arising from manufacturing variability. At the same time, the formulated controller is very computationally efficient, with the MPC implemented at 5 KHz on a Simulink embedded target, via which we empirically demonstrate controlled hovering on a RoboBee.

Automated summary

The proposed controller combines model-based methods with model-free data-driven methods, achieving better performance and stability in controlling micro aerial vehicles like RoboBee.