Bio-inspired real-time dynamics prediction and flight control for fixed-wing drones

Clear sky gusts threaten the safety operation of fixed-wing drones. Their signals may only be picked up by pressure or acoustic sensors, and are often noisy and sparse. On the other hand, it is difficult to separate gust information from the vehicles' own dynamics, and to exacerbate, precious time for controllers to react may already have elapsed. In this talk, we use data-driven approaches to predict the drones dynamics into the short future, in the wake of a vertical leading wing as a gust generator. The clear separation of gusts and the vehicles' own dynamics is made possible by the forward protruding multi-hole probes into the clean flow, and is inspired by how the Narwhal whale uses its long tusk to detect salinity signals upstream. We show that the vehicle's dynamics, even in the presence of unseen, continuous gusts, can be faithfully captured and predicted for a short time in advance by a pretrained multi-layer perceptron, which gains invaluable time for controllers to design optimal action to stabilize the drones. We showcase that even the simplest flight controller would benefit from this short horizon dynamics prediction, and such a framework may be applicable for all future fixed-wing flight controller design.