Data driven prediction of bat flight kinematics and trajectory

Bat flight is unique among animal species due to their patagium wings which consist of a thin membrane connecting wing bones. This allows bats to have many degrees of freedom, typically more than birds or insects. In this work, we quantify the dimensional complexity of three different flights, straight flight, ascending right turn, and U-turn made by a great round-leaf bat (Hipposideros armiger). We employ proper orthogonal decomposition (POD) to determine the relevant number of modes that capture the essential kinematics associated with each flight. The POD is performed for the full flights and also on a flap-wise basis to investigate if the dimensional complexity of single flaps varies across the different flights. A study is performed to identify the marker points on the bat body that exhibit independent kinematic behavior across different flight maneuvers, with the goal of identifying the regions or joints that contribute most distinctly to the overall motion. Finally, we propose and test a deep learning architecture for predicting global flight trajectory given the wing motion and inversely predict the wing motion in the body frame required to effectuate an observed flight trajectory. These findings offer valuable insights for developing bat inspired unmanned aerial vehicles.