Trajectory Generation and Control for UAVs Using Quintic PH Curves and Kolmogorov-Arnold Networks (KANs)

This paper presents a novel methodology that utilizes quintic Pythagorean-Hodograph (PH) curves for trajectory generation of unmanned aerial vehicles (UAVs) and Kolmogorov-Arnold Networks (KANs) for tracking problem. PH curves enable exact arc length computation and ensures smooth transitions in positions, velocities, and accelerations throughout the trajectory. The UAV's movements along the generated trajectory are regulated by KAN-based controllers. A separate controller is designed for each axis. We validate through simulations, indicating that a 40% improvement in tracking accuracy can be attained with KANs over conventional multilayer perceptrons (MLPs) under wind disturbances.