KINEMATIC CHARACTERIZATION OF BEE FLIGHT MODES USING DEEP LEARNING FOR PROSPECTIVE FLUID FLOW ANALYSIS

This study investigates wing kinematics across different flight modes of the Western Honeybee (Apis mellifera), including take-off, fanning, hovering, forward flight, and landing, using high-speed video capture and deep learning-based pose estimation. Wing motion was recorded in both free-flight and controlled laboratory settings and analyzed using DeepLabCut to reconstruct two-dimensional views of wing trajectories from a single-camera setup. These trajectories provide insights into the aerodynamic mechanisms underlying various flight behaviors. For example, earlier research has established that during takeoff, rapid, high-amplitude wingbeats generate thrust and lift, while hovering exhibits phenomena such as vortex shedding and potential wake recapture for energy-efficient lift generation. Landing prioritizes stability, lift reduction, and aerodynamic braking through altered wing kinematics, while ventilatory fanning-a thermoregulatory behavior-optimizes airflow generation with minimal energy expenditure. This work reveals kinematic trends and the effect of wing flexibility in the different flight modes in bee flight. By employing marker-less deep learning tools for detailed pose estimation, this study not only elucidates the biomechanical intricacies of bee flight-and especially their variance across notably distinct flight modes-but also lays the groundwork for future computational fluid dynamics analyses and bio-inspired robotic designs.