Pontryagin Neural Network Optimized Terrain Following Paths for Supersonic Atmospheric Vehicles

Author(s):

Bradley T. Burchett

Abstract:

For many years, tactical aircraft have used terrain following or nap-of-the-Earth flight to avoid enemy detection. With advances in fixed wing drones, such tactics are used by fast, small, unmanned vehicles as well. This report explores path planning for such vehicles using Pontryagin Neural Networks—a method previously used for space and low-speed atmospheric vehicles. This work includes several extensions from previous work, such as more realistic 3D flight dynamics and realistic supersonic aerodynamics. Multiple obstacles are combined into a single, smooth 2D function using Fourier sine series. By encoding actual Earth topography using the Fourier approximation, we demonstrate optimal flyable trajectories over actual terrain.