Optimisation of unsteady flight in learned avian perching manoeuvres

Abstract

Perching at speed is amongst the most challenging flight behaviours that birds perform1,2, and beyond the capability of current autonomous vehicles. Smaller birds may touchdown by hovering3-8, but larger birds typically swoop upward to perch1,2 - presumably because the adverse scaling of their power margin prohibits slow flapping flight9, and because swooping transfers excess kinetic to potential energy1,2,10. Perching is risky in larger birds6,11, demanding precise control of velocity and pose12-15, but it is unknown how they optimize this challenging manoeuvre. More generally, whereas cruising flight behaviours such as migration and commuting are adapted to minimize cost-of-transport or time-of-flight16, the optimization of unsteady flight manoeuvres remains largely unexplored7,17. Here we show that swooping minimizes neither the time nor energy required to perch safely in Harris hawks Parabuteo unicinctus, but instead minimizes the distance flown under hazardous post-stall conditions. By combining motion capture data from 1,563 flights with flight dynamics modelling, we found that the birds choice of where to transition from powered dive to unpowered climb minimizes the distance from the landing perch over which very high lift coefficients are required. Time and energy are therefore invested to maintain the control authority needed to execute a safe landing, rather than being minimized continuously as in technical applications of autonomous perching under nonlinear feedback control13 and deep reinforcement learning18,19. Naive birds learn this behaviour on-the-fly, so our findings suggest an alternative reward function for reinforcement learning of autonomous perching in air vehicles.